- lp-body-content
- first-letter
Developing countries are under intensifying pressure to create jobs, strengthen resilience, and keep growth alive—just as budgets tighten, and development aid comes under strain. The gap between what is needed and what is available is no longer looming on the horizon. It is here, driving the choices governments must make right now.
Closing that gap requires something harder to measure than dollars or disbursements: trust, coordination, and a shared willingness to put country priorities first. Cofinancing— when multiple partners pool resources behind a common program—is one of the most powerful ways to do this.
This idea was the basis of our recent event, The Power of Co-financing, which brought development partners together in Casablanca. Under the auspices of Morocco's Ministry of Economy and Finance, the World Bank Group and Agence Française de Développement (AFD Groupe) convened governments, multilateral development banks, bilateral partners, private sector actors, and national development banks to take stock of how co-financing is working in practice and what needs to change to make it work better.
Participants agreed that co-financing, when done well, allows countries to pursue investments that no single institution could support alone—reducing fragmentation, lowering transaction costs, and aligning external financing behind national strategies rather than institutional agendas. But they were equally clear that this does not happen automatically. Procedures need to be simplified and aligned. Teams need incentives to work across institutional boundaries. Financing must be coordinated and predictable. And success must be measured by outcomes delivered, not commitments announced.
From Principles to Practice: Voices from the Field
To better illustrate what co-financing can accomplish in practice, the Casablanca event featured a series of Ignite Talks. These short presentations showcased practitioners working on co-financed projects across organizations, sectors, and countries, sharing concrete examples of what co-financing looks like on the ground.
Six practitioners share their experiences below—from water partnerships in Morocco to energy access in Mozambique, from transformative urban infrastructure to municipal finance reform. Each story is different. But the same thread runs through all of them: when development partners align around a shared goal and put the client country first, more becomes possible.
00:04 Hello everybody.
00:05 My name is Timurbach.
00:07 I work for Eyeddin Rabat,
00:08 and today I will give you a team talk.
00:12 Let's dive straight into my story,
00:14 The Power of water partnership.
00:17 Well,
00:18 for those of you who don't know Morocco,
00:20 climate change and human activities have created serious negative impacts
00:25 all over the kingdom for the past 7 years.
00:28 And even if this year,
00:29 this is especially true for water supply,
00:31 and even if this year
00:33 has been very humid,
00:34 as you know,
00:35 as we talked,
00:35 there are floodings in the northern part of Morocco.
00:39 For the past 7 years,
00:40 water shortages were seen
00:42 in the kingdom,
00:44 in cities,
00:45 agriculture,
00:46 and livestock.
00:48 Well,
00:49 since 1992,
00:50 AFD has been working with national and local water utilities,
00:55 and we still are,
00:57 but in 2024,
00:59 everything changed.
01:00 I guess you could say the Moroccan government asked us
01:02 to come to the dance and to partner with them.
01:06 Well,
01:07 uh,
01:07 to
01:09 understand how we could fit in the picture and what could be our added value,
01:12 we held many,
01:14 many discussions with the water ministry
01:16 and water institutions.
01:18 We also held discussions with the World Bank,
01:20 and the World Bank is already providing a $350 million
01:25 program for results to implement the national water policy.
01:29 To be candid with you,
01:31 AFD has a choice to make
01:34 going with business as usual,
01:36 that is AFD
01:38 alone,
01:38 or as the mayor of Freetown just put it,
01:41 doing the usual road.
01:43 Or we can join forces and work together
01:46 and try to,
01:47 to
01:48 raise to match the Moroccan ambition by working together.
01:54 Well,
01:54 the decision was a no-brainer
01:56 for sure.
01:57 We decided to join forces and to work together as AFD motto is world in common.
02:02 Together we are stronger,
02:04 Leon fela force on the Francais.
02:07 But what is this initiative about?
02:10 First,
02:10 it's about improving
02:13 knowledge of water resources.
02:15 You manage what you know and you know what you measure.
02:19 It is also about
02:21 extreme event
02:22 management.
02:24 Climate change forces us to work on the two sides of the coin
02:27 at the same time droughts for the past 7 years in Morocco and floods
02:32 as we are seeing right now.
02:35 At the end,
02:36 I guess the,
02:36 the main objective of the initiative is adaptation to
02:39 climate change and protection of biodiversity both in surface waters
02:43 as well as groundwaters.
02:46 Let me give you a few concrete examples of what the initiative will do.
02:50 We will strengthen hydrological networks.
02:53 We will work on industrial pollutions.
02:55 We will also work
02:57 to strengthen early warming systems against flood
03:00 and have natural-based solutions to protect against extreme events,
03:06 as well as reuse waters,
03:08 reinforcing the water police,
03:09 la police the law,
03:11 etc.
03:13 How will we do that?
03:15 Well,
03:15 AFD
03:17 has work on the preparation to appraisal and implementation,
03:20 but we decided to build
03:23 to do that in a Team Europe spirit.
03:25 So we invite our dear German and European colleagues,
03:28 KFW as well as the European Union.
03:32 And we also invited a newcomer,
03:35 our Italian counterpart,
03:36 the CDP,
03:37 the Casa de de depositia Prestiti,
03:39 to join the squad and to work with us
03:41 to try to match the ambition of the Moroccan government.
03:47 100 million
03:48 AFD plus 100 million KFW plus 100 million CDP plus
03:53 48 million from the European Union's total 348 million total
03:58 as a policy based loan.
04:01 But maybe the most important part of my story,
04:03 what will all this mean for Morocco?
04:06 Well,
04:07 first,
04:08 together,
04:08 together with what the World Bank is providing,
04:10 we are bringing on the table more than
04:12 700 million to implement the national water policy.
04:16 But
04:17 even more than money,
04:18 we are bringing expertise as well as interaction with the European institutions.
04:24 We are 4 institutions,
04:26 but for the Moroccan government,
04:27 we are
04:28 AFD is a one-stop shop,
04:30 so we are adding the numbers but decreasing the burden,
04:33 and it will cut tons of red tape.
04:37 And finally,
04:38 it's a little bit technical,
04:39 and I don't have the time to go into the details,
04:41 but we are providing the Moroccan government
04:44 with the EU grants,
04:46 what we call the financing not linked to cost modality,
04:48 which is an innovation not only in Morocco but worldwide for
04:52 and for the European Union.
04:55 It's we need the time to conclude the story.
04:58 First,
04:58 if there is one word that I want you to take away from this talk,
05:03 this is the one,
05:04 cooperation.
05:06 Cooperation is a powerful word.
05:08 You think better,
05:09 bigger,
05:10 and you,
05:11 when you,
05:12 when there is cooperation.
05:14 You raise expectation and you raise ambition with cooperation.
05:19 And finally,
05:19 let me conclude on a personal note.
05:22 Before ID,
05:22 I was a former diplomat.
05:25 I've been working on the climate change negotiations
05:27 to what became in 2015 the Paris Agreement,
05:30 and I've learned three key crucial elements from those negotiations.
05:36 I've carried,
05:37 I've carried them on with me
05:39 on to this day and to this initiative.
05:43 Listening,
05:43 inclusiveness,
05:45 and trust are key elements to success.
05:48 Thank you very much.
The Power of Water Partnerships
Timothée Ourbak, Head of Natural Resources and Biodiversity, AFD Groupe
When four partners came together to finance a water investment in Morocco, coordination did not add complexity, it made things simpler. AFD Groupe served as a "one-stop shop" that reduced administrative burden and streamlined engagement for the government. Drawing on lessons from the Paris Agreement climate negotiations, the speaker reflects on what makes partnerships truly work: listening, inclusiveness, and trust built over time.
00:00 We have the privilege of working with the ONCF,
00:04 the Moroccan Ministry of Economy and Finance,
00:07 and the Moroccan Ministry of Transport and Logistics
00:10 in the con in the context of the
00:12 Greater Casablanca Mobility and Logistics hub program.
00:17 They really are the most capable partners we could wish for,
00:21 and it's a real pleasure working with them.
00:24 So the ONCF
00:25 owns and operates Morocco's railway.
00:29 Every one of you should take the train in Morocco.
00:32 It's a really good service and it's becoming
00:34 even better under ONCF's leadership.
00:37 I agree,
00:38 Dominic.
00:39 The team at ONCF seem born for what they do,
00:43 and it shows in their excellent work.
00:45 They definitely are naturals,
00:47 but the same can't be said for me.
00:49 In fact,
00:50 I was never born to be a world banker.
00:53 Early in my career,
00:55 in fact,
00:55 I worked as an airborne firefighter
00:58 in the Western
00:59 United States.
01:01 This paid for engineering school.
01:03 It made for some really cool pictures that you can see here.
01:07 Uh,
01:08 er but most importantly,
01:09 it taught me a lot about the truly extraordinary things
01:14 that highly motivated and capable teams can accomplish by working together.
01:20 I love that I'm standing here with a former firefighter who became a
01:23 transport expert and then a task team leader at the World Bank.
01:28 And what's even funnier is
01:30 that I was never born to be a banker either.
01:32 In fact,
01:33 my background is in athletics.
01:36 I played semi-professional football for a very long time,
01:40 and although I retired from that career
01:43 a long while ago,
01:44 and I obviously never became a professional,
01:47 this taught me very early how important it is to
01:50 have a team when you want to accomplish extraordinary things.
01:55 So you see,
01:56 both Lena and I,
01:57 we came from backgrounds
01:59 where
02:00 working in teams was never a question,
02:02 never a topic of debate.
02:04 It was always the only way
02:07 to go.
02:09 And
02:10 looking at it from today's perspective
02:12 is actually not that different for large complex infrastructure projects.
02:17 You need to join forces
02:19 and you need a team
02:20 because you simply cannot achieve the same results alone.
02:24 That's right.
02:25 And in the case of the Casablanca program,
02:28 the KFW
02:30 and the World Bank were teaming up
02:32 because Morocco's vision to transform mobility is truly big
02:37 and bold.
02:39 It's centered on an $8.4 billion investment program,
02:44 and within this larger program,
02:46 there is a $1.1 billion component
02:51 for a new
02:52 uh commuter railway system in Casablanca.
02:56 This is similar in its technical scope
02:58 to a German S-Bahn system
03:00 or a French RER type system.
03:03 So,
03:05 let's technically put,
03:06 a community railway is a high frequency train service
03:10 that connects outer lying areas to central districts.
03:14 We cannot stress enough what this system is going to do for the people in Casablanca.
03:19 It will provide sustainable transport access to over half a million residents.
03:25 It will reduce CO2 emissions and help to fight air pollution
03:29 while more jobs,
03:31 more schools,
03:32 and more hospitals will become accessible within 45 minutes using this new system.
03:39 Commuter railways truly are powerful tools for development.
03:44 As Lena might say,
03:46 they level the playing field for mobility across income groups.
03:50 Rapid,
03:51 reliable
03:53 and affordable access
03:54 to the city center
03:56 is truly a game changer for people who live in outer lying districts.
04:00 That's especially true
04:02 when they need transport solutions other than buying
04:06 and using private cars.
04:09 Do you remember,
04:10 Dominic,
04:11 when we first got in touch about this program,
04:14 we agreed very early on that we would provide the best
04:17 support to ONCF in implementing this system if we team up.
04:22 Yeah,
04:22 I,
04:23 I do remember that,
04:24 but uh
04:25 I also remember
04:27 we were both a little bit,
04:29 how do I say this nicely,
04:30 we were a bit concerned,
04:32 yeah,
04:32 that getting
04:33 two
04:34 large
04:36 complex
04:37 organizations to align,
04:39 well,
04:39 it might be a little bit like.
04:42 Trying to to herd reluctant tortoises.
04:45 You,
04:46 you know,
04:47 theoretically that might be possible,
04:50 but,
04:50 um,
04:51 you know,
04:51 it just might not happen quickly
04:53 and it might not be
04:55 particularly elegant
04:56 when it does happen.
04:58 Right.
04:59 So
05:00 in German we would probably say
05:02 we didn't want to
05:04 for schlimbesser anything by teaming up or trying to align.
05:09 Oof,
05:09 you got me there.
05:12 So,
05:13 I think Ver Schlim Bessen is one of our better German word creations.
05:18 It describes something that you're trying to simplify or improve,
05:22 and although the attempt is well meant,
05:24 you accidentally end up making everything worse or more complicated.
05:29 Yeah,
05:29 absolutely none of that.
05:31 So,
05:33 how did the World Bank and KFW avoid messy complexity?
05:37 Well,
05:38 thankfully the two institutions
05:40 got together and they signed a co-financing framework agreement in 2024,
05:46 and that's really helped a lot.
05:48 Right.
05:49 It was said before,
05:50 we're gonna say it again.
05:52 It provides simplicity.
05:54 It allows us to rely on the same standards.
05:57 It allows the two banks to share documentation.
06:01 We share,
06:01 for example,
06:02 appraisal and environmental and social due diligence,
06:05 and we share key mechanisms
06:07 such as funds flow,
06:09 audits,
06:10 program monitoring,
06:11 or disbursement procedures.
06:14 And thanks to this model.
06:16 Clients
06:17 such as Morocco
06:19 are able to access more support
06:21 for virtually the same level of effort.
06:26 If you ask us,
06:27 we think this approach works well.
06:30 Why?
06:31 Once again,
06:32 it simplifies life.
06:34 And
06:35 having a smooth administrative machinery
06:38 allows us
06:39 to focus more on the program
06:42 and on the partners we're working with.
06:44 You know what,
06:45 it also
06:46 provides for complementarity.
06:48 So for example,
06:49 in Casablanca,
06:51 the World Bank's financing is prioritizing institutional modernization,
06:56 whereas KFW
06:58 is prioritizing green infrastructure.
07:01 This bit of diversity between us
07:03 allows for a comprehensive coverage
07:07 across different needs.
07:09 Right.
07:10 And the same also goes for the resources
07:12 that both banks have beyond the program financing.
07:16 For example,
07:17 in our case,
07:18 World Bank will provide technical assistance on fares and ticketing,
07:22 and KFW will be providing technical assistance
07:25 on digitalization and enterprise resource management.
07:29 So,
07:30 look,
07:30 I think as you can see,
07:32 we have a lot of confidence in the co-financing model
07:36 and its ability to help countries achieve their objectives.
07:41 And in fact,
07:42 you know,
07:43 those of you in the audience who take key decisions,
07:47 well,
07:47 you should be asking people like us
07:50 to do more for you
07:51 together.
07:53 Because
07:53 co-financing
07:55 can and should become a norm,
07:57 especially when you need big teams
07:59 to support big,
08:01 ambitious things.
08:03 And
08:04 don't get us wrong.
08:06 Despite all excitement,
08:08 we're not saying the model is perfect.
08:10 So
08:11 our ask is pretty straightforward.
08:14 Keep doing what you're doing for example today.
08:17 Keep exchanging.
08:19 Help us refine the model,
08:21 make it even simpler.
08:22 Make it more seamless.
08:24 And also
08:26 expand the number of people who have experience in
08:28 teaming up for big projects like Dominic and I.
08:33 It's like with football or any other sports really,
08:36 if you think about it.
08:37 You need practice
08:39 because practice makes you a better player
08:41 and a better teammate.
08:43 Yeah.
08:44 And speaking of teammates,
08:46 we want to acknowledge Mr.
08:47 Nabil Samir,
08:49 who's not with us on this stage today.
08:51 He's a,
08:51 he's a key one of our teammates.
08:53 And furthermore,
08:55 we want to thank again the Ministry of Transport and Logistics,
08:58 the Ministry of Economy and Finance,
09:00 and most definitely,
09:01 we want to thank the ONCF
09:03 for their strong leadership on the CASA hub program.
09:07 As KFW and the bank seek to do more together,
09:11 having that strong client side leadership helps
09:14 ensure that we ourselves remain aligned.
09:17 And I would like to thank once again you,
09:19 Dominic,
09:20 and the World Bank Group for being our teammates,
09:23 not only in this specific program but in so many others.
09:27 Thanks again,
09:28 Lena.
09:28 You and the KFW team have been terrific.
09:31 Thank you very much.
09:33 Don't forget to take the train to Morocco.
09:35 Hey.
Greater Casablanca Mobility and Logistics Hub
Dominic Pasquale Patella, Senior Transport Specialist, World Bank, and Lena Keicher, Portfolio Manager, KfW Development Bank
Large infrastructure projects test the limits of what any single institution can deliver. The World Bank and KfW are working together to help Morocco transform urban mobility through the CasaHub program. At the heart of their collaboration is the KfW–World Bank Co-Financing Framework Agreement, which allows both institutions to rely on shared standards and procedures—cutting complexity for teams while reducing the burden on the government, and ultimately helping to deliver better services for commuters and businesses.
00:01 That it's going to prosper in the long run.
00:02 So,
00:03 the prosperity of the marketplace is pretty crucial for the platforms to
00:07 achieve long-term profit maximization.
00:09 So nowadays,
00:10 as a result,
00:10 we do see a lot of these leading e-commerce platforms like Amazon or Alibaba,
00:14 they engage in a lot of efforts to support their third-party sellers.
00:19 And so in this paper,
00:20 we are going to look at
00:21 a particular effort the platform undertake which
00:24 is really a large-scale business training program.
00:26 And the goal of the program is to help
00:28 the firms overcome the growth barriers when they enter,
00:31 and we are going to study the welfare impact of this program
00:34 on different participants in the market.
00:37 So,
00:38 thinking broadly about the interventions to support the SME I think that's a um
00:44 really a leading
00:45 topics in many of the development research.
00:48 But
00:48 based on the existing evidence,
00:50 especially if you're thinking about the context of business training,
00:52 I think there's still
00:54 quite a bit of uncertainty left.
00:56 I think,
00:56 um,
00:57 most recent meta-analysis by David McKenzie,
01:00 uh,
01:00 summarized the impact on business training.
01:02 I think overarching the conclusion is
01:04 there are small positive impacts on the profit,
01:07 but because each individual study tend to have rather
01:10 small sample size and there's significant underlying variations,
01:14 it's a little bit hard to
01:15 detect um
01:16 that impacts for individual studies.
01:19 And we also,
01:20 the other worry that we often case have is what if the
01:23 success of one group come as a cause of the other group.
01:26 So for example,
01:27 what if um all these revenue and profit
01:29 growths come as stealing business from the competitors.
01:32 So,
01:33 then the idea is really to separate out the business stealing
01:36 and um being able to say something about what if there's,
01:39 are there any possibility of market expansion.
01:41 So in the previous analysis,
01:43 we tend to generate some market-level variations
01:46 and to see um if we can vary.
01:48 For example,
01:49 um,
01:49 in the experimental study,
01:51 um,
01:53 researchers randomized
01:54 the various treatment intensity of a business training
01:57 program across different rural markets in Kenya.
02:00 And so I think the finding there is the
02:02 business training foster market expansion but at limited cost of
02:05 spillover on the competitors.
02:08 And in this program,
02:09 we are going to actually take a different approach
02:11 by looking at the other side of the market.
02:14 Which are the consumers who interact with these firms.
02:17 I think the consumers are a relatively understudied group,
02:21 but by directly looking at their experience,
02:23 we will try to address the question such as,
02:26 um,
02:26 could the consumer losing out due to potential negative selection on the firm side
02:31 where the training
02:32 keep um lower quality and efficient sellers on the market longer.
02:36 So,
02:36 uh,
02:37 what are the impact on the consumers?
02:40 So,
02:41 in a nutshell,
02:41 what we are trying to achieve in this paper is to
02:45 use a randomized control experiment that we implement in
02:48 partnership with this leading e-commerce platform in China.
02:51 And we will try to answer three questions.
02:54 So the first thing we want to know is,
02:56 can the training actually lift the growth barrier that the new solar face?
02:59 And if so,
03:00 uh,
03:01 what are the,
03:02 what is the underlying mechanism?
03:03 What is the,
03:04 um,
03:04 what is the approach?
03:06 And then the second thing we want to find out
03:08 is how does the training actually affect consumers' experience,
03:11 um,
03:12 especially when the consumers now interact with different types of firms.
03:15 And lastly,
03:16 we also want to quantify the welfare implications of the training and,
03:19 um,
03:19 think about,
03:20 uh,
03:20 decompose that and think about the impact on different
03:23 participants in this market.
03:27 So to answer this question,
03:28 we are going to adopt the following set of the empirical strategies.
03:32 The first thing we want to look at is,
03:34 of course,
03:34 um,
03:35 leverage the experimental variations that we have from the RCT
03:39 and to directly and causally estimate the impact of the training on the sellers.
03:43 And because the particular design of the program,
03:46 um,
03:46 the way we set up the experiment is we
03:48 randomize access to the training as a whole package
03:51 when the new seller just entered the market.
03:54 And so in our final example,
03:55 we have slightly over 700,000 sellers
03:57 and we assigned 25% of them to the treatment group.
04:01 And so to really,
04:02 and then to really study the impact on the consumer side,
04:05 we are going to be taking advantage of this very detailed
04:08 consumer and seller match search and browsing data.
04:10 So with that data,
04:11 we are able to explore the variations in
04:13 the composition of sellers that the consumer visit
04:16 when they search for product on the platform.
04:18 And so then we can compare consumers' experience
04:20 when they interact with different type of sellers.
04:23 And lastly,
04:24 we also build a structural model to quantify the welfare.
04:26 So in the model,
04:27 we characterize consumer demand and also the
04:29 matching between the consumer and sellers.
04:31 And we use a counterfactual exercise to think about
04:34 what if the training is not being made available to these sellers.
04:38 And so now,
04:39 let me give you an overview of what we found to answer that three
04:43 man-made research question that we,
04:44 we have.
04:46 So the first,
04:47 um,
04:48 uh,
04:48 so the first answer I want to provide is really on the seller side.
04:51 So overall,
04:52 we found these treated new sites are earning slightly higher revenues
04:56 and,
04:57 um,
04:57 that happened because they're attracting more visitors to their site.
05:01 So given that change on the performance,
05:03 we look at the outcome,
05:04 we look at their strategy,
05:06 and we see that
05:07 most of the strategy change concentrated on the fact that
05:10 these treated news are to become more engaged in marketing and promotion,
05:14 and to a small extent,
05:15 they also improve their customer service quality.
05:18 So that's on the seller side.
05:20 And because the traded new seller attract more visitors to their site,
05:23 that implies the consumer now become more likely to
05:26 interact with these new sellers because of the training.
05:29 So then the question we want to know is
05:31 how consumers' um experience compare when they
05:34 interact with different type of sellers.
05:37 And,
05:37 uh,
05:37 our main takeaway is really that the consumer actually
05:40 have better matching outcomes with the treated new sellers
05:43 as opposed to,
05:44 in particular,
05:44 the incumbents.
05:46 And here we have 3 pieces of supporting evidence.
05:49 The first thing we found is that the consumer are just
05:51 more likely to make a purchase when they visit new sellers,
05:54 either treated or controlled
05:55 compared to the cases when they only visit incumbents.
05:59 And when visiting both new sellers and
06:01 the incumbent during the same search session,
06:03 the consumers are more likely to choose
06:05 these treated new sellers over the incumbents.
06:08 And lastly,
06:09 when the consumers are now making purchase from the new sellers,
06:12 they actually didn't sacrifice their purchase quality
06:15 because they are no more likely to
06:16 request returns or refund or um they're as likely to make a repeat purchase.
06:22 So,
06:22 so now,
06:23 I mean that's uh what we found on the reduced form,
06:25 but eventually we,
06:26 we also want to quantify the welfare impact.
06:29 So,
06:30 for this part,
06:31 we found um the training generated a small,
06:33 about 0.1% increase in the consumer surplus
06:36 and to a less extent also um higher s higher total revenues.
06:40 And the reason that the training can generate higher consumer surplus
06:44 is really by reducing the search frictions that these new sellers are facing
06:48 and as the training improves alignment of visibility and quality.
06:52 So by that,
06:53 I simply mean the training makes these higher quality new
06:56 sellers more likely to be found by the consumers.
06:59 And I'm going to give you a much precise definition
07:01 about what we mean by the quality in this compounds.
07:05 And as an overview of the results,
07:07 um,
07:07 uh,
07:07 like,
07:08 let me give you,
07:09 um,
07:10 Um,
07:11 a sense of where we think this paper fit in the literature.
07:14 And then broadly,
07:15 we will contribute to three main strands of the literatures.
07:18 So the first line,
07:20 um,
07:20 we are trying to,
07:21 we're speaking to is this literature investigating the
07:23 nature of firm dynamics and growth barriers.
07:25 And here,
07:26 we are really highlighting the importance of search frictions
07:29 for these new firms in the online market and that echoes recent finding
07:33 about the demand side friction and the importance of accumulating customers.
07:37 And also the fact that um adopting a growth-oriented strategy by investing in more
07:43 marketing and accumulating customers
07:45 um really
07:46 proves the potential of that particular approach
07:49 compared to the other type of approach which is really on cost saving.
07:52 I think that's
07:53 also echoes some of the,
07:54 even though we can't really compare the effects
07:56 of these two strategies directly as the Anderson Chandy
07:59 Zia paper did
08:00 in their offline experiment,
08:02 but at least in this online context,
08:04 we're highlighting that
08:05 marketing could be a very important approach.
08:08 And compared to the earlier paper by Bai Chen Liu and Xu,
08:11 I think
08:12 what we are
08:13 trying to do is we not only identify that
08:15 search and information friction is an important barriers,
08:18 we actually also go one step further
08:20 and telling you an effective strategy to address that friction,
08:23 which is really this business training program.
08:26 So then we contribute to the second literature uh uh looking
08:29 at the effectiveness of business training on helping the SMEs.
08:33 So here,
08:33 I think we're making two contributions.
08:35 The first is really on the empirical side where
08:38 we provide perhaps one of the first evidence looking
08:41 at um the impact of a large-scale business training
08:44 for the online firms.
08:45 But second,
08:46 I think the more important contribution we're making here
08:49 is really to think about a new welfare channel
08:51 and to think about the impact of the training that could help on the consumer side.
08:56 Because the training now reduces search frictions
08:58 on the consumer side that eventually improves the consumer's
09:01 um outcomes by
09:03 getting better matches.
09:05 And so for the last part of the literature,
09:08 um,
09:08 that we think we're also broadly,
09:10 um,
09:11 concerned about the role of the platforms and the process of digitization
09:15 in relation to the economic development.
09:17 I think our overall findings that the training actually helps
09:21 the new sellers and the consumer as well
09:23 show that the platform should be
09:25 And should and actually can be more active in
09:28 its governments and in supporting the services participant in housing
09:32 because that's incentive compatible
09:33 with its own agenda.
09:34 So that's,
09:35 um,
09:36 that implies
09:37 um the type of
09:39 the type of collaboration that should be really
09:41 be fostering between the public and private sector.
09:46 I mean that's the point where I could stop and take questions.
09:49 OK,
09:49 Patricia,
09:50 so I,
09:51 yes,
09:51 this is a good place
09:53 to pose.
09:53 Uh,
09:54 anyone has questions for Patricia?
09:57 I don't see any hands here.
10:04 Oh,
10:04 OK,
10:05 let's wait.
10:07 OK,
10:07 it looks like.
10:09 There is none,
10:10 so you can,
10:12 OK,
10:13 so I,
10:13 I'm sure people will have more questions when I start to talk about the contest,
10:16 but
10:17 honestly,
10:18 please ask me a question.
10:19 Uh,
10:19 I'm sorry.
10:21 Uh,
10:21 I was seeing 3 hands,
10:23 so I'm wondering your ability to see hands at the top of your participant list.
10:28 OK,
10:28 they were at the very bottom.
10:29 So,
10:29 uh,
10:30 let's give them a chance then.
10:31 So it hasn't been moving.
10:32 They are still at the very bottom.
10:33 OK,
10:34 so,
10:35 uh,
10:36 key,
10:36 why don't you ask the first question.
10:39 Yes.
10:40 Uh,
10:40 can you hear me?
10:42 Uh,
10:42 yes,
10:43 yes,
10:43 go ahead.
10:45 OK.
10:45 Now,
10:46 my question is,
10:46 uh,
10:47 do you have,
10:47 uh,
10:49 information about the type of product
10:51 that
10:52 may have differential impact in the experiment?
10:56 Certain products may be just a one-shot,
10:58 um,
11:00 purchase,
11:01 so you do not require
11:03 repeated
11:04 interaction.
11:05 Other products may be.
11:08 Is more long-lasting.
11:10 It require more long-term kind of relationship.
11:13 So it's a
11:18 Uh,
11:22 Sorry,
11:22 I,
11:22 I,
11:22 I didn't hear the end of the question,
11:24 but I think
11:25 regarding the differential impact on different sectors and different industry,
11:29 I think,
11:29 um,
11:29 there are two,
11:30 so I have to
11:31 answer to,
11:32 to that question in two parts.
11:33 So first,
11:33 we look at the differential treatment effect for firms in different industry,
11:38 and we don't find a differential impact on that front.
11:41 And then on the consumer side,
11:42 um,
11:43 you will see,
11:44 you,
11:44 you will see more clearly about our the empirical approach because we,
11:47 we actually took,
11:48 uh,
11:48 we actually include the
11:50 search keywords as a fixed effect.
11:52 So right now,
11:52 uh,
11:53 everything is comparing within the type of product that is happening,
11:56 but I totally agree,
11:57 the type of heterogeneity would be something very
11:59 interesting to look at down the road.
12:03 Thank you.
12:05 OK,
12:05 Paulo.
12:07 Next.
12:08 Hi,
12:08 Patricia.
12:09 Uh,
12:10 so I just,
12:10 I have a question about,
12:12 uh,
12:12 what,
12:13 what can we learn from the intervention.
12:14 So there are many potential entrants and many platforms.
12:18 So is it,
12:19 this,
12:19 the results will point to
12:21 something that can be done at,
12:23 at the country level by the government,
12:25 or can we,
12:26 can we
12:27 expect that to inform that kind of policy or not?
12:30 Uh,
12:30 and also,
12:31 can you tell us a bit about the costs
12:33 versus the benefits of this type of intervention?
12:36 OK,
12:37 great.
12:37 So,
12:37 I think my policy recommendation for this question is
12:41 for
12:42 any single e-commerce platform,
12:43 so for any single type of onboarding program that you may,
12:46 you may,
12:47 the country may want to implement,
12:48 it's worthwhile to think about what happened after they get on board.
12:51 And I always think it's important.
12:53 And then it's become important for the for the country
12:56 to partner to partner with these platforms to,
12:59 you know,
12:59 to help the
13:00 participant overcome that particular barriers.
13:02 I think a lot of the skills that we are trying to train um the.
13:05 Here are specific to these
13:08 e-commerce sellers on this platform,
13:09 but they're also generalizable components for other sellers.
13:12 For example,
13:12 the importance of marketing.
13:13 I think that's some of,
13:14 some of the transferable,
13:15 uh,
13:15 message.
13:16 But then I think the main takeaway is really the importance of thinking about,
13:20 you know,
13:21 like close,
13:21 like helping people like thinking about like
13:23 you need to implement strategy to help people
13:26 after they've got,
13:27 got on board.
13:28 Getting on board is just not enough.
13:29 And uh think of,
13:30 and then the importance of building like private
13:33 partner partnership.
13:34 So on the cost and benefit analysis part,
13:36 um,
13:38 overall I think the additional revenues that are generated by these,
13:41 the 700,000 salaries that we check,
13:43 um,
13:44 in our experimental analysis about like $2 million
13:47 but the cost of the programs really involved paying the salary of some engineer.
13:52 So
13:53 even for this sample,
13:54 the,
13:54 the,
13:55 the benefit already
13:56 outweighs the cost.
13:57 And since the cost,
13:58 uh,
13:59 since the,
14:00 the cost of further implement the program and to just expand it to everyone,
14:04 it's going to continue to go down.
14:06 Um,
14:07 if you just expand,
14:08 expanded
14:09 due to the low dissemination costs in this online environment.
14:11 I think
14:11 even from the platform's perspective,
14:13 this is a
14:15 profitable business to engage in.
14:22 OK.
14:23 Uh,
14:23 I don't see any more hands,
14:26 so you can continue.
14:27 OK.
14:28 So,
14:29 yeah,
14:29 thank you for your question and like
14:30 please just um ask a question about clarification
14:34 type of thing,
14:34 especially when I talk about context.
14:36 I,
14:36 I could
14:37 totally overestimate people's knowledge about this particular setting.
14:40 I'm,
14:40 I'm just a little bit too familiar with it.
14:42 But anyway,
14:43 so now let me tell you about the context and the and the design
14:46 of the training and before moving on to the results on the seller,
14:49 consumer,
14:49 and the welfare.
14:51 So we are working with the leading e-commerce platform in China,
14:54 which shall not be named,
14:55 but you probably already guessed.
14:57 So this platform currently has about 9
14:59 million active sellers and over 700/700 million
15:03 active consumers.
15:04 The platform is mostly domestic.
15:06 Some of us are doing engage in like import export,
15:09 but I would say the vast majority of transactions happen within the border.
15:13 And um the platforms account for about 62% of market share in
15:17 e-commerce in 2019 and that's about 16% of all retail in China.
15:21 So
15:22 it is a very meaningful market to look at by itself.
15:25 And,
15:26 um,
15:26 and the major difference between these platforms and Amazon,
15:29 which I think people are more familiar with,
15:32 is really the fact that this platform does very limited direct
15:34 sales to the consumer but mostly holds the third-party sellers.
15:38 And then the platform also source revenues
15:40 from the sellers by selling them advertising,
15:42 charging their commissions,
15:43 and also,
15:44 also offering supplementary service which we
15:46 actually analyze in another project.
15:48 So,
15:48 so from the incentive on the platform,
15:51 It's just important for these,
15:52 um,
15:53 for the,
15:54 for the platform to make sure that the seller can,
15:56 can grow and can survive because otherwise they will lose their stream of profit.
16:01 And so they do engage in a lot of efforts to
16:03 make sure that happens.
16:05 So then,
16:05 what does it mean to be a seller on the platform?
16:08 And here,
16:08 um,
16:09 I want to highlight two things.
16:10 So one thing is becoming a seller is easy but difficult at the same time.
16:14 The easy part will come from the low entry cost.
16:17 Basically,
16:17 right now,
16:18 everyone with a valid
16:20 ID card,
16:21 as uh like an individual ID card or or business registry
16:25 and a bank account can set up the
16:28 account to be,
16:28 to be a seller on the platform in say another 5 minutes.
16:31 And the monetary and effort cost of becoming a sellers to enter the market is,
16:36 is very low.
16:37 And most of these sellers are very small.
16:40 A lot of them just operate by one person and um
16:42 they tend to be retailers rather than manufacturers or wholesalers.
16:46 And they offer,
16:47 as I was talking about,
16:48 they offer very diverse set of products.
16:50 So,
16:51 um,
16:51 the idea is everything that is legal to sell in this country,
16:55 you can find it on the platform,
16:56 including a house.
16:57 So,
16:59 that's kind of crazy.
16:59 But
17:00 the hard part about becoming a seller in the platform is really to
17:05 Attract visitors and actually to earn money.
17:08 And because there are a lot of sellers on the platform,
17:11 the competition is very intense and the main type
17:13 of competition is really on attracting consumers' attention.
17:18 So in this process,
17:19 both of uh platforms and sellers are taking effort and the platforms,
17:22 what the platform does is it try to improve
17:25 the efficiency of the search and recommendation algorithm to,
17:28 to achieve the most efficient matching outcomes.
17:31 And then the seller can actually actively engage and
17:34 um
17:36 engage in that process
17:37 by
17:38 um doing search advertising,
17:39 by doing more promotions,
17:40 and right now the trending thing is to hire an,
17:43 an influencer and celebrity
17:45 to advertise the product on your behalf.
17:47 So a lot of actions happen on the uh on the server side,
17:50 um,
17:51 but
17:51 that also suggests the intensity of the competition there.
17:55 So,
17:56 given all this challenge the new sellers are facing,
17:59 um,
17:59 then the platforms,
18:00 do you want to think about ways to,
18:03 you know,
18:03 help these firms overcome the growth barrier.
18:05 So at least they know how to operate on this platform.
18:08 So which comes with the discussion about
18:10 our business training program.
18:13 So the training actually borrows quite a bit of lessons from the literature,
18:17 but in a nutshell,
18:18 um,
18:19 I will summarize the training as a customized task-based system
18:23 where
18:23 the training is organized as a sequence of individual tasks,
18:27 and
18:27 these tasks,
18:29 each of them tackles a specific challenge that the seller may face
18:33 in the growth trajectory.
18:34 And the customization part comes in because the platform actually
18:37 use an algorithm to dynamically match the sellers to the task
18:42 based on their performance at the moment.
18:45 So then each seller is going to get
18:47 assigned with a customized list of tasks and so that
18:51 assignment process actually I would say achieve two goals.
18:54 So one thing is
18:55 um
18:56 It really serves a very structured reminder to,
18:59 to help these firms point out what are the something
19:01 that they should be working on at that particular moment.
19:04 So it's like identifying where the problem is.
19:07 And,
19:08 um,
19:08 and then
19:09 after identifying the problem,
19:10 if the sellers want to take up the task and they're going to work on that challenge,
19:14 they can take up the task and then access the corresponding
19:18 tutorials and other supporting material that is associated with this tasks.
19:23 And so if the sellers are able to complete the task on time,
19:28 they're going to earn a small rewards and that rewards
19:30 usually comes in the forms of some short-term access to a
19:34 um supplementary service like,
19:36 um,
19:37 uh,
19:38 a little tools to help you print
19:40 multiple shipping labels with one click and those,
19:42 um,
19:43 so some,
19:44 sometimes service to help you streamline the business operation.
19:48 And so that's the basic structure of the,
19:50 of the design of the task and um of the training program.
19:53 And one thing I do want to highlight is this training program
19:56 is completely independent from any other aspect of the platform operation.
20:00 So for example,
20:01 the search and recommendation algorithm does not use any information
20:05 about from sellers participation in the training to determine the ranking.
20:10 So now,
20:11 so that's the design,
20:12 but now let me tell you what exactly the training is trying to teach these sellers.
20:16 So I would summarize the training as a very,
20:19 as focusing on very practical online business operational skills
20:22 and um broadly we classify that into three main categories.
20:26 So the first type first type of task
20:30 teaches the seller basic techniques to set up their own online business.
20:33 So that includes
20:35 identify them the performing tricks to,
20:37 uh,
20:38 when the seller posts product.
20:39 So for example,
20:40 what are the languages should be used,
20:41 how to,
20:42 uh,
20:43 shoot a better photo,
20:44 and how to write a better description,
20:46 and so on and so forth.
20:46 So that's a basic setup type.
20:49 And then the second main focus is really to teach a
20:52 seller necessary marketing skills and techniques to attract more visitors.
20:56 So that could include uh helping the sellers to
20:59 um get an idea about designing better titles for their product
21:03 or getting,
21:03 um,
21:04 um,
21:05 to find a better
21:06 keywords to bid on
21:07 so that,
21:08 um,
21:08 they can get a higher ranking in the search outcome.
21:11 So that's the second type of marketing and promotion.
21:14 And then the last part of the training really focused on improving the
21:18 customer service quality and man and manage customer relationship.
21:22 So that includes um when the consumer
21:24 make contacts with the customer service agent,
21:26 what are the ways you should be
21:29 addressing this consumer,
21:29 what kind of language you should be using,
21:31 and a very interesting task actually helps us out to set up an AI
21:34 system so that AI assistant can handle some of the basic customer inquiries.
21:39 And so this is quite different from the typical type of me.
21:52 It's specific
21:54 to the e-commerce setting.
21:57 So that's the basic content of the task.
22:00 And then
22:01 everything here we're talking about is completely online and then the
22:05 way the seller access it is through this official seller portal.
22:08 So let me show you a demo of how they look like.
22:10 So
22:11 every seller's,
22:12 um,
22:12 if you're dedicated to run your business,
22:15 um,
22:15 you essentially need to install this app.
22:18 Um,
22:19 so that they can actually communicate with the platform.
22:22 And so then the training will show up as one of the module here on the front page
22:27 and you will see some of the tasks being assigned to you at this particular moment.
22:31 Um,
22:31 and then once you decide to take up um the task,
22:35 you can see,
22:36 uh,
22:36 what is the target,
22:37 why you are being assigned with that task,
22:39 what is the current level of performance,
22:40 you keep track of the performance and you can also get access to the,
22:43 to the goal you want to reach,
22:45 and then the supporting material.
22:46 So everything here is online.
22:48 And also because this app is very essential to Sara's day to day operation,
22:53 the additional cost of participation in the training
22:56 is actually quite low.
23:01 And just a quick overview of um the training program.
23:04 And because the training program is designed in a customized way,
23:09 the way we couldn't really randomize the access to the,
23:12 to the training
23:13 um
23:14 by individual tasks.
23:15 So then the way we set up experiment
23:17 is we randomize new sellers access to their training
23:20 as soon as they complete the registration process
23:22 with the platform and we think about the training as a whole package.
23:26 And um so in our study cohort,
23:28 we assigned 25% of them to receive the access.
23:31 So our cohorts cover the cell who entered
23:33 the platform between May and mid-August in 2019.
23:36 So that's pre,
23:37 pre-COVID.
23:40 So now,
23:41 looking at the take-up rate,
23:42 and here we define the take-up as taking up some of the tasks.
23:46 The take-up rate is about 25%,
23:48 which is certainly not very high compared to some of the offline training.
23:53 But one reason of the low take-up is because a lot of it is Sellers are,
23:58 I would,
23:58 we only call them the potential sellers since they just
24:01 enter the platform to take a look.
24:03 In fact,
24:03 um,
24:04 less than 50% of the sellers
24:06 post any product for sale
24:08 within the next 9 months.
24:10 So that already expanded a lot of the,
24:12 um,
24:12 low take-up
24:13 there.
24:15 And um we also find there's quite a bit of
24:16 variations as you imagine in seller's participation in the training.
24:21 So measuring by how many tasks the seller ended up taking,
24:24 you see there are quite a bit
24:25 of seller who just take up one task and to check it out.
24:29 But there are a lot,
24:30 um,
24:30 there's another group of very intense player who took up over 10 tasks and then the,
24:35 um,
24:35 one seller ended up taking over 50 tasks over,
24:39 over the 9 month period,
24:40 right?
24:40 So there's like quite a bit of variation down there.
24:44 And then
24:45 the other thing I do want to mention
24:47 is think about the long-term versus short-term gains.
24:49 So
24:50 most of the actions about
24:52 training participation really concentrate during the first month of entry.
24:55 In fact,
24:56 over
24:57 like nearly 50% of the tasks were taken up during that very first month.
25:00 And there's definitely selection into who will become a participant,
25:04 um,
25:04 on the,
25:05 um,
25:06 on the demographic
25:07 and,
25:07 uh,
25:08 on the characteristic side
25:11 as well.
25:13 So,
25:13 and that's an overview of um
25:16 of the training program and the experimental design.
25:19 So before moving on to the results,
25:21 let me
25:21 just give you a quick overview of what kind of data we have access to.
25:26 So our main data source are the administrative data that this platform regularly
25:30 collects and we have information on the seller and the consumer side.
25:34 And on the seller side,
25:35 we know their performance,
25:36 we have some indicator for their strategy,
25:38 characteristic,
25:39 and their participation in the training.
25:41 So the performance part,
25:42 um,
25:43 We,
25:44 we certainly have some Patricia,
25:44 yeah,
25:46 uh,
25:46 before,
25:46 uh,
25:47 you,
25:47 you tell us about your date,
25:49 I think this is probably a good time to pause for a bit and,
25:51 uh,
25:52 see if there are questions.
25:54 Uh,
25:54 Miriam,
25:55 can you go ahead and ask a question?
25:57 Yes,
25:57 yeah,
25:57 hi,
25:57 thank you so much.
25:58 Um,
25:59 Patricia,
25:59 I had two questions.
26:00 One,
26:00 when you just showed,
26:01 you know,
26:02 a lot of them only take one task,
26:04 is that typically the first task,
26:06 or is,
26:06 is there any pattern in terms of like which task
26:09 they typically do?
26:10 And my other question is,
26:12 Like,
26:12 to which extent is this information available elsewhere?
26:15 Like,
26:15 is there some sort of help section on the platform where
26:18 sellers could in theory have gotten a lot of this information,
26:21 or
26:22 it,
26:22 it goes back a bit to this question of,
26:24 well,
26:24 why don't people just,
26:25 you know,
26:26 seek out this type of information that
26:28 they can gain through business training themselves.
26:31 Yeah,
26:31 yeah,
26:31 quick question.
26:31 So,
26:32 um,
26:32 on the first question.
26:33 Um,
26:34 we,
26:35 there's a,
26:36 a little bit of variations about like which tasks they are,
26:38 but,
26:39 uh,
26:39 for those who only take up one task,
26:41 that task is very much likely to be the
26:43 one that teaches very basic skill about store setup.
26:47 Um,
26:47 so there are a lot of variation within that category,
26:49 but I think that's we account for the majority of,
26:52 you know,
26:52 like the,
26:53 uh,
26:54 The limit,
26:55 the the low participation type.
26:58 So that's,
26:58 that's first.
26:59 And then
27:00 a lot of these other information,
27:01 whether the information is available elsewhere,
27:03 the answer is yes.
27:05 If you go search hard enough,
27:06 these are definitely not like rocket science.
27:08 You can definitely,
27:08 you can find it
27:10 perhaps not through the platform,
27:11 but if you just Google it on the internet,
27:13 there's a possibility to find a similar type of knowledge.
27:16 So,
27:17 so
27:17 that as a result,
27:18 I would say um the training.
27:21 Um,
27:22 the training for,
27:23 for the sellers who
27:24 could potentially look into that question and then the training really does,
27:28 um,
27:29 the,
27:29 what the training achieve is really
27:32 structure that knowledge
27:34 and tell and,
27:35 uh,
27:35 really simplify the,
27:36 you know,
27:37 the process of actively seeking that knowledge,
27:39 but instead providing the relevant information to the seller at relevant time.
27:43 I think this is a major,
27:45 um,
27:45 difference that this uh this particular training is doing.
27:50 But of course,
27:50 there are a lot of sellers who
27:52 perhaps don't have the skill of actually do the online searching.
27:55 Actually,
27:55 Google doesn't
27:56 exist in China.
27:57 So,
27:57 uh,
27:58 perhaps they,
27:58 they don't have great access to other search engines,
28:00 so there could be some knowledge gap.
28:02 And then for these group of sellers,
28:04 then I would,
28:04 I would argue there could be
28:07 real knowledge um being shared by this training,
28:09 but
28:10 it's
28:10 hard for us to actually separate these two possibilities.
28:17 Xvi,
28:18 you're next.
28:20 Hi,
28:20 yes,
28:20 uh,
28:21 uh,
28:21 thanks.
28:21 Um,
28:22 so,
28:22 so I guess,
28:23 uh,
28:23 could you tell us a bit more about
28:25 the contract between the platform and these vendors?
28:29 I mean,
28:30 is there a heterogeneity?
28:31 Is there,
28:32 is there a way where,
28:33 where,
28:34 you know,
28:34 uh,
28:35 by paying a bit more,
28:36 you know,
28:36 by,
28:36 by,
28:37 by,
28:37 you know,
28:38 giving away
28:40 A larger share of the,
28:42 you know,
28:42 revenues of a transaction,
28:44 you know,
28:44 I can get
28:45 a better placement in the in the web,
28:48 or,
28:48 you know,
28:49 higher in the list of products that are selling after,
28:52 you know,
28:52 after a search.
28:53 I mean,
28:53 is there any heterogeneity there that,
28:55 that,
28:55 that,
28:56 um,
28:56 you know,
28:56 that vendors can exploit?
28:59 So within the sample of seller that we look at,
29:01 I would argue there's,
29:02 there's none.
29:03 Like people face a same contract as long as they're in the same product category.
29:08 But broadly speaking,
29:09 there are really two types of sellers on the platform.
29:11 So the type of seller we look at are the
29:14 seller without a brand who register as an individual with the small firms.
29:18 And so they account for the majority of the seller on the market,
29:20 but
29:21 they are a small group of sellers
29:23 who are really the brand.
29:24 And the seller.
29:25 So,
29:25 for example,
29:25 Nike and Uniqo and this like internationally
29:28 recognized brand are also on those platforms.
29:30 So they have their special demarcation,
29:32 they have their special contract,
29:33 and so they're sort of competing with the,
29:35 with the small seller in the same market.
29:37 So that's,
29:38 um,
29:38 so there are different types of contracts.
29:40 But within our experimental example,
29:41 we only focus on the small sellers
29:44 who,
29:44 um,
29:45 who are not branded,
29:46 um,
29:46 who,
29:47 who have basically the same type of set of contracts.
29:51 And so,
29:52 seller can spend money to get better rankings in the search outcomes,
29:55 and this is actually one of the main channels
29:57 that we're going to be talking about right now
29:59 is
30:00 they spend money on advertising
30:01 and they do promotions.
30:07 Uh,
30:08 Art,
30:08 your next,
30:09 uh,
30:09 by the way,
30:09 uh,
30:10 once your,
30:11 uh,
30:11 questions have been answered,
30:12 please lower your hands.
30:14 Thanks.
30:15 Um,
30:15 hi,
30:16 Patricia.
30:16 Um,
30:16 I wanted to follow up a little bit on your exchange with Miriam.
30:19 I'm trying to understand the information that is,
30:22 uh,
30:23 conveyed through the training.
30:25 Because,
30:25 and,
30:26 um,
30:27 I,
30:27 I'm not sure if this is a good example,
30:28 but imagine one of your tasks is,
30:30 you know,
30:31 following up on customer complaints.
30:33 Then
30:34 your training gives advice on how to follow up with customer,
30:37 uh,
30:38 complaints.
30:38 But I was wondering if you would get a similar effect
30:41 just by reminding entrepreneurs that customer complaints are a
30:45 bad thing and you should deal with them,
30:46 right?
30:46 So you could imagine having
30:48 a treatment where for some guys,
30:49 you just say,
30:50 hey,
30:50 you should pay attention to customer complaints,
30:52 and the other one is,
30:53 and here's how you do it.
30:55 And it'd be interesting to know if
30:57 those would give very different things.
30:59 Yeah,
30:59 I totally agree with that.
31:01 So then,
31:01 so
31:02 really by assigning this individual task to the sellers,
31:04 the seller,
31:05 the tasks already serve as a reminder to say,
31:07 hey,
31:07 this is an important area to work on.
31:09 For example,
31:10 handling customer complaints.
31:11 Right?
31:11 In the ideal experiment,
31:12 we,
31:13 we,
31:13 we would really want to have a group
31:15 that just have the reminder but not the,
31:17 you know,
31:18 access to the tutorials.
31:20 We proposed that idea.
31:21 Um,
31:21 unfortunately,
31:22 that didn't,
31:22 the,
31:23 the platforms thing it's like a waste of energy.
31:25 And so eventually that didn't go through,
31:26 but
31:27 I totally agree that's something interesting to look at.
31:30 So,
31:31 uh,
31:31 but in the current structure,
31:32 it's just,
31:32 we can't really address that.
31:36 OK,
31:37 Patricia,
31:37 now you can,
31:38 I don't see any more questions.
31:39 You can,
31:41 uh,
31:42 move ahead.
31:42 Thanks.
31:43 OK,
31:43 so let me finish up with my discussion about data.
31:46 The data part,
31:46 um,
31:47 I,
31:47 I guess the main question,
31:49 uh,
31:49 sorry,
31:49 the main point I really want to make on the data is,
31:52 um,
31:52 is the limitation part.
31:54 So we're good,
31:54 we do have pretty good measure on the outcomes
31:56 on revenues and the number of visitors and.
32:00 Um,
32:00 things like that and also some observable quality metrics,
32:03 but what we don't have is unfortunately,
32:05 we don't know how much they're spending
32:07 on
32:07 advertising and promotions and,
32:09 uh,
32:10 of course,
32:10 their cost structure
32:11 and that poses some of the limitation and worry.
32:14 And so on the strategy side,
32:16 we know the type of products they're offering.
32:18 We,
32:18 we,
32:18 we were going to be investigating a lot more in the nature of the product.
32:22 And we know the pricing and we have some proxy
32:24 for their participation in marketing but not exactly on spending.
32:28 Right.
32:28 So,
32:29 um,
32:29 the characteristic side,
32:31 because as you recall since we assigned the,
32:33 um,
32:33 the training to the seller immediately after the register,
32:36 so,
32:37 uh,
32:37 the type of baseline information that we have hands,
32:40 we got a hands-on is
32:41 restricted to those collected during the registration process,
32:44 which is a little bit limited.
32:46 And so,
32:47 on the consumer side,
32:49 uh,
32:49 we're going to be constructing the all the observations
32:52 and the measures from the very detailed search,
32:54 browsing,
32:55 and purchasing record,
32:56 which I'm going to be a lot more precise when
32:58 I talk about the um results on the consumer part.
33:03 Now let's move on to the,
33:05 to the results sections,
33:06 right?
33:06 So the first thing we of course want to know is
33:09 what are the effect of this training on the cellar side.
33:11 And,
33:12 um,
33:13 and as I was mentioning before,
33:15 so we look at the seller who registered between May and mid-August,
33:18 uh,
33:18 in 2019.
33:19 This is when the program was first launched.
33:21 And during that period,
33:22 we assigned 25% of them to the,
33:24 to receive the treatment and we tracked these sellers
33:27 over the subsequent nine months after entry.
33:29 Um,
33:30 so in the end,
33:31 we have slightly over 700,000 new sellers.
33:35 And,
33:35 um,
33:36 and,
33:36 um,
33:36 so we construct a balanced panel for these news hours.
33:40 So,
33:41 OK.
33:41 So we first,
33:42 um,
33:43 the first thing we want to know is,
33:44 of course,
33:45 is the training useful,
33:46 and we estimate the ITT effect with the following,
33:49 um,
33:50 specification.
33:50 We're on the left-hand side,
33:51 we're interested in the outcomes or strategies and,
33:54 uh,
33:54 on the right-hand side,
33:56 the outcome of the variable of interest is,
33:58 of course,
33:58 whether they're assigned to the treatment or the
34:00 control group and then we also control for
34:02 the cohort
34:03 relative months and also initial sector fix.
34:08 So the first thing we track is we look at the take-up rate and uh so the take-up
34:13 um
34:13 defined by taking up any of the tests during that period
34:16 is about 25%
34:18 and actually none of the new sellers,
34:20 uh sorry,
34:20 none of the new cell from the control group take up the experiment,
34:23 so at least that part is pretty successful.
34:27 So now,
34:28 um,
34:28 looking at the main outcome measure,
34:30 so overall I think the training has a positive impact on sales's performance.
34:34 So the main outcome measure here we're looking at is really on the revenue side.
34:39 So on the extensive margin,
34:40 we see the training generate about 1% increase in the
34:44 likelihood for the seller to earn some positive revenues.
34:47 And um
34:48 because the distribution of the revenue measure is very much skilled,
34:52 so we use the log measure instead.
34:54 So here we found um the revenue um
34:58 Total revenue increased about 1.7%.
35:00 And if we look at the TOT estimate to
35:02 think about impact of actually participating in the training,
35:05 so that um the training participants earn about 6.6% higher revenue.
35:11 And the other thing that you,
35:12 we
35:12 also work um that we care about is,
35:15 is the retention rate,
35:16 and here the retention is really measured by
35:19 um whether you're making
35:20 any positive revenue during that particular month.
35:22 So as you can tell,
35:23 the retention rate,
35:24 um,
35:25 on average for the amount of control group is very low.
35:28 Um,
35:28 actually,
35:29 by the end of the 9th month,
35:31 just slightly over 10% of the sellers are still active in this market.
35:35 So looking at the treatment,
35:36 in fact,
35:36 we see actually
35:38 uh the training is most effective during
35:39 the initial months to keep these hours slightly
35:42 longer to
35:43 in the
35:44 2nd and 3 months and then the
35:46 results um disappeared gradually over time.
35:50 So
35:51 And that's the,
35:53 I think that's the temporal variations on the treatment in fact,
35:55 but as you recall,
35:57 most of the cells are most active to participate
35:59 in the training also in the earlier months.
36:02 And,
36:02 so,
36:03 OK,
36:03 so given this impacts on the revenue side,
36:06 um,
36:07 The next thing we want to do is we want to decompose that and then think about
36:11 um
36:13 Like how the revenue is earned.
36:15 So by decomposing revenue,
36:16 you can think about,
36:17 we can decompose this into
36:19 number of visitor tracks times
36:21 how many of these visitors eventually become buyers and
36:24 multiply that by the pricing,
36:25 right?
36:26 So that,
36:26 that gives you the revenues.
36:28 So
36:28 by looking at number of visitor attract and the conversion rate which
36:32 is essentially defined as share of visitors who eventually make a purchase,
36:36 most of the action and impact actually
36:39 are concentrated on attracting more visitors.
36:42 We don't see the significant outcomes on the conversion side.
36:47 And what we also don't see a significant difference is uh observed quality.
36:51 So
36:52 here we measure the quality in terms
36:54 of the customer rating these sellers are getting
36:56 and we also look at whether the
36:58 consumers are requesting refunds.
37:00 They're filing complaints through the valid rules and,
37:02 um,
37:03 and on these dimension by comparing the average sellers,
37:06 average
37:07 pretty sellers to other controlling that we
37:09 actually don't find a significant difference.
37:14 Right.
37:15 So,
37:16 Um,
37:17 so that's what we found on the outcome side,
37:19 but given these outcomes,
37:21 then of course,
37:21 we're interested in,
37:23 um,
37:23 the strategy.
37:24 So what does treated news are do differentially
37:27 that leads to these diff these changes.
37:30 So overall,
37:31 we find the changes in cellar strategy are pretty,
37:34 are quite consistent with the focus of the training,
37:36 with one exception.
37:38 And this expression,
37:39 uh this uh exception is really the fact that
37:42 even though a lot of the seller took up the tasks that are related to
37:46 setting up online stores and like to
37:49 the task that really to incentivize and encourage sellers
37:51 to post product and participate in this market,
37:54 there's nothing
37:55 on that
37:56 um particular dimension.
37:57 So
37:57 in this table,
37:58 what I'm presenting is um
38:01 It's the estimated treatment in effect converted to percentage
38:05 for different strategy and outcome measures.
38:08 So on the market participation side,
38:10 we don't find these treated news out to
38:12 be more active in posting product or paid deposit
38:15 or um also I'm not showing here is they're not posting more products
38:19 um on the intensive margin as well.
38:22 But they're doing one thing different.
38:23 So they are becoming a lot more active on marketing and promotions.
38:28 So by that,
38:29 I mean these sellers have more product,
38:31 participating in the paid advertising,
38:33 and they also have higher share of visitors coming from the paid channels.
38:37 And,
38:37 um,
38:38 and on the promotion side,
38:40 they're more likely to engage in limited-time sales event.
38:43 So all of these actions could benefiting sellers by
38:47 getting them a better ranking in the search outcomes.
38:49 So let me just give you a
38:51 What what's that,
38:51 what exactly that means.
38:52 So
38:53 when you search for keywords,
38:54 so I'm searching for coffee
38:55 here on the platform.
38:56 You have
38:57 code,
38:57 you have listing by
38:59 where this is a search ad.
39:00 You have a listing where this is a promotion event that seller participate.
39:04 It's like a limited time sales,
39:06 and you can also get a bunch of regular organic
39:09 search results.
39:10 So what the training actually do with
39:12 uh CL are participating in marketing promotion
39:15 is they're really changing their position and the rankings
39:18 um in this list.
39:19 So if you ranked higher,
39:20 you're just more likely to be found by the consumers.
39:23 So once you click on the page,
39:24 you can access a lot of details.
39:26 And one thing I do want to mention is
39:28 people engage in very complex set of pricing strategies.
39:31 So they offer discounts,
39:32 they offer coupons,
39:34 so you can share the links with your friends
39:35 who get additional coupon and they also give out,
39:37 give,
39:38 uh,
39:39 um,
39:41 they also give out gifts and samples.
39:43 So,
39:43 uh,
39:43 um,
39:44 the strategy space is very complex in this setting.
39:47 Um,
39:48 and right,
39:48 and that's also something that we want to have a better knowledge
39:52 about,
39:52 um,
39:53 down the road.
39:55 So that's on the marketing and promotion side.
39:58 And then the other side that we found a small change is on the service part.
40:04 So as you recall,
40:05 the last part of the training really teaches
40:07 the seller how to engage with their customers.
40:09 So on that front,
40:10 we found
40:11 these sellers,
40:12 the trading seller actually have value shorter response time when
40:15 the consumer make contacts with the customer service agent.
40:18 But that's,
40:19 but more importantly,
40:19 they have a higher conversion rate among the seller who ended,
40:22 sorry,
40:23 among the consumer who ended up contacting the customer service agent.
40:27 And,
40:27 um,
40:28 and so that's the other area of change.
40:30 And then the last thing we also want to check is really on the pricing front.
40:33 We,
40:34 we don't see a significant difference on the pricing about
40:37 um between the treated and control new seller.
40:40 For one thing is becau perhaps because pricing is
40:42 not the main focus of the training program,
40:45 but also because the pricing can be very complex and we should,
40:48 and we will be looking into more details on this end.
40:52 So now let me give you a quick summarize about what we found on the south side.
40:56 I think
40:57 overall we found Patricia,
41:00 yes,
41:00 uh,
41:00 let's pause for a bit and
41:02 allow people to ask questions.
41:04 Cairo,
41:04 you go first.
41:06 Thank you,
41:07 thank you,
41:07 Patricia.
41:08 A couple of clarification questions.
41:10 Um,
41:11 you know,
41:12 on the revenue,
41:13 I love revenue,
41:14 is this the revenue coming from online sales only,
41:18 or,
41:18 or is the total revenue of the firm?
41:20 Because I'm wondering,
41:21 for example,
41:21 if a firm is,
41:22 you know,
41:23 being more aggressive online.
41:25 Uh,
41:25 and expenses of selling less,
41:28 let's say
41:28 in a physical store or something like that.
41:32 OK.
41:32 So most of these firms,
41:33 as far as I know,
41:35 uh,
41:35 most of these firms don't have an offline presence.
41:37 So we implement a surveys,
41:38 a survey to,
41:40 and the survey is not that well represented,
41:42 but for the small,
41:43 very small salary,
41:44 I,
41:44 I wouldn't imagine like
41:46 less than 10% of these firms have an
41:48 offline presence.
41:49 So
41:50 most of them are just,
41:51 so every,
41:51 I think whatever we're covering here
41:54 is a pretty good representation about
41:56 their presence in the online market,
41:57 at least on this particular platform.
42:03 David,
42:04 you're next.
42:06 I just had a quick clarification on the pricing.
42:09 You said one of the sort of marketing channels they were
42:12 using was these temporary price discounts and things like that,
42:15 and so,
42:16 um,
42:17 and,
42:17 and all these special offers,
42:19 and so the what the prices could be quite complex,
42:22 I guess,
42:23 depending on whether you're paying,
42:25 you know,
42:25 with the discount and whether you're getting something free.
42:27 So,
42:28 when you're measuring price,
42:29 what are you actually measuring there,
42:30 and are you taking account of these marketing,
42:32 Um,
42:33 yeah,
42:33 we,
42:34 yeah,
42:34 we look at the actual price that the consumer paid out to the seller.
42:38 So of course the free sample part,
42:40 that part we can't control for,
42:41 but,
42:42 but that the price that we look at already
42:45 incorporate all of these discounts and coupons and
42:48 yeah.
42:50 So is your sense then that,
42:52 that uh
42:53 that those are just too small to show up in the price,
42:56 there's just enough other things going on in,
42:58 in price,
42:58 or that those discounts really aren't that
43:01 important for,
43:02 for most
43:03 um firms as part of attracting
43:05 more customers,
43:05 cos why is it not showing up in price if,
43:08 um,
43:09 you know,
43:09 they are using these discounts?
43:11 So that's a good question,
43:12 but I would say
43:14 the participate the average participation rate about this training,
43:18 about this promotion is very,
43:21 very,
43:21 very low.
43:22 So this is a dependent variable means like very,
43:24 very few.
43:26 Sellers participate in that,
43:28 um,
43:28 in that,
43:29 um,
43:29 in that marketing tool.
43:30 So,
43:31 um,
43:31 I,
43:31 so
43:32 I think part of it is really because,
43:34 um,
43:36 Of the effect could be small on the pricing side.
43:40 And even if they incorporate,
43:41 so because here we are looking at the average price that um
43:44 buyers are making,
43:46 but if the sellers,
43:47 if
43:48 um by offering discount and by offering different type of,
43:51 you know,
43:52 coupons and other promotion strategy,
43:54 you actually incentivize the consumer to spend more on the store
43:58 by,
43:58 rather you lower the price on one product,
44:00 but overall they spend more from this particular seller overall.
44:03 I think that,
44:04 um,
44:05 that is another thing that could be happening.
44:11 Uh,
44:12 uh,
44:13 I don't see any more hands,
44:15 so
44:16 you may continue,
44:17 Patricia.
44:18 OK.
44:19 So let me summarize on this part.
44:21 So on this part,
44:22 we,
44:23 um,
44:23 we see there's,
44:24 there is a small revenue increase on the seller side and that's really
44:27 coming from the fact that they invest more in marketing and they're more,
44:31 they do more promotions
44:33 and there's also small incremental.
44:34 Customer service quality.
44:35 But
44:36 you may already know is that a lot of these additional,
44:39 these changes in strategy could come at additional cost for these new sellers.
44:44 And since we don't have appropriate measure of their,
44:46 their spending and their cost,
44:47 and the impact on the profit is a little bit ambiguous.
44:50 And if we think about,
44:51 take it
44:53 a little bit cynical
44:54 view about marketing.
44:55 So what if everything is just zero-sum?
44:57 So if my success comes at the cost of the other person,
45:00 so what are the implications on the consumer side?
45:02 So,
45:03 to answer that,
45:03 this type of concern,
45:04 we really want to look at
45:06 um
45:06 consumers' experience.
45:09 Which brings us to the discussion about the next part,
45:12 the consumer side.
45:13 So,
45:14 uh,
45:15 before talking about the outcomes,
45:16 I really want to make it clear about how exactly the training
45:20 work and how exactly the training could change the consumer's experience.
45:25 So here I'm showing in a,
45:27 in a graphical way illustrate that,
45:29 you know,
45:30 the
45:30 the allocation of consumer's attention.
45:32 So
45:33 before the training,
45:34 we have
45:34 some of the new sellers treated or controlled
45:37 being found by the consumers when they search for the product on the platforms.
45:41 And,
45:42 uh,
45:43 empirically,
45:43 vast majority of these sellers ended up to be incumbents and
45:46 only a very small share of them are new sellers.
45:49 So uh
45:50 actually running into a new seller is a very rare event for the consumers.
45:53 So since the training allows these three-day news
45:56 to attract more visitors to the site,
45:59 That implies some of their consumers' attention,
46:03 especially if you think about changing the order of the search ranking,
46:06 really get reallocated
46:08 from the incumbents,
46:09 um,
46:09 because there are 90% of the seller on the market
46:11 to these traded new sellers.
46:13 And of course,
46:14 there are,
46:14 um,
46:15 I'm not drawing out the error,
46:16 but there could be some reallocation from this control new seller to the
46:19 treated new seller,
46:20 but
46:20 in the structured exercise,
46:22 we found actually most of the reallocation really
46:24 happened between the incumbents and the treated sellers.
46:28 So then the question we want to know is now there's
46:30 a change in the selection process about which sellers ended up
46:34 being,
46:35 ended up being found by the consumers.
46:37 So we want to know
46:39 what the consumers' experience when they interact
46:41 with this different group of sellers.
46:44 And so,
46:45 OK.
46:45 So to answer that question,
46:46 I'm going to
46:47 do it in 3 steps.
46:49 The first thing we want to know is when
46:51 the consumer visits new seller in the search section,
46:53 do they just more likely to make a purchase overall from
46:57 any of the sellers they ended up visiting.
47:00 And then the second question we want to make sure is
47:02 if they are making a little bit more purchase,
47:05 is it really drive by the new sellers or do they happen to,
47:08 you know,
47:09 interact with a different group of incumbents who are perhaps different,
47:12 um,
47:12 from the incumbents that don't,
47:15 don't show up at the same search session as
47:18 the other,
47:19 right,
47:19 as the other incumbents.
47:21 And so the last part we want to check is
47:24 um if the consumers are purchasing from the new sellers,
47:27 what's the implication on the quality of their purchase.
47:31 To answer that question,
47:32 I'm going to really explore the variations in
47:35 the composition of the sellers that the consumer visit
47:37 when they search for particular keywords and uh spend certain
47:41 effort on search.
47:42 So let me give you a visual um demonstration about the type of data we have access to.
47:48 So we have access to the search and browsing data,
47:52 and the data
47:53 um
47:55 is reorganized in the sense that we have a lot of different consumers.
47:59 They could be searching for a very diverse set of keywords like cellphone cases,
48:04 um,
48:04 sweaters,
48:05 and so on and so forth.
48:06 So when they search for these keywords,
48:08 as I was showing in the previous graph,
48:10 they're going to get a list of sellers and that's all coming in,
48:14 of course,
48:14 in a particular ranking.
48:15 And so given the rankings,
48:17 um,
48:18 these consumers are going to be clicked on a specific set of sellers.
48:23 And so
48:24 we don't know um the precise
48:28 data,
48:28 the precise list of sellers that the consumer got exposed to,
48:31 but what we do know is which are the sellers
48:34 these consumer ended up clicking on and go to the
48:37 product detail page.
48:39 I think,
48:39 so this is the
48:40 data that,
48:41 this is the browsing data that we're going to be
48:44 considering here.
48:45 And so for the vast majority of these cases,
48:48 these consumers are going to be interacting with the incumbents
48:52 and occasionally some consumers are going to run into
48:54 a control new sellers who are treating new sellers.
48:57 And if they find a match,
48:59 they're going to make a purchase from one of these sellers.
49:03 And so
49:04 given this data structure,
49:06 we have two types of variations.
49:07 So we have
49:08 same consumers,
49:09 they can search for very different type of keywords
49:12 and we also have the same keywords where they,
49:14 these keywords are being searched by different consumers.
49:17 So that allows us to address some of the consumer-specific
49:21 idiosyncrasy and also
49:24 um seller like uh keywords or where you can think about as product-specific
49:28 um differences.
49:30 Um,
49:31 I'm going to refer to the set of the sellers that the consumer ended up
49:35 visiting as the consideration that that these consumer face,
49:39 um,
49:39 when they search for the product.
49:41 And so,
49:43 we also,
49:44 we're going to be referring to consumer search
49:46 effort as how many sellers they ended up visiting
49:49 in that particular search section.
49:51 It's actually very important to consider the search effort because
49:55 spending extra effort could have
49:57 different implications on the probability of purchase,
50:00 right?
50:00 Because
50:01 if the consumers decide to spend more effort on search,
50:04 that could means they really want to Look for a match and so they really want to,
50:08 so that could,
50:08 that increase their probability of making a purchase overall.
50:11 But that could,
50:12 but
50:12 having to go through more sellers
50:15 could also means they are finding bad draws in the previous search.
50:19 So,
50:19 um,
50:20 in the end,
50:20 we,
50:21 we think it's very
50:22 important to consider
50:24 to add the search effort into part of the fixed effect.
50:29 So
50:30 this is the basic structure of the data on the consumer
50:33 side that we have access to and then I'm now,
50:36 now going to
50:37 um
50:38 show you how we answer that question by constructing
50:40 relevant data set from this particular data set.
50:43 And so,
50:44 actually,
50:44 before going there,
50:45 the one thing I do want to mention is,
50:47 of course,
50:48 the seller,
50:48 the platforms
50:50 is doing a lot of customization in the search and uh recommendation process.
50:54 So
50:54 each consumer,
50:55 even when they search for the exact same products,
50:57 they're going to get a different list of the sellers.
50:59 So,
51:00 um,
51:01 so as a result,
51:02 we really don't have the ideal experiment where
51:04 we randomly match the consumer with the new sellers.
51:07 And,
51:07 uh,
51:07 empirically,
51:08 we do find the consumer who visit these new sellers
51:12 just tend to be more experienced.
51:14 They spend more,
51:15 they search more,
51:15 so
51:16 they are considered,
51:17 I would imagine as a safer type
51:19 from the perspective of the platform.
51:21 So,
51:22 um,
51:23 so to address that type of selection challenge,
51:26 we will
51:27 use different types of relevant fixed effect whenever um whenever makes sense.
51:34 So if there's no
51:36 is there any que uh yeah,
51:37 please let me know if there are any
51:38 clarification question about the data structure and,
51:41 and so on and so forth.
51:43 But um
51:44 if there's no question at this moment,
51:45 yeah,
51:46 let me tell you how we answer these three questions,
51:48 um,
51:49 step by step.
51:51 So for the first question looking at the market expansion part,
51:54 the data we uh we construct is the following.
51:56 So we find a particular date,
51:59 search keyword,
52:00 and effort combination.
52:01 And with that particular combination,
52:03 we identify the consumer who only visit
52:06 the incumbents and those who visit at least one new seller.
52:10 And so we collapse each of these observation to a consideration set
52:14 and then we look at the
52:16 outcome which is
52:17 to the consumer make a purchase
52:19 from any of these cells in the set and
52:21 we explore the variations in the particular consideration set.
52:25 And here,
52:26 uh,
52:27 we run the following specification where on the left-hand side,
52:30 the outcome of interest is whether a purchase happened,
52:34 um,
52:34 from that side.
52:35 And then the,
52:37 on the right-hand side,
52:37 we're interested in whether
52:39 any treated new sellers will control new sellers show up in that consideration set.
52:44 And we also include a very rich set of controls to control for
52:48 consumers' characteristics.
52:49 For example,
52:50 the recent spending or their search intensity,
52:52 and we also control for,
52:54 you know,
52:54 some,
52:55 some of the characteristic of the seller who showed up in that task,
52:58 especially the,
52:59 the pricing level.
53:00 And relaxing that kind of control on the seller
53:02 side actually doesn't change the result that much.
53:04 But
53:05 Importantly to really address the selection concern,
53:08 we add the consumer fixed effect
53:10 and we also add the
53:11 fixed effects that captures the variations by keywords
53:15 and search effort and also
53:16 time.
53:17 So,
53:17 and then everything left is really the
53:19 consumer and product-specific variation which is,
53:23 um,
53:23 which is
53:25 And so with that particular type of variations,
53:27 um,
53:27 it's hard to predict whether
53:29 interacting with new sellers are going to
53:31 downward,
53:32 upward or like you know,
53:34 downward bias,
53:35 um,
53:36 the results.
53:38 So
53:39 Now,
53:40 let me tell you what we found.
53:41 So,
53:43 Here is a table um summarizing
53:45 the,
53:46 um,
53:46 the results from the previous spec specification.
53:49 So on the first column,
53:50 we look at
53:51 whether purchase
53:52 uh from the consideration that happened on the same day.
53:55 And then here we extend the time horizon a little bit and to,
53:58 to check if any purchase happened in the same week.
54:01 And we also look at the total spending and we measure the total spending in,
54:05 in log scale.
54:07 So,
54:08 um,
54:08 looking at the frozen circus,
54:11 um,
54:11 and comparing the to the baseline case
54:14 which is only interacting with the incumbents,
54:17 we found actually interacting with new sellers either treated or controlled
54:22 significantly increased the purchase probability.
54:25 Here,
54:25 the magnate of increase is about 2%.
54:29 So that in uh the magnitude,
54:31 that magnitude should be small,
54:32 the percentage term,
54:33 but I think
54:34 once they have perhaps forgot to mention we'll talk about the platform
54:38 in
54:38 in 2019,
54:39 the platform facilitate over
54:43 $100 billion US dollars
54:45 of transactions.
54:46 So
54:47 even a small increase in the efficiency
54:50 can be translating into something very large,
54:53 uh,
54:53 in absolute magnitude.
54:56 So,
54:57 on the generating purchase part,
54:59 interacting with new sellers,
55:01 um,
55:02 Means the purchase is more likely to happen compared to the case with incumbents.
55:06 And then we also want to know if there's a difference between interacting
55:10 with the pre-date and controlling so so we compare the coefficient on these
55:13 um two variables here
55:15 and um
55:16 And we're looking at the same-day purchase,
55:18 we don't find a significant difference.
55:20 We
55:21 actually looking at the purchase property overall,
55:23 we don't find a significant difference between
55:25 the pre-date and control new seller.
55:27 And then here,
55:28 um,
55:28 we're looking at the total spending,
55:29 the
55:30 pre-date new sellers are doing slightly worse than the control new sellers,
55:33 but still they are doing better than the incumbents
55:35 which really mitigate some of the
55:38 earlier concern about potential negative selections
55:41 due to the training.
55:42 So
55:42 even the training really expand the
55:46 News out to treated news out visibility
55:48 uh among the consumers that didn't,
55:51 that didn't uh really negative hurt the consumer's experience on this platform.
55:57 That's the first part.
55:58 And then on the second part,
56:00 um,
56:00 we look at the allocations,
56:02 and here the data we use is to,
56:05 um,
56:05 is to,
56:06 is restricted to this kind of cases.
56:08 We're in the Patricia,
56:10 yeah,
56:11 so we're getting to the 1 hour marks.
56:12 This would be a good time to pause and
56:15 allow people to ask question.
56:16 I see Toan,
56:17 you have your hand raised.
56:20 Go,
56:20 you go first.
56:21 Yeah,
56:22 so,
56:22 uh,
56:22 uh,
56:23 um,
56:24 I'd like to understand the empirical strategy.
56:26 So here,
56:27 the identification assumption is
56:31 Whatever the good,
56:32 a given consumer will spend
56:35 as much effort.
56:36 So conditional ongoing on the platform and
56:39 so I,
56:39 you,
56:39 you're ruling out the case where I really want to buy a sweater.
56:43 And I'm just browsing for a case for iPhone X because
56:47 I,
56:47 I like to spend time.
56:48 So,
56:49 so your,
56:50 your
56:51 consideration set is,
56:52 is,
56:53 is consumer-specific.
56:54 It cannot be consumer times good specific.
56:57 Otherwise,
56:58 you don't identify your coefficients.
57:00 Am I,
57:00 am I correct?
57:01 Do I,
57:01 to understand the,
57:02 the empirical strategy that way?
57:04 Yeah,
57:04 so we are
57:06 coding constant search keywords,
57:08 right?
57:08 So everything is comparing conditionally you are searching this
57:12 exact consumer are searching for the exact keywords.
57:15 What are the difference when you ended up
57:17 interacting with a different composition of the cell,
57:18 I think that's uh
57:19 where the identification come from.
57:21 So we try to
57:22 control for the idiosyncrasy
57:24 between product category and among the consumers.
57:27 Yeah,
57:27 but you don't allow that idiosyncrasy to interact
57:31 between consumer goods.
57:31 We,
57:31 we have a look at uh like heterogenetic
57:34 result.
57:42 OK,
57:43 I don't see
57:43 any more hands,
57:44 so you can continue.
57:47 OK,
57:47 I thought there was some discussion in the chat function,
57:49 but I,
57:50 if there are any questions,
57:51 just please ask me directly.
57:53 I think it's easier,
57:54 um.
57:55 But,
57:55 OK,
57:56 so let me go back to this part of the analysis where the,
57:59 I think hopefully the uh the first part is clear and then,
58:02 and then on the second part,
58:03 we look at consumers' choices
58:05 within the consideration set,
58:07 um,
58:07 when,
58:08 especially when both incumbents and new sellers appear in
58:11 the same set and the purchase ended up happening.
58:13 So the question is,
58:15 who the consumer ended up choosing from this set.
58:18 And so here's the
58:20 We can really look at the um
58:23 We can really think about the choice within the set,
58:26 so we do include the conservation side fixed,
58:28 fixed effect.
58:29 Um,
58:30 and this is what we found.
58:32 So we,
58:33 we found the consumers
58:35 when facing incumbents and the new seller at the same time,
58:38 they are significantly more likely to choose
58:41 these treated new sellers
58:42 over the incumbents.
58:43 Whether we measure that by,
58:45 you know,
58:45 by the,
58:46 by the purchase probability and the total spending,
58:48 especially if we look at the,
58:50 you know,
58:50 the,
58:50 the impact on the,
58:51 on the same thing.
58:53 And so also comparing the treated new seller to the control new seller seems to be
58:58 the case that here the treated new sellers
58:59 are doing better than the control new sellers.
59:01 And here the training actually has two functions.
59:03 The one thing is the training changes selection of which seller ended up showing up
59:08 in this consideration set.
59:10 But also,
59:11 as you recall,
59:11 since the training treated new sellers are
59:14 having slightly better customer service quality,
59:16 especially their conversion rate when the consumer
59:18 make contact with the customer service agent.
59:20 So there could be some premium back here
59:22 and um we can't really distinguish between selection and treatments,
59:26 in fact,
59:26 precisely.
59:29 And so for the last part of analysis,
59:31 we,
59:31 what we also care about is we want to know what are the quality difference
59:35 when consumer choose different type of sellers.
59:38 So
59:38 here we look at the case when the purchase
59:41 actually happened and we look at the post-purchase experience
59:44 um when they are made from different sellers.
59:47 And,
59:47 um,
59:48 ideally,
59:48 we definitely want to include the consideration set fixed effect
59:52 uh because um
59:53 most seller only pick one,
59:55 sorry,
59:56 most consumer only pick one seller from a particular set,
59:59 we instead switch back to the previous specification and use consumer fixed effect
1:00:02 and we also control for like search keyword effort and date fixed effect.
1:00:05 And we also include the size of order which is an important determinant,
1:00:09 um,
1:00:09 about,
1:00:10 uh,
1:00:11 uh,
1:00:11 about the subsequent cases.
1:00:13 So what we found,
1:00:15 so we found
1:00:16 this um
1:00:18 In a nutshell,
1:00:18 there's no adverse effect on the quality of the purchase from the new sellers.
1:00:22 So looking at the probability of consumer requesting refund and or returns
1:00:26 which
1:00:27 um
1:00:28 Ideally,
1:00:29 we want these measures to be as low as possible
1:00:31 and we also look at the possibility of them making repeat
1:00:33 purchase in the subsequent months and here we do want to
1:00:36 have,
1:00:37 um,
1:00:37 have a,
1:00:38 have a high number here because that imply retention,
1:00:41 right?
1:00:41 So along all of these metrics,
1:00:43 we don't find a significant difference between the treated
1:00:46 and treated news hours in particular with the incumbents.
1:00:49 And the control news hours seem to be doing
1:00:51 slightly better on the requesting the rebounded side,
1:00:55 um,
1:00:55 either both compared to the incumbents and to the
1:00:58 And so the treating cell,
1:01:00 so
1:01:00 there could be some evidence of negative selection in this context,
1:01:03 but overall if the more meaningful
1:01:06 reference group here is encompassed,
1:01:08 I'm less of a con I'm less concerning about this
1:01:11 type of negative selection between the treated and controlled.
1:01:16 So now let me just quickly summarize what we found on the consumer side.
1:01:20 So we,
1:01:20 we think
1:01:21 uh from the reduced form analysis,
1:01:23 we think the consumer actually benefit from interacting with these new sellers
1:01:26 and uh since they're more likely to make a purchase that generate market expansion
1:01:30 and um because they're purchasing from the new seller for
1:01:33 the incumbents that implies the market got reallocated from these
1:01:36 uh incumbents to new sellers
1:01:38 but um we don't find a significant impact on the quality of the purchase.
1:01:43 And so just to link that back to the training,
1:01:45 so the training does two things.
1:01:47 Um,
1:01:47 the training improves customer service quality to a small extent,
1:01:50 but what the training really does is,
1:01:53 um,
1:01:53 it,
1:01:53 it really increases the visibility of these new sellers.
1:01:57 And so these things have become much more likely to be found
1:02:00 because they're attracting more business.
1:02:02 So then the question um on the welfare part is to
1:02:05 understand what are the implication of this training on the consumers,
1:02:08 sellers,
1:02:08 and incumbents.
1:02:10 So,
1:02:11 Do we need to pause that at this moment,
1:02:13 or should I,
1:02:13 um,
1:02:14 where I can directly go into the discussion about
1:02:16 how we can actually quantify and characterize the welfare?
1:02:19 Yeah,
1:02:20 and let's pause for a bit and see if there are any questions.huan,
1:02:22 I see your hand,
1:02:23 that's from your previous question or you,
1:02:26 you have a question?
1:02:27 Oh,
1:02:27 OK.
1:02:30 OK.
1:02:30 Uh,
1:02:31 I don't see any harm,
1:02:31 so you can continue,
1:02:32 Patricia.
1:02:33 OK,
1:02:34 so now let me tell you how we set up the structural part of the analysis.
1:02:38 So,
1:02:38 um,
1:02:39 we set on a model in a very,
1:02:41 you know,
1:02:42 uh,
1:02:42 pretty straightforward and simple way because we really want
1:02:44 to focus on the main friction in this market which
1:02:47 essentially is the third friction.
1:02:48 So,
1:02:49 so the model has two parts.
1:02:50 On the first part,
1:02:51 we characterize the consumer demand and um the goal of that is to
1:02:54 recover the seller-specific analy quality.
1:02:57 So this is how the model looks like.
1:02:59 So we think about the consumers when they enter the market,
1:03:02 they're going to be with search,
1:03:03 they search for particular keywords,
1:03:05 then they visit a particular set of the sellers
1:03:07 which right now we hold as something constant and we
1:03:10 don't exactly model the data generation process of who ended up
1:03:14 into these consideration sets,
1:03:15 um,
1:03:15 for this part of analysis.
1:03:17 So,
1:03:18 so the only thing the consumer need to decide is
1:03:20 which of the seller gave her the highest utility,
1:03:22 and the utility
1:03:23 is going to be depend on seller's pricing and some of the strategy like
1:03:28 number of products they're offering or their customer rating.
1:03:30 But very importantly,
1:03:31 we're going to think about this,
1:03:32 um,
1:03:34 This CJ which captures the post-treatment quality and this is the outcome,
1:03:38 this is the parameter that we're mostly interested in to recover and,
1:03:42 um,
1:03:43 from this particular setup,
1:03:44 this is really the demand residual that
1:03:46 um
1:03:47 um in this specification.
1:03:49 So
1:03:50 If we make the standard assumption and assume a large error for the,
1:03:54 for the,
1:03:55 for the,
1:03:55 for the error term and we also normalize the
1:03:57 outside option to opt not to purchase to zero,
1:03:59 we got a,
1:04:01 a familiar setup
1:04:02 where uh we can simply characterize the consumer's likelihood to make a purchase
1:04:07 from a particular seller J in her set in the following way.
1:04:11 And so one thing you may worry about is really what if the seller's strategy are
1:04:17 going to be correlated with this estimated with uh no,
1:04:20 sorry,
1:04:20 with the with the CCJ.
1:04:22 So
1:04:22 to address that,
1:04:23 um,
1:04:24 selection problem,
1:04:25 so it's an endogenous induction problem,
1:04:28 uh,
1:04:29 we use a set of instrument to
1:04:31 instrument for the uh uh for the,
1:04:33 for the strategies that our
1:04:35 and the instrument is actually pretty unique to this market.
1:04:39 So
1:04:39 The Platforms have implemented a lot of rules
1:04:43 and um they try to regulate the market by
1:04:46 uh punishing the behavior such as charging very excessive price or
1:04:50 char or posting a lot of product just to get exposure
1:04:54 uh in the search rankings,
1:04:56 so
1:04:57 And then there's variations in terms of how stringent
1:05:00 the platforms are implementing these type of rules.
1:05:03 So the instrument we use is we,
1:05:05 um,
1:05:06 we take,
1:05:06 we,
1:05:07 we explore the temporal and also
1:05:09 product-level variations in the platform's rule enforcement
1:05:12 and we also use the training assignment.
1:05:14 So
1:05:15 these instruments are going to,
1:05:16 so we use these instruments in the first stage to
1:05:19 um to estimate,
1:05:20 um,
1:05:21 To,
1:05:22 to estimate seller strategy on pricing and product and we jointly estimate
1:05:26 the first stage with the
1:05:28 previous specification
1:05:29 looking at the consumer demand side.
1:05:32 And because we have this very nice logic formula,
1:05:35 we can
1:05:35 actually write down the consumer welfare and the sales revenue in a,
1:05:39 in a close one way.
1:05:40 And we also,
1:05:41 so,
1:05:41 and then we estimate
1:05:43 um this part of the model with the simulated maximum,
1:05:46 maximum likelihood
1:05:47 on the consumer and seller search session sample
1:05:50 that um that we talked about
1:05:52 in the reduced form analysis.
1:05:55 That's the first part of the model.
1:05:57 So to close the model,
1:05:58 we um also characterize a matching between the consumer and seller,
1:06:02 and here
1:06:02 we can take a very,
1:06:03 um,
1:06:04 a flexible approach to think about,
1:06:07 uh,
1:06:07 how many visitors that the that the seller can attract in the
1:06:10 current period and we assume that it's going to be dependent on
1:06:13 number of visitors that the consumer,
1:06:14 that the seller track in the previous period and also their conversion rate.
1:06:18 So then training affect this process by changing the lack of visitors
1:06:22 and then the quality is going to change the conversion rate.
1:06:26 So with the model,
1:06:27 then the main channel that we really want to focus
1:06:30 on is the search frictions and allocation of consumers' attention.
1:06:34 Basically,
1:06:35 um,
1:06:35 what we found
1:06:36 is that the
1:06:38 new sellers have higher underlying quality,
1:06:41 higher estimated CCJ
1:06:43 than the
1:06:44 um than the incumbent.
1:06:46 So here we look at a separate example for
1:06:49 Seller with a purchase record and those with also among these two examples,
1:06:53 the treated and controlled new seller have higher procedures and incumbents.
1:06:57 So because these higher quality news sellers are
1:06:59 not getting enough attention from the consumer,
1:07:02 so that causes a problem that generates a search.
1:07:05 And since the training really
1:07:07 um make these training participants more likely to be found
1:07:11 and more likely to appear in the conservation set,
1:07:13 the training reduces their function.
1:07:15 So to think about what happened if the,
1:07:17 if the training is not being made available,
1:07:20 we
1:07:20 just limit training participant access uh appearance in the consideration set.
1:07:25 And so this is what we found on the reb um.
1:07:28 Uh,
1:07:28 in the two cases,
1:07:29 but compared to the baseline case when the training is available.
1:07:33 So without the training,
1:07:34 we'll see there's a small decrease about less than
1:07:36 1% 0.1% decrease in the consumer welfare and it's a small extent
1:07:40 also a drop in the total sales revenue.
1:07:42 So
1:07:43 again,
1:07:43 um,
1:07:44 I think in percentage terms of changing the welfare is now quite substantial,
1:07:48 but
1:07:48 think about,
1:07:49 but there,
1:07:49 um,
1:07:50 think about the size of the platform and also the news are only account for
1:07:53 a rather,
1:07:54 a very small section of the market.
1:07:56 I think this,
1:07:57 um,
1:07:57 number is
1:07:58 already quite meaningful using it and it
1:08:01 definitely generate that definitely justify the
1:08:03 cost of implementing the program.
1:08:07 And
1:08:07 so,
1:08:07 so
1:08:08 Patricia,
1:08:09 so you,
1:08:10 you have less than 5 minutes.
1:08:12 So we just focus on more important parts.
1:08:14 All right,
1:08:15 thanks.
1:08:15 Yeah,
1:08:16 I'm almost there.
1:08:16 So,
1:08:17 OK,
1:08:17 so the next thing we'll also look at is we're worried about business building.
1:08:21 So the question is who's still.
1:08:23 who the business still,
1:08:24 uh,
1:08:24 the new sellers steal the business from.
1:08:27 And looking at the changes in market share the
1:08:29 treat the treating new seller have a significant drop
1:08:32 and that really come from the incumbents.
1:08:34 We actually don't see a significant change on the market share
1:08:37 of the controlling sellers.
1:08:40 And um
1:08:41 OK.
1:08:42 Let me skip the discussion on that part,
1:08:44 um,
1:08:45 by going to the conclusion.
1:08:47 So
1:08:47 I think um from the previous analysis that I discussed,
1:08:51 I think we have
1:08:52 3 major pieces of conclusion.
1:08:54 So
1:08:54 the first thing we want to show is that new sellers
1:08:57 in this e-commerce market certainly faced quite substantial growth barriers.
1:09:02 And so that implies
1:09:03 the type of onboarding program that many of the
1:09:06 government agencies are currently implementing might not be.
1:09:09 So,
1:09:10 besides onboarding,
1:09:11 what they
1:09:12 really should be considering is to think about the ways,
1:09:15 for example,
1:09:16 the business training program to really
1:09:17 address these search and information frictions.
1:09:20 And because the training has a potential
1:09:22 to reduce the search frictions on the platforms
1:09:25 that ended up improving consumers' experience
1:09:29 and because the consumer now got better matching outcomes as
1:09:32 they're more likely to interact with these higher-quality news hour.
1:09:35 So,
1:09:36 um,
1:09:36 so that,
1:09:37 and that eventually foster market expansion.
1:09:40 given the findings on the sellers on the consumer side,
1:09:44 what I imply for the platform is that the platform could actually
1:09:48 be more active to
1:09:51 help the
1:09:52 participants they're hosting and to really lift the growth barriers because
1:09:55 improving,
1:09:56 improving experience of the condoman sellers are going to
1:09:59 be consistent with the platform long-term profit maximization agenda.
1:10:02 And thinking about it from the policy perspective,
1:10:05 when we think about implementing um
1:10:08 different interventions to support these firms,
1:10:10 especially
1:10:11 in,
1:10:12 as we enter the um the digital economy territory,
1:10:16 it's quite,
1:10:17 um,
1:10:17 it could make sense to think about
1:10:19 fostering
1:10:21 more closely connected
1:10:22 relationship between the private and um
1:10:25 public
1:10:26 uh pub um public.
1:10:28 Private and public sector
1:10:29 precisely because the platforms occupy this very unique position to
1:10:34 reach,
1:10:34 implement,
1:10:35 and disseminate information and program at
1:10:38 low cost and very large scale.
1:10:41 And that's actually the end of my talk,
1:10:44 and one minute early,
1:10:45 so
1:10:46 is there any questions to ask,
1:10:47 please?
1:10:48 OK,
1:10:48 so I
1:10:50 did see a hand from Alvaro earlier.
1:10:52 Alvaro,
1:10:53 do you want to ask your question?
1:10:55 Um,
1:10:56 I think Patricia basically address it at the end because,
1:10:58 I mean,
1:10:58 my sense is you're saying,
1:11:00 what you're saying is that the welfare gains
1:11:02 come from the lowering of the search costs,
1:11:04 but
1:11:05 from the point of view of,
1:11:06 of the platform is,
1:11:07 it's
1:11:08 there,
1:11:08 it's
1:11:10 from the profit maximization to make sure that there are more matches.
1:11:12 So,
1:11:12 I mean,
1:11:13 why we,
1:11:14 do we want to,
1:11:14 you know,
1:11:15 figure,
1:11:15 worry about the training per se,
1:11:17 which at the end,
1:11:17 it seems that it's generally more marketing,
1:11:19 but the platform itself should be,
1:11:21 you know,
1:11:23 Reducing these barriers at the end because you want more,
1:11:26 more and better matches so there's more transactions.
1:11:29 I think you already mentioned that.
1:11:31 Right.
1:11:32 Maybe I'm not sure this is exactly addressing the problem,
1:11:34 but I think the,
1:11:35 um,
1:11:36 I,
1:11:36 I think my take is this.
1:11:38 So on the short run,
1:11:39 the platform is definitely benefiting from marketing,
1:11:41 and then the platform is already spending a lot of effort to,
1:11:44 you know,
1:11:44 improve the social and uh social ranking algorithm to,
1:11:47 you know,
1:11:47 improve the efficiency.
1:11:48 But there's a limit.
1:11:49 So the platforms actually
1:11:51 know less than you expect about these sellers.
1:11:54 So they won't,
1:11:55 since the,
1:11:56 since the platform will never have perfect information about the sellers,
1:11:59 it's very hard to,
1:12:00 for them to,
1:12:01 to figure out who are the high-quality seller
1:12:03 unless through some sort of review preference.
1:12:06 So by encouraging these new sellers to invest in
1:12:08 marketing and to generate more experiment on their end,
1:12:12 that also benefits the platforms so they can use that extra
1:12:16 information in the er uh in the algorithm as well.
1:12:19 OK,
1:12:20 yeah,
1:12:20 thanks,
1:12:21 Patricia.
1:12:21 Thanks,
1:12:22 everyone.
1:12:22 So it is time,
1:12:23 but I see there is one more question from Kyo.
1:12:26 Kyu,
1:12:26 you can stick around and speak to Patricia.
1:12:29 Uh
1:12:30 So,
1:12:31 thanks everyone.
1:12:36 Thank you.
1:12:36 Bye.
1:12:38 Thank you,
1:12:38 Patricia.
1:12:39 Thank you for coming.
1:12:41 Yeah,
1:12:41 like,
1:12:42 please let me know if there are any questions.
1:12:44 If there's anybody else with questions,
1:12:46 you can stick around and
1:12:48 speak with Patricia.
1:12:50 OK.
1:12:50 Thanks.
1:12:51 Yeah,
1:12:51 I hear there's a discussion about search,
1:12:54 uh,
Accelerating Sustainable and Clean Energy Access Transformation (ASCENT) in Mozambique
Jenny Chao, Senior Energy Specialist, World Bank
Mozambique has made remarkable strides in expanding electricity access—and development partners have played a central role in making that progress possible. Coordinated investments are helping drive jobs, economic activity, and growth across the country. The program is part of the broader ambition of Mission 300: the joint initiative to bring electricity to 300 million people in Sub-Saharan Africa by 2030. That goal is only achievable through partnership—with multiple institutions aligning behind a shared target and co-financiers stepping in to accelerate momentum.
00:05 Jacques Bartoni Chercole bonjour,
00:06 good afternoon
00:08 everyone.
00:09 So I'm Shaima Beelwali,
00:10 senior urban specialist at the World Bank in Morocco.
00:13 I am Zuita Benani,
00:15 local governments and financial institutions
00:17 specialist at the French Development Agency
00:20 in Morocco,
00:22 and we're going to tell you
00:24 about our partnership with the government of Morocco
00:27 under
00:28 the municipal Performance Program.
00:32 So first,
00:32 let us give you some quick background elements.
00:36 Over the past two decades,
00:37 Morocco has achieved significant socio-economic progress.
00:41 Cities have in this context been really key engines of this growth trajectory.
00:47 In parallel,
00:48 the country has also embarked on a transformational decentralization agenda
00:52 where local governments now play a critical role
00:56 in contributing to the country's development vision.
00:59 So in this context,
01:00 municipalities
01:02 now play a key role and are responsible for
01:05 delivering inclusive and quality access to basic services,
01:09 and that's the good news.
01:11 But municipalities are still facing capacity
01:15 challenges to fulfill their mandate.
01:18 Let me give you some quick figures to illustrate that.
01:22 It has been estimated
01:24 that
01:25 $33 billion in investments are needed
01:28 for urban infrastructures and services
01:31 in Moroccan cities between 2017 and 2027.
01:37 Around 70% of those costs are expected to be covered
01:41 by municipalities.
01:44 However,
01:45 they have only been able to invest
01:47 about 20%
01:49 out of the expected amount yearly,
01:52 and that's where we came in.
01:54 In 2020,
01:56 the government of Morocco
01:58 through the Directorate of
02:00 Local Governments,
02:01 DIC
02:03 partnered with the World Bank and the French Development Agency.
02:07 Together we rolled out an innovative program for results.
02:13 It targets approximately 100 urban municipalities
02:17 that represents about 50% of the total population.
02:22 The aim was
02:24 to improve the quality of services,
02:27 and we're doing it by enhancing
02:29 municipal,
02:31 institutional,
02:31 and financial capacities.
02:35 So basically,
02:36 first,
02:36 the journey began with developing a common
02:39 understanding of what good municipal management means.
02:42 We helped develop a framework that
02:44 really captured key ingredients of municipal performance
02:48 in that we had about 20 indicators that ranged from
02:52 how well municipal budgets are executed
02:54 to
02:55 how critical services such as solid waste are managed.
02:59 Really this allowed the government as well as development
03:02 partner to develop a standardized approach where we could assess
03:07 where municipal capacity lies
03:09 and the remaining gaps to bridge.
03:11 This is really for us one of the main value added of the World Bank and IFDS
03:17 partnership under the program.
03:19 Right.
03:21 Next,
03:21 we operationalized the framework
03:24 and conducted a series of annual performance assessments.
03:29 This allowed us to measure progress in municipal capacity.
03:34 We also carried out
03:35 complementary training
03:37 and technical assistance in areas where weaknesses were identified.
03:43 This included,
03:44 for example,
03:45 courses in local finance management,
03:48 as well as
03:50 good
03:51 measures in procurements.
03:54 So basically our partnership helped mobilize about $550 million
03:59 to improve municipal investment capacity.
04:03 Out of that 400 million in co-financing between the World Bank and IFD.
04:08 Basically,
04:09 it was the first time that Morocco introduced a performance-based subsidy schemes
04:15 for municipalities.
04:16 It is actually
04:17 simple and effective.
04:19 The better you perform,
04:21 the more money you get
04:22 to improve actually your capacity in
04:24 terms of investments in urban infrastructure.
04:27 So up until now,
04:29 an envelope of about $350 million was transferred
04:33 to about 100 municipalities and as Zubeida said,
04:36 covering.
04:37 About 50% of the population,
04:41 um,
04:41 and these investments that that were financed,
04:43 so included urban improvements,
04:46 namely in roads,
04:47 in public street lighting,
04:49 in green spaces,
04:50 but also
04:51 on the jobs creation perspective,
04:53 development of
04:54 economic and and and commercial areas,
04:57 as well as from the social side,
04:59 so social and sports facilities.
05:02 Another aspect of the program
05:04 is to improve management practices
05:07 in terms of transparency.
05:10 93% of targeted municipalities
05:13 now publish their financial statements,
05:16 as you can see here for Casablanca.
05:20 That compares to less than 12% at the beginning of the program.
05:26 86% of targeted municipalities have now established
05:31 a well functioning grievance redress mechanism,
05:34 and 67%
05:36 have set up
05:38 citizen satisfaction systems.
05:41 And finally,
05:42 beyond the IFD and World Bank partnership,
05:45 the program also helped municipalities partner together for greater impact.
05:50 It actually created a framework where municipalities would join forces
05:55 to deliver more efficiently
05:57 mutualized services such as urban transport,
05:59 hygiene,
06:00 or solid waste management.
06:02 It would actually then be for municipalities
06:05 and opportunities to come together to work
06:08 on those selected services more efficiently.
06:12 We wanted to highlight one final advantage of this partnership before we leave you.
06:18 Working together
06:19 on the program
06:21 allowed us for more alignment and greater impact.
06:25 The World Bank and AFD's standalone project
06:29 further contribute to the objectives under
06:32 this program.
06:34 So
06:35 for instance on the World Bank side,
06:37 activities that actually were
06:39 directly piloted with the city of Casablanca
06:42 have been further scaled up under the municipal
06:44 Performance Program to be about 100 municipalities.
06:48 These namely included,
06:49 for instance,
06:50 greater support to improve revenue mobilization.
06:54 On the AFD side,
06:56 the program complements our long-standing
06:59 partnership with the Fond
07:01 Equipment Communale FEC,
07:02 the local government bank.
07:05 AFD supports FEC through financing,
07:08 technical assistance on environmental,
07:11 social,
07:12 and climate issues.
07:13 In addition,
07:15 it gives targeted support to the most vulnerable municipalities.
07:20 And that's how we make co-financing work with each other,
07:24 yes,
07:24 and also with our partner,
07:25 the government of Morocco.
07:27 Thank you.
07:27 Thank you.
Morocco Municipal Performance Program
Chaymae Belouali, Senior Urban Development Specialist, World Bank, and Zoubida Bennani, Local Governments and Financial Institutions Specialist, AFD Groupe
When designed well, co-financing can strengthen the institutions that make services work for citizens over the long term. The World Bank and AFD Groupe developed a shared framework for measuring and improving municipal management—combining coordinated financing with complementary technical assistance and training. The result is a model that goes beyond building assets to building capacity: helping local governments perform better, manage resources more effectively, and deliver more for the people they serve.
Co-financing is one way multilateralism can respond to a constrained financing environment. And the evidence that co-financing works in practice, and can work at scale, exists. By putting countries first, aligning incentives, and acting collectively, development partners can deliver at the scale the moment demands.
RELATED
- lp-body-content
- first-letter