00:02 So we're gonna have 6 lightning talks,
00:03 uh,
00:04 uh,
00:04 to,
00:05 to follow,
00:05 uh,
00:05 session,
00:06 uh,
00:06 session 4.
00:08 I'm gonna MC,
00:09 uh,
00:10 the,
00:10 the lightning talks kinda get things,
00:11 uh,
00:12 moving along.
00:13 Um,
00:13 each lightning talk,
00:15 uh,
00:15 is gonna be 7 minutes,
00:17 and I'm gonna be mean to you if you go over time,
00:21 um,
00:21 but you know,
00:22 I hope that they're informative,
00:24 uh,
00:24 in,
00:25 in,
00:25 in a short,
00:25 uh,
00:26 in a short amount of time.
00:27 Um,
00:27 OK,
00:27 so we're gonna start with,
00:29 uh,
00:29 Marissa Noii from ID.
00:31 is 42 and I'm gonna let kind of the panelists
00:33 up here
00:34 get back to,
00:35 get back to their seats.
00:36 And just to quickly note that the reception this evening at 5 will be in the atrium,
00:40 which is up two floors,
00:42 the main,
00:43 um,
00:43 the main floor of the World Bank ground level.
00:45 So
00:46 that's which is exciting I've been told.
00:49 OK,
00:50 awesome.
00:53 Yeah.
00:55 And
00:56 if you
01:00 Alright.
01:01 Hi everyone.
01:02 My name is Marissa Nowicki,
01:03 and I work with Ideas 42.
01:05 Knowing I'm the only person keeping you from wine right now,
01:08 I'm gonna go through this quickly,
01:11 um.
01:11 Uh,
01:12 so today,
01:12 I'm gonna just give you a brief introduction on who we are,
01:15 who IDEAS 42 is.
01:17 Um,
01:17 we're going to talk briefly about three different case
01:20 studies and projects that we did during COVID,
01:23 um,
01:23 and how we pivoted to use mobile data collection
01:26 very briefly,
01:27 um,
01:28 and then some key insights that we learned from this experience.
01:32 So for those that don't know,
01:33 Ideas 42 is an independent nonprofit firm.
01:38 It was originally founded in 2008 by a
01:40 group of academics that worked in behavioral economics,
01:42 and they wanted to make sure that their work was applied to real world situations
01:47 and used for social good.
01:49 Essentially we study
01:50 human decision making and behavior.
01:55 Um,
01:55 and what we do is first we help define a behavioral problem.
01:59 We study the context
02:01 to understand why the desired behavior is or is not occurring.
02:05 Um,
02:05 we design a behaviorally informed intervention,
02:08 we test that solution and then we scale successful interventions.
02:12 And throughout our methodology,
02:14 data is crucial.
02:15 It's important to identify the behavioral barriers,
02:19 to create effective behaviorally informed interventions,
02:21 and then to make sure that our designs are working.
02:24 Um,
02:25 and during the COVID-19 pandemic,
02:27 we needed to utilize new and creative data collection methods so
02:30 that way we could pivot and continue doing our work.
02:35 So I'm gonna briefly talk about 3 case studies where we pivoted,
02:38 um,
02:39 and in these instances,
02:41 instead of doing our normal in-person activities,
02:44 we decided to do IVR or interactive voice response,
02:47 which is essentially when a pre-recorded message,
02:51 um,
02:52 is spoken to you and then you click a number which we've all interacted with,
02:55 you know,
02:55 uh,
02:56 for English,
02:56 please press 1.
02:58 I think we've all,
02:58 we've all experienced that and then also telephone interviews.
03:03 So the first one is from the DRC,
03:05 um,
03:05 and in one of our projects,
03:07 we are working to help beneficiaries of cash transfer programs,
03:11 um,
03:11 use this money to build resistance against economic shocks.
03:15 We knew we needed a really large sample for
03:17 this to run a fully powered randomized control trial,
03:19 so about 5000 was our goal,
03:21 and that that would require a lot of resources.
03:23 And as this was newer for us,
03:25 we decided to first do a pilot.
03:27 Make sure that we could
03:29 get the reach that we actually wanted,
03:31 so we ran a pilot IVR survey with about 200 randomly sampled people
03:36 to assess baseline response rate and if there was
03:38 anything we could do to improve that experience.
03:40 The good news was about 67% of people answered the phone
03:44 and about 20% completed the survey.
03:46 And a few key insights from this,
03:48 um,
03:49 were
03:50 that,
03:50 you know,
03:50 there were some technical difficulties,
03:52 which is,
03:53 uh,
03:53 I think a common theme that we want to be mindful of who can participate,
03:56 who might have a little bit of difficulty participating.
03:59 Uh,
03:59 some interesting notes on the language,
04:02 so the local language Langala was preferred for speaking and listening,
04:05 whereas French was preferred for written messages.
04:08 So having an option to have it in the local language,
04:11 but then the option to switch to French if they would like.
04:14 We saw that times of day,
04:16 um,
04:16 and the days of the week that we called were really important,
04:19 so calling during weekend work,
04:21 weekday afternoons
04:23 had the best response rate.
04:24 So understanding when people wanted to be called,
04:27 what languages were the most important,
04:29 and then of course offsetting any potential costs,
04:31 we did this through airtime so that people,
04:34 uh,
04:34 did not have to pay anything to participate.
04:36 So we use this to refine our approach and we
04:39 did ultimately run the RCT and got the 5000.
04:42 Um,
04:43 completed responses,
04:44 so that was a success that went well.
04:47 Um,
04:47 around the same time,
04:48 we were also working in Kenya,
04:51 um,
04:51 and prior to the pandemic,
04:52 we had been in the middle of working with the Kenyan
04:55 government to enhance the impact of social protection cash transfers,
04:59 and then when travel restrictions were in place,
05:01 our work had to stop.
05:03 Um,
05:03 at that time we had been in the process
05:06 of rolling out a wide scale randomized control trial.
05:09 We had already done the trial in 3 of the 4 intended counties,
05:13 um,
05:14 and the data collection was scheduled to begin a couple months later
05:18 and so to make the.
05:19 of the situation,
05:20 we decided to do something that was a contact-free approach,
05:22 so using mobile data collection this time through
05:25 phone interviews,
05:27 um,
05:27 and the goal was to assess,
05:29 among other questions,
05:31 we also had to pivot our research approach and our questions,
05:34 um.
05:35 But the goal was to see if the
05:37 behavioral interventions that we designed in a pre-COVID context
05:40 were still useful to those beneficiaries during the pandemic.
05:44 Uh,
05:44 and we were able to see using this,
05:48 uh,
05:48 using this tool that people who had been included in the interventions
05:52 were more likely to have concrete savings goals and were
05:54 more likely to actually be working towards those savings goals.
05:57 So although it wasn't a full replacement for in-person surveys,
06:01 it was a useful tool to help us see,
06:03 uh,
06:04 that there was an effect.
06:08 And then in the last project I'm just gonna briefly talk about,
06:11 we've been working with the World Bank to adapt a behaviorally
06:13 designed business training program which
06:15 we call the Financial heuristics training
06:18 to an audience of female small business owners in Ethiopia.
06:21 And initially we would,
06:23 or what we would usually do would be
06:24 to go to Ethiopia and do contextual reconnaissance,
06:28 um,
06:28 some behavioral diagnostics,
06:30 carry out the interventions in person,
06:32 but of course we couldn't do that,
06:34 so we had to adapt,
06:37 um,
06:37 and part of what we did was we hired someone
06:39 locally to conduct interviews over the phone and see,
06:43 uh,
06:44 get some baseline data,
06:45 see how we could refine our approach and how we could.
06:47 Adopt
06:48 it,
06:49 um,
06:49 and there were lots of lessons learned about this project,
06:52 about how people appreciated callback numbers and options,
06:55 um,
06:56 actually the value of just a step up from IVR but
07:00 giving them a little personal interaction was really valuable to them.
07:03 Um,
07:03 they appreciated it and of course this wasn't again a perfect substitute
07:07 but
07:07 was useful for us to collect useful data and then
07:11 pivot,
07:11 um,
07:12 and adjust our approaches.
07:14 So just I realized that was a lot of information very fast,
07:17 um,
07:18 in summary.
07:19 This is a really good tool to have in your tool belt.
07:22 Um,
07:23 I don't think this is a replacement for in-person data collection,
07:26 but it is a good way to complement data you're already collecting,
07:30 or
07:31 You know,
07:31 in the unprecedented times we're living in,
07:33 if you need to have a stopgap measure,
07:34 this is a good way to,
07:36 um,
07:37 to achieve what you need to achieve.
07:38 And then just some simple considerations that we found useful to think about
07:41 if you're considering using those tools were
07:44 the sample size,
07:45 the level of detail needed,
07:47 and the potential value add
07:49 of personal interactions.
07:53 All right,
07:54 thank you for everyone.
07:55 Uh,
07:55 if you have any questions,
07:57 feel free to check out our website or send me an email.
08:06 myself.
08:09 OK,
08:09 so our next speaker,
08:11 uh,
08:11 is Vivek,
08:12 um,
08:12 and would you mind pulling up the,
08:14 the
08:16 link,
08:16 link.
08:18 OK,
08:19 and this is what you wanted and needed.
08:22 Yes,
08:22 uh,
08:22 I'll start with the slides first though.
08:24 OK,
08:24 great.
08:25 Just on the other.
08:26 Thank you.
08:28 So when you wanna switch it,
08:30 just like slide it.
08:31 Perfect.
08:31 All right,
08:32 Vivek Saranni from Atlas AI.
08:34 Thank you.
08:35 Hello everyone,
08:36 I'm Vivek Sarani,
08:37 head of applied data science at Atlas AI.
08:39 Uh,
08:40 before jumping into
08:41 some of the technical components of the talk,
08:43 a quick word on our company.
08:45 Since,
08:45 uh,
08:46 I'm not affiliated with the university or the bank,
08:49 we are a software engineering company that uses
08:52 remote sensing,
08:53 uh,
08:53 machine learning,
08:54 and,
08:54 and ground data sources to build
08:56 high resolution data sets and then translate that into
08:59 urban and rural applications for both the public
09:02 sector as well as the private sector.
09:05 So in this talk,
09:06 I'll focus on this idea of where people spending their time and money.
09:09 It's shown up in many ways
09:11 today,
09:12 but the main difference I would say between many of the talks today and this one
09:16 is that we're particularly focused on urban service design.
09:20 So how can the public sector or private companies focus on Where to
09:24 increase capacity or deliver
09:26 additional services.
09:27 So probably Alice's talk comes closest
09:30 to this and thinking about transportation applications
09:33 in the Lagos context.
09:34 And so the,
09:35 the,
09:35 the easy key question to hold on to in,
09:38 in this,
09:38 this idea is where are people spending their time and money.
09:42 And we're looking at mobility data in the Nairobi
09:45 context.
09:46 We're also combining this with high resolution data sets,
09:49 uh,
09:49 for example,
09:50 the asset wealth
09:51 index,
09:52 which is an indicator of poverty.
09:54 In,
09:55 in Nairobi,
09:55 this is a proprietary data set that we built ourselves,
09:58 and then curated sources on urban,
10:00 urban built environment,
10:02 for example,
10:03 Google,
10:03 Google open buildings,
10:05 but also other types of scraped information like points of interest data sets.
10:10 And then a number of the methods which uh researchers in this room are,
10:14 are familiar with,
10:15 uh,
10:15 including
10:16 safeguards for,
10:17 for the data
10:18 and then also some of the optimization techniques for,
10:21 for modeling.
10:24 So,
10:24 uh,
10:24 I'll skip over some of the contextual slides since,
10:26 uh,
10:27 at this point in the day we're all very
10:28 familiar with some of the challenges and issues,
10:31 but again,
10:32 uh,
10:32 the idea that I'll focus on here in this talk is
10:35 that we're interested in not just the location properties of mobility.
10:39 Also the interaction of that mobility information with the built environment,
10:43 whether it's physical assets,
10:45 assets like roads and transportation infrastructure
10:48 or built places,
10:50 stores,
10:50 retail environments,
10:52 homes,
10:53 or other types of facilities.
10:58 So,
10:58 in,
10:59 in the study,
10:59 we're looking at 3 years' worth of mobility data for all of Kenya.
11:04 Uh,
11:05 but for today,
11:05 I'm focusing on
11:07 about 3.5 million observations from 1 week alone.
11:10 And within the Nairobi envelope,
11:12 that's about almost 2 million
11:15 observations from 150,000 unique devices.
11:19 So all the caveats of sample bias and the,
11:22 the,
11:22 the,
11:22 the limits on who owns smartphones
11:24 obviously hold in this case as well.
11:27 Ah,
11:27 but we're gonna try to dig a little bit deeper into them.
11:31 Uh,
11:31 the proprietary Atlas AI data is,
11:34 um,
11:34 1 kilometer by 1 kilometer resolution,
11:37 uh,
11:37 indices on asset wealth,
11:39 spending power measured in dollars per person per day,
11:42 uh,
11:42 calibrated to the international purchasing power parity metrics,
11:46 uh,
11:47 from studies by the World Bank and,
11:48 and others,
11:49 uh,
11:50 and then also details on infrastructure access including electrification.
11:54 Uh,
11:54 built environment,
11:55 access,
11:56 for example,
11:56 access to healthcare,
11:57 schools,
11:58 and so on.
11:59 And so by looking at this at high resolution,
12:01 we're able to triangulate this with the mobility data,
12:03 uh,
12:04 and trace,
12:04 uh,
12:05 the pathways as well.
12:07 Here's an example of uh Google open buildings for Nairobi.
12:11 It's probably one of the highest coverage,
12:13 uh,
12:13 data sets for,
12:14 uh,
12:15 for building footprints in the area.
12:17 Uh,
12:17 and then the OSM road networks which are also well developed,
12:20 uh,
12:21 for this part of Kenya.
12:24 Uh,
12:24 just a quick overview on methods,
12:26 uh,
12:26 combining these data sources,
12:28 so we run through,
12:29 uh,
12:30 some of the typical techniques.
12:31 Here,
12:32 uh,
12:32 we're thinking about being able to,
12:34 to set up these methods in a way that can process data at scale.
12:38 So,
12:38 as I said,
12:39 we're going to go back and look at 3 years' worth of information.
12:42 So our,
12:42 our machine learning pipelines and our,
12:45 our archival uh or uh architectures need to be
12:48 able to consume that amount of information on a,
12:50 on an ongoing basis.
12:51 So if we were to
12:52 expand this to a country scale or to multiple cities
12:55 on the continent,
12:56 for example,
12:57 and what comes out of this is the typical indicators
13:00 that we've looked at in terms of dwell times,
13:02 visit frequencies,
13:03 routing across infrastructure,
13:06 proximity to,
13:07 to various public and private services.
13:11 So what this shows is uh one device and its pathways,
13:16 so one trip,
13:17 one day,
13:17 and one week.
13:19 Uh,
13:20 I've shown it all for one week here for,
13:22 for privacy concerns,
13:23 so that way we aren't able to,
13:25 to pinpoint
13:25 the individual's
13:27 base location or work location.
13:29 Uh,
13:30 we run through the typical steps of
13:32 stop identification to identify the unique locations.
13:36 There's a massive amount of preprocessing
13:38 required here to make sure we're actually
13:40 picking up unique locations as opposed to the
13:44 jitter that's often observed in GPS points.
13:48 We look at a number of destinations of interest,
13:51 so retail and leisure.
13:52 Uh,
13:52 this is a large mall in the Lamington area of Nairobi
13:56 that attracts multiple devices and users
13:59 to that location,
14:00 and then over the course of a week,
14:02 we observed that there are about 200 unique
14:04 visitors from this part of the data sample
14:07 to that location and,
14:08 and those retail facilities of interest.
14:11 So what this helps us understand is not just where
14:13 the people coming from across the urban fabric of Nairobi,
14:16 but in their
14:17 leisure time or their work time,
14:18 where they spend that time and money.
14:22 So I'll switch over to um.
14:31 3 fingers.
14:33 Oops.
14:38 Thank you.
14:39 So,
14:39 for today,
14:40 uh,
14:40 we put together a,
14:41 a demo where we can look at that data interactively.
14:44 So imagine that you're a government entity
14:46 and trying to do road infrastructure planning
14:48 or uh urban or if you're a telecom company
14:51 trying to understand where existing signatures come from,
14:54 but you know that there's a population that hasn't received uh self service yet
14:59 as you're thinking about expanding.
15:00 Your towers or
15:02 a repeater coverage,
15:04 this type of analysis can be useful.
15:07 And so once the safeguards have been built in,
15:09 we're able to then
15:11 showcase
15:12 some of the individual traces as well,
15:15 to be able to pick up information like where the visits are and then be able to
15:20 aggregate to these.
15:22 So,
15:22 uh,
15:23 one of the advantages of planning for this architecture at scale is
15:26 that we're able to deliver the insights directly to a variety of users
15:30 as opposed to,
15:31 uh,
15:31 expecting that every organization that needs this information
15:34 has the capability to,
15:35 to crunch this information on
15:39 on their own or implement the research methods in
15:40 the way that we as researchers would do.
15:44 So,
15:44 uh,
15:45 I'm probably out of time,
15:46 so I'll stop there.
15:47 Um,
15:48 uh,
15:48 one of the main sort of policy implications that
15:51 we haven't touched on today that I'll speak to
15:53 is the idea of procurement.
15:55 For example,
15:56 during COVID,
15:56 we often receive calls from the bank
15:58 on can we use this type of information to continue
16:01 to support bank operations,
16:03 uh,
16:03 and also from ministries who were focused on other
16:06 areas of public need like access to healthcare.
16:09 To be or for example,
16:11 to,
16:11 to help deliver
16:12 uh COVID vaccines in relation to where they were observing activity of people,
16:17 uh,
16:18 and so on in triangulation with,
16:20 uh,
16:20 population density,
16:21 spending patterns and asset wealth.
16:23 So there are a number of applications there,
16:25 um,
16:26 that,
16:26 uh,
16:26 aren't sort of causal inference oriented but,
16:29 uh,
16:29 more designed for immediate services.
16:31 Thank you.
16:39 Thank you,
16:39 Vivek.
16:40 Um,
16:40 our next speaker is Barbara Kramer from IPRI.
16:48 Yeah,
16:48 thank you so much for having me.
16:49 Um,
16:50 also,
16:51 um,
16:52 it's just been really inspiring to see so many great talks today,
16:55 um,
16:55 and I think this is a good time for me
16:58 to put in a plug for the journal where I'm,
17:01 uh,
17:01 editor of the Social Science section.
17:03 Which is uh development engineering.
17:05 and development engineering,
17:07 um,
17:07 publishes work um on rigorous research,
17:11 on innovations to address global poverty
17:14 problems or global development problems.
17:17 Um,
17:17 and then that includes a lot of the mobile data work that has been presented today.
17:21 Both mobile data in terms of strengthening policies and programs,
17:24 but also using mobile data to track
17:27 impacts um of different interventions or to track
17:30 um outcomes and.
17:32 And so in that light also,
17:33 it's good to mention that we currently have a special issue out
17:36 on tracking progress towards the sustainable development goals
17:41 um and if anybody is interested in submitting a paper um do talk to me
17:46 or to Sam Fishman,
17:47 um,
17:48 who is also part of the editorial team and and keeps us as editors uh on track.
17:54 Um,
17:54 so having said that,
17:55 let me now turn my eyes on the ground and talk about,
17:59 uh,
17:59 crop monitoring.
18:01 Uh,
18:01 through smartphone pictures of plots.
18:03 Um,
18:03 this is,
18:03 uh,
18:04 a project,
18:05 um,
18:05 that is joint work with Koon Hufkins,
18:08 who is with Ghent University.
18:10 Um,
18:10 Lillian Weissaka and Benson Juguna,
18:13 who are with Acre Africa,
18:15 um,
18:15 the implementing partner for this work,
18:17 um,
18:17 Acre Africa is an insurance broker and does a lot of work around,
18:21 uh,
18:22 smartphone-based data collection.
18:24 Um,
18:24 and then we're working with,
18:25 um,
18:25 Dwara e-registry,
18:27 which is,
18:27 um,
18:28 an Ag FinTech startup in India,
18:31 um,
18:31 and the Kenyan Agricultural Livestock Research Organization.
18:36 Um,
18:36 so why crop monitoring?
18:38 Why are we talking about,
18:39 you know,
18:39 getting our eyes on the ground on farms?
18:42 Um,
18:42 smallholder farmers,
18:43 they depend
18:44 on,
18:45 uh,
18:45 agriculture for their subsistence,
18:47 and in developing countries,
18:48 a lot of agriculture is rain fed and thereby is exposed,
18:52 uh,
18:53 tremendously to weather risks.
18:55 Um,
18:56 and,
18:56 and this includes droughts,
18:58 floods,
18:58 also,
18:59 you could think of pests and diseases.
19:01 Um,
19:02 and,
19:02 and monitoring this,
19:03 it's just really,
19:04 really important
19:05 to help improve early warning systems,
19:07 um,
19:08 to,
19:09 to provide agricultural advisory services that would allow farmers
19:12 to take preventive measures to minimize damage
19:16 from these extreme weather events,
19:17 um,
19:18 but also financial products such as credit
19:20 or crop insurance to help improve adaptation,
19:23 uh,
19:24 to and resilience to climate change.
19:26 Um,
19:27 now,
19:28 advances in remote sensing have been enormous in the last few years
19:33 to really improve monitoring of crop phonology and,
19:36 and productivity,
19:38 but there's still a number of challenges that are remaining,
19:41 including some technical aspects.
19:43 For instance,
19:43 cloud cover in a rainy season,
19:46 um,
19:46 is is is limiting the accuracy of some
19:48 of the satellite-based algorithms to predict productivity.
19:53 Then there's,
19:53 um,
19:54 resolution issues.
19:55 For small plots,
19:56 there's intercropping in the sense that in a lot of smallholder farming settings
20:01 and as a measure to adapt to climate change,
20:04 we see that farmers are growing,
20:05 for instance,
20:06 beans and maize
20:08 in one plot and how to make sense of that when you're using
20:11 satellite remote sensing,
20:13 how to estimate productivity in those cases.
20:16 Then there's issues that are more social.
20:19 People don't always trust what's being read from satellite images.
20:23 It's not very tangible
20:25 and so understanding becomes an issue
20:28 as well.
20:28 And finally,
20:29 on the empirical side,
20:30 there's a lack of ground truth data
20:32 and a lot of models that have been developed,
20:35 um and a lot of validation studies that have been
20:37 done are focusing on major cereals such as maize,
20:40 wheat or
20:42 uh rice,
20:42 but not on
20:44 other crops that are very important for dietary diversity
20:48 um and for vulnerable populations and more indigenous crops.
20:53 So,
20:53 with that in mind,
20:54 um,
20:54 we started a project in 2016,
20:57 um,
20:58 that is using smartphones
21:00 to,
21:01 we call it smartphone-based near surface remote sensing.
21:04 Um,
21:05 that is using smartphone pictures to monitor crops,
21:07 and it works as follows.
21:09 A farmer can download an application
21:11 on his or her phone,
21:13 uh,
21:13 register a site using an initial picture,
21:16 and then over the course of the season,
21:19 take repeat pictures of that same site,
21:21 and there's visual aids in the smartphone application
21:23 that help tracking this.
21:24 There's also GPS coordinates and timestamps being added,
21:28 so that when the photos are uploaded to the server,
21:31 um this information is all available.
21:33 And then in the application that we're focusing on crop insurance,
21:36 there's actually experts that are reviewing the images and
21:39 indicating whether this farmer has suffered damage and should receive
21:42 an insurance payout.
21:44 And we did a formative evaluation of this in India
21:47 whereby we found that indeed it is possible to track
21:51 uh or to to do crop monitoring through this approach,
21:54 especially to identify crop growth stages,
21:57 um,
21:57 which is useful in the case of advisories
22:00 and to um.
22:02 Uh,
22:02 to,
22:03 um,
22:03 uh,
22:04 identify severe damage which is useful in the case of insurance,
22:08 and then the question became can we do this in
22:09 other settings and so we started working on this in,
22:12 in Kenya,
22:13 um,
22:13 and can we further automate image processing.
22:16 And so in that light,
22:17 um we did a project at a very large scale collecting images
22:23 in Kenya um
22:25 through
22:26 in Kenya where smartphone ownership is a lot a lot lower
22:29 than in the setting in India where we were working,
22:32 um,
22:32 we worked through champion Farmers,
22:34 um,
22:34 which are local opinion.
22:35 Shapers who were equipped by Acre Africa with a smartphone and trained
22:40 on how to send in these images
22:42 and they registered a network of more than 100 farmers per champion
22:47 and out of those we had 20 farmers for whom they took regular photos
22:52 um over the course of the season
22:55 for 3 seasons in a row.
22:57 And all these images were labeled,
22:59 um,
23:00 and in terms of crop growth stage,
23:02 type and extent of damage.
23:04 So,
23:05 um,
23:06 that resulted in a data set of more than 13,000 maize images,
23:10 uh,
23:10 for the three seasons combined.
23:12 Those have all been labeled,
23:14 um,
23:14 and
23:16 out of those,
23:16 we have more than 11,000
23:19 images.
23:20 That are useful that are out there actually
23:23 on the web
23:24 um the images were also used to train
23:26 convolutional neural networks um because so using computer vision
23:30 to detect the labels that we had
23:33 um uh created
23:35 or that the agronomists had been assigning to the images
23:39 um and and other seasons are being added over time.
23:42 So,
23:42 the reason that I was presenting this is that I think this is a great
23:46 data set for those who are interested in agriculture AI and who have remote sensing
23:52 um
23:53 applications who are looking for Ground truth data sets.
23:56 Um,
23:57 this data is out there on the web.
23:59 Um,
23:59 I have a web a website indicated here.
24:02 Um,
24:03 and so also if you have further questions about the data set,
24:06 uh,
24:07 do feel free to get in touch,
24:09 um,
24:09 because we're happy also to share,
24:11 for instance,
24:12 a smartphone application if people want to collect this type of data
24:15 in other locations.
24:16 So,
24:17 that's it.
24:17 Thank you so much.
24:24 Thank you so much Berber.
24:25 It seems a recurring theme is that taking pictures
24:29 of stuff with your cell phone is really great,
24:31 whether it be bacteria or
24:34 worms.
24:34 Sorry,
24:35 I can't,
24:36 I don't remember exactly which.
24:37 Sorry,
24:37 Maria,
24:38 um,
24:38 OK,
24:38 so our,
24:39 our next,
24:39 uh,
24:40 presenter is Sarraj Nair.
24:41 Uh,
24:41 he's been mentioned a couple times in the conference and now here is the man himself,
24:46 not to put you on the spot.
24:47 Thank you.
24:50 Um,
24:52 Oh,
24:52 awesome,
24:52 thanks.
24:53 Um,
24:54 all right.
24:55 Good evening,
24:56 everyone.
24:56 I guess we're almost done.
24:57 Wine,
24:58 woo-hoo.
24:59 Um,
25:00 yeah,
25:00 I'm excited to talk to you today about,
25:02 um,
25:02 ongoing work with Josh Blumenstock from UC Berkeley.
25:05 Um,
25:05 we've been looking at the impact of cash transfers on
25:07 mobile money usage and then more broadly digital financial inclusion.
25:11 Um,
25:12 in terms of motivation,
25:13 I guess we've heard several times today that in the last couple of years,
25:17 there's been a massive digitization of social protection.
25:20 Um,
25:21 I think this is the last accurate number.
25:23 I have over 150 programs delivered aid
25:26 via digital channels during the COVID-19 pandemic,
25:29 um,
25:30 and a large number of these programs relied on mobile money to deliver this aid,
25:34 especially in,
25:34 um,
25:34 sub-Saharan Africa.
25:37 Our paper focuses on this idea that
25:40 delivering aid via mobile money or mobile money transfers
25:42 could serve as an on-ramp to broader financial inclusion
25:46 by providing access to very basic financial services for low-income individuals
25:51 who are often excluded from the traditional banking system.
25:54 So,
25:55 this paper very specifically simply looks at this question
25:57 of whether short-term exposure to these digital cash transfers can
26:01 drive the longer-term usage of the broader mobile money ecosystem.
26:05 I guess it complements this very
26:07 quickly growing set of papers that generally look
26:10 at the impacts of mobile money transfers.
26:12 Most of those papers focus on
26:14 welfare outcomes such as consumption or
26:16 women's economic empowerment or intra-household bargaining
26:19 or on efficiency gains for firms and employers.
26:22 Much of this literature
26:24 finds that short term exposure to
26:25 cash transfers doesn't really change financial behavior
26:28 in the long term.
26:29 Um,
26:30 at least in the past,
26:31 pre-pandemic.
26:31 We think this is going to change after the
26:34 evidence from the COVID-19 studies come out.
26:38 Um,
26:38 the context we're focusing on is that of the OBC program in Togo.
26:42 Um,
26:42 Emily spoke at this,
26:44 uh,
26:44 spoke about this in the morning at length,
26:46 so I'm not gonna spend too much time on this.
26:47 I just wanna mention that we focus on a phase of the program.
26:51 Um,
26:51 that,
26:51 that had a randomized rollout.
26:53 So a random set of individuals were assigned to a treatment group
26:56 who began receiving cash transfers in November 2020 for a five-month period,
27:00 following which the control group received transfers.
27:02 So we leveraged this randomization to study these impacts on mobile money usage
27:07 that I'm going to talk about.
27:08 Um.
27:10 Yeah,
27:10 that's the only thing I wanted to mention here.
27:12 Um,
27:12 I won't spend any more time.
27:14 We,
27:14 we use data from uh one of the largest telecom operators in Togo.
27:18 Uh,
27:18 this is a rich data set that
27:20 gives us information on when individuals sort of
27:23 withdraw and deposit money into their mobile money accounts,
27:25 but it also gives us rich insight into their saving behavior,
27:28 into the network of financial transfers,
27:30 and how they're using the broader mobile money
27:32 ecosystem to make digital payments and purchase airtime
27:35 and to pay for things like cable and water and electricity and things like that.
27:40 Um,
27:41 yeah.
27:42 So before I jump into the main results,
27:43 I think
27:44 my,
27:44 the thing I really like about the analysis so far has
27:47 been that simple descriptive stuff provides us a lot of insight.
27:51 So,
27:51 I'd like you to focus on this panel on the top where I'm simply plotting out
27:55 the fraction of individuals making at least one
27:58 mobile money transaction in a given month.
28:00 Remember that the transfers we're interested in began in November 2020.
28:05 So,
28:05 prior to that,
28:06 basically,
28:07 roughly 20% of our sample was making a mobile money transaction.
28:11 The moment people start receiving these transfers,
28:13 you see this huge spike in activity in their mobile money accounts which
28:17 um
28:18 kind of sustains for the five months during which they
28:20 receive transfers and then there's a decline of sorts.
28:23 The interesting thing here,
28:24 um,
28:24 and I think this is what differentiates these results from,
28:26 I guess,
28:27 stuff that's out there in the past is Panel B where
28:30 we're able to show that
28:31 A lot of this increase in mobile money usage is actually driven by
28:36 an increase in the use of wider ecosystem
28:38 services such as bill payments and transfers and
28:41 other things,
28:42 so people are not simply using their accounts to withdraw cash
28:45 withdraw cash and then spend it in cash locally.
28:47 There's actual ecosystem-wide transactions happening in this setting at least.
28:51 Right?
28:51 And so this sets up
28:53 sort of my
28:54 one results side.
28:55 I promise not to show tables,
28:57 so
28:58 it's all text.
28:58 Um,
28:59 but yeah,
28:59 so basically,
29:00 what happens when individuals receive transfers in a month,
29:02 they receive transfer,
29:03 we see this like
29:04 62% increase in the
29:06 usage of their mobile money accounts relative to baseline levels,
29:09 that's a 300% increase.
29:11 Um,
29:12 the nice thing,
29:12 the encouraging thing rather is that these increases
29:15 are ecosystem-wide,
29:17 so there's people actually making digital transactions during
29:20 this period.
29:20 Um,
29:21 what's,
29:22 what we are trying to look at and get a
29:24 better sense of is what happens in the longer term.
29:26 So after the cash transfers conclude,
29:29 do these gains persist or does everyone go back to using cash?
29:32 And I think at least our initial results
29:34 suggest that there are modest gains in mobile money
29:37 activity,
29:37 mobile money account usage in the longer term.
29:40 Our data
29:41 lasts for about 4 months after the cash transfer program ends.
29:44 What I'm showing you here are results from an event
29:46 study analysis where we're comparing individuals who received a cash transfer
29:50 to individuals who never received one.
29:53 So,
29:53 relative to baseline levels of
29:56 mobile money usage,
29:56 these results show the same thing.
29:58 There's this large initial increase in the months
30:00 during which they receive a cash transfer,
30:02 and then there's a drop-off of sorts in mobile money account activity,
30:05 but
30:07 Even 4 months after the cash transfers conclude,
30:10 individuals who receive transfers remain much more likely to use mobile money than
30:15 individuals who don't receive these transfers.
30:17 So,
30:17 that's the broad pattern we see.
30:19 Um,
30:19 we think there are some specific,
30:21 um,
30:21 factors in Togo that contribute to these increases.
30:24 Um,
30:24 Togo,
30:25 they subsidized the transaction costs to withdraw money,
30:28 at least initially.
30:29 Uh,
30:29 we think this helped.
30:30 There's a dense agent network and the Togo government was
30:32 also running these financial literacy programs even before the pandemic,
30:36 all of which could have helped familiarize people with some of these aspects and
30:40 um.
30:42 Lead to the results that we're seeing.
30:43 What are the implications of this?
30:45 We think the NoviC program,
30:46 based on rough back of the envelope calculations,
30:49 added
30:50 at least 80,000 customers to the ecosystem in the long term.
30:53 Um,
30:54 and again,
30:54 using like reference numbers available from exercises that
30:57 have been done in West Africa and Kenya,
30:59 we think
31:00 this means up to 80 million
31:02 um Togolese francs in additional revenue and potentially up
31:05 to 160 million in terms of true economic value.
31:08 Uh,
31:08 added to the economy.
31:09 But
31:10 these are back of the envelope calculations,
31:12 we're trying to be better about this,
31:13 um,
31:14 but yeah,
31:15 that's my results.
31:16 Um,
31:17 thank you.
31:18 Feedback is welcome.
31:19 This is the first time we're presenting these results,
31:20 so we would love to hear your thoughts.
31:23 Thank you.
31:30 Thank you so much,
31:31 Sarraj.
31:32 Um,
31:32 next up we have,
31:33 uh,
31:34 Winnie Mugogu from Busara.
31:36 Uh,
31:36 we're really excited to have her and thank her for,
31:38 thank her for coming all the way to Kenya to pre from Kenya to,
31:42 uh,
31:42 to present,
31:43 uh,
31:43 these,
31:44 uh,
31:45 results,
31:45 and I'm hoping that while I was talking,
31:49 Napur would come to change the slides,
31:50 but that has not happened.
31:51 All right.
31:57 Oh,
31:57 there we go.
31:58 Cool.
32:04 Um,
32:04 did it all by myself.
32:05 All right.
32:06 All right,
32:06 Winnie,
32:06 the floor is yours.
32:12 OK.
32:12 Hello,
32:13 everybody.
32:13 Um,
32:14 today I'll be presenting on um moving from
32:16 in-person to mobile data collection in Kenya.
32:19 I should have said at Busara,
32:21 but the at Busara is silent.
32:23 Um,
32:24 I'm a research specialist at Busara Center for Behavioral Economics,
32:27 which is in Nairobi,
32:28 Kenya.
32:29 So I'll just give a background on Busara.
32:32 Um,
32:33 methods,
32:33 research agenda that we have going on and
32:35 some studies that use mobile data collection.
32:38 So I'll highlight a lot on the lessons learned from these studies
32:41 more than the specific results.
32:43 If anybody wants,
32:44 is interested in any of the studies and the specific results,
32:47 they can consult me on that.
32:49 So,
32:50 background.
32:50 Um,
32:51 Basically,
32:52 BUSARA Center um is a research institution
32:55 and a consultant um they do consulting work
32:58 with researchers and organizations to advance and apply
33:01 behavioral science in pursuit of poverty alleviation.
33:03 That's the mission of BUSARA.
33:05 So basically,
33:06 we have a nice
33:07 behavioral lab in Nairobi,
33:09 Kenya.
33:10 Possibly the only one in Africa,
33:12 but I could be wrong,
33:13 so
33:14 anyone can correct me on that one.
33:16 but we um conduct some behavioral studies in Nairobi,
33:19 Kenya.
33:20 In March 2020,
33:21 I moved from my home country,
33:23 which is Malawi
33:24 to Kenya.
33:25 And I arrived in Kenya and a few days later,
33:29 there was the first COVID case.
33:30 A few days later,
33:32 they said we're not going to work next week.
33:34 So I never actually got to see this lab and I never got to experience um
33:38 all the wonderful experiments that people carry out.
33:41 But as a research institution,
33:43 of course,
33:44 we had to somehow carry out the research.
33:46 We needed Data.
33:47 Um,
33:48 as such,
33:48 yeah,
33:49 the pandemic,
33:50 it came in the way of um
33:51 data,
33:52 in-person data collection
33:53 as most of the studies today have been motivated by,
33:55 you know,
33:55 the pandemic
33:57 and so
33:58 we had to think of how do we continue to collect data,
34:00 otherwise,
34:00 we can't carry out our operations.
34:03 And so
34:04 we decided,
34:04 well,
34:05 remote or mobile data collection
34:07 and this is in a low-income country setting um
34:11 We had the subject pool of respondents in
34:14 Kibera and Kwangware,
34:15 which are
34:17 um urban slums in Nairobi,
34:18 Kenya.
34:19 If you see a lot on slums in the news,
34:21 it's probably in Kibera,
34:22 very um the largest urban slum in Africa,
34:25 actually.
34:26 Um,
34:27 and so as we were thinking around how do we continue um
34:31 data collection,
34:32 that's when we came up with this methods research agenda.
34:35 So I had the methods research agenda at BASARA
34:38 and basically,
34:39 we are,
34:40 we have a few questions that we're trying to answer.
34:42 I've just selected two of the most prominent ones
34:44 um that we're trying to answer with this methods research agenda.
34:48 It's how we can contextualize,
34:50 measure
34:51 and improve access.
34:53 Response,
34:54 attention,
34:54 comprehension
34:56 of surveys
34:57 and experiments for populations in the global South.
34:59 And this is going to be specifically be applied to remote
35:03 data collection because that's what happened during the COVID pandemic.
35:06 We couldn't do in-person
35:07 data collection.
35:08 So,
35:08 we're looking at,
35:09 you know,
35:10 still improving access,
35:12 still getting good quality data,
35:14 still getting um
35:16 you know,
35:16 good attention even though the,
35:17 the
35:19 research is being carried out remotely.
35:21 And then how do you,
35:22 how do the answers to these questions vary across gender,
35:25 racial,
35:25 national,
35:25 and economic groups?
35:26 I think a lot of the discussion today as well has been,
35:29 you know,
35:29 around
35:30 um do we see differences in gender when we're using mobile data collection.
35:34 These are some of the questions that we're going to be answering
35:36 and some of the studies that I'm going to share.
35:39 Um,
35:40 begin to answer those questions.
35:42 It's going to be carried out over the next 3 years.
35:44 If anybody's interested in collaborating to answer these questions,
35:47 um feel free to do so.
35:49 So,
35:50 One of the ways we innovated was through
35:52 what we called Kite.
35:54 Kite is um which sought for Knowledge and Insights tool for experiments.
35:59 So basically,
35:59 we wanted to continue doing this behavioral econ um experiments
36:03 but on,
36:04 on a mobile app.
36:05 Um,
36:06 so
36:07 It was created in-house in Bossara
36:09 and one of the studies that we um
36:12 we did on
36:14 the app was a financial safety nets and attention study.
36:17 By that,
36:17 it was
36:18 how are people coping over the
36:21 COVID pandemic financially.
36:22 I think these are important questions to ask at a time where
36:24 we couldn't visit the people in uh like we couldn't go in person
36:28 but we wanted to know how they're doing financially.
36:30 Um,
36:31 and then we added in this part of attention.
36:34 Now,
36:34 attention is because
36:35 when you're thinking of remote data collection,
36:37 you don't have the enumerator or
36:40 the FO to the field officers to
36:42 keep people's attention going.
36:44 So how are we so sure that the um with remote data,
36:46 people aren't just clicking away at answers and then submitting?
36:49 I'm sure we've done that sometimes when you get an email
36:52 asking you to complete a survey and just kinda click,
36:54 click,
36:54 click submit.
36:55 Um,
36:56 and so
36:57 we try to Answer
36:59 questions like
37:00 the sources of income people used before and after the pandemic,
37:04 but also the attention
37:05 part would be would ask questions like,
37:07 were you born before 1940.
37:09 The inclusion was people who were a specific age and
37:11 there was no way they should be born before 1940.
37:14 That's how we would measure something like attention.
37:17 So,
37:18 we ran this with um a target sample of 500 people but you only reached 375.
37:23 Um
37:24 getting people to download the app was quite a challenge um and so
37:28 we're actually pivoting
37:30 how we're going to use Kite so that we include a lot of people
37:33 um
37:35 given how difficult it was to get people to download it.
37:38 Um,
37:39 this is a study of,
37:40 uh,
37:40 chlorine,
37:41 uh,
37:41 the study's actually called Can Simple Psychological
37:43 Interventions Increase Investment in Preventative Health.
37:46 I've included this study but it's by Kate Orkin and Annette John.
37:49 Kate Orkins at Oxford.
37:51 Um,
37:51 collaborated with Bosara.
37:53 When I'd arrived,
37:54 we were supposed to do a follow-up survey
37:57 in person.
37:58 We had to pivot to,
38:00 um,
38:00 a phone survey.
38:02 We managed to reach
38:03 3,194 women in rural
38:06 Kenya,
38:06 in
38:07 Bungoma and Kakamega.
38:09 And to reduce attrition,
38:10 we went and tracked respondents on the ground,
38:14 but they still did the surveys over the phone
38:17 and for those who didn't have a phone,
38:19 the people who did the tracking,
38:20 they borrowed the respondent the phone so that
38:22 they can still take part in the survey.
38:24 So that's one of the ways in which you try to
38:25 include people who would otherwise have not been in the survey.
38:28 Um,
38:29 I'll skip this,
38:30 uh,
38:30 study on we use SMSs because there's the one study
38:33 which we found very similar results that was just presented,
38:36 um,
38:37 and cause of time.
38:38 Um.
38:40 And then we've also tried um
38:43 to do call tricks and
38:45 compare that to phone surveys.
38:46 I'd actually initially wanted to present on this study
38:49 on using the LT experiment,
38:50 which is an indirect method of measuring intimate partner violence,
38:55 um comparing it over result um findings on court tricks and
38:59 on a phone.
39:00 It has taken quite some time to start data collection because of IRBs and you know,
39:04 the sensitive issue of IPV
39:06 but we finally got an approval.
39:07 We're starting data collection this week
39:09 if anybody's interested to find out.
39:11 Um,
39:12 just the idea behind this um study with the list experiment is that
39:16 um with intimate partner violence,
39:18 with sensitive topic.
39:19 topics,
39:19 people aren't so open to share about their experiences
39:23 and so the idea with this methods research agenda is that we
39:26 want to make sure that we are collecting good quality data.
39:30 And so
39:31 today,
39:31 the idea is how do these things,
39:33 how do these things affect policy?
39:35 What,
39:35 uh what we're trying to get from these studies is basically that
39:38 Policy must be guided by evidence
39:41 and can we trust the evidence?
39:43 We can only trust the evidence if they come from good methods
39:46 and that's why we looked at things like this.
39:49 So um that's why I had to share.
39:52 Like I said,
39:52 there's so much I could have shared
39:54 specifics of the study,
39:55 but
39:56 yeah,
39:56 I just didn't have time for it.
39:57 Um
39:58 thank you.
40:05 That's a great quote.
40:07 Uh,
40:07 you can only trust the evidence if it comes from good methods.
40:10 Um,
40:10 so our last speaker of the day,
40:12 uh,
40:12 is gonna be,
40:13 uh,
40:13 Zach White from GSMA,
40:15 which,
40:15 as we've seen in,
40:16 in,
40:16 in several of the presentations of prior speakers,
40:18 you know,
40:19 really makes possible a lot of,
40:20 uh,
40:20 a lot of the research.
40:21 So,
40:21 um.
40:23 Yes.
40:24 All right,
40:24 great.
40:24 Thank you.
40:34 This is it.
40:35 That's it.
40:43 Fuck.
40:44 Uh,
40:45 so I really am standing in the way
40:47 of the last thing.
40:49 But I'm gonna do 3 things today which is just introduce the GSMA and
40:52 then move on to 3 frameworks that we use to look at partnerships,
40:57 uh,
40:57 and particularly around mobile big data.
41:00 And then I'm gonna talk through some of the cases
41:03 where the GSMA played a very active role in facilitating those
41:07 in the context of the COVID-19 response.
41:10 And you can count them down as 321,
41:13 should you choose to.
41:15 Uh,
41:15 so the GSMA and talking to people it seems about 50/50,
41:18 uh,
41:18 those who know us,
41:19 those who don't.
41:20 It's the industry association for the mobile industry,
41:23 so our members are about
41:25 800 of all of the major mobile network operators and associated ecosystem players.
41:31 Uh,
41:32 we convene the industry,
41:33 there's industry services and events,
41:35 but then also within the mobile,
41:37 uh,
41:38 the GSMA is the Mobile for Development Foundation
41:41 that sits at the interface of development and the industry.
41:45 And I am one of,
41:46 I work with the Digital Utilities Program which is
41:48 one of 9 kind of thematic programs working across,
41:52 you know,
41:52 agriculture,
41:53 mobile money.
41:55 Uh,
41:56 uh,
41:56 and many others that
41:58 my brain can't recall recall right now.
42:00 Uh,
42:00 and within the utilities program we focus on energy,
42:03 water,
42:03 transport,
42:04 sanitation,
42:05 and waste management,
42:05 those services.
42:07 And what we do in,
42:08 this is similar across all of the
42:11 programs and portfolios is most importantly,
42:14 one is we give innovation grants uh to predominantly startups and SMEs.
42:18 To date we've given about 100 grants
42:20 totaling nearly 20 million to about 100
42:24 organizations.
42:25 Facilitating with those about 80 partnerships with MNO's and I
42:29 think the kind of headline numbers are within 2 years,
42:32 those organizations went on to raise over half a billion,
42:35 something like 530 million
42:37 and impact 36 million people.
42:39 Alongside that we do research and insights and uh convening uh.
42:44 For the industry.
42:46 So
42:47 that's the first bit down.
42:48 Number 2 is the frameworks we use to think about uh partnerships.
42:53 And this is the first one which is also,
42:55 it's a process but it's also the different stakeholders
42:58 and it's also the technical aspects that come in.
43:00 So starting with the MNO data,
43:03 it distinguishing between the different types of data that come
43:05 through and being clear about which you need and then
43:08 the process that goes into making those useful within the warehouse.
43:12 Secondly,
43:13 where it comes in
43:14 the third party data and how we've talked
43:16 about combining those and the specific analytics you need
43:19 to get those useful
43:20 uh packages out for the decision makers.
43:23 And of course these can be different people across different stages.
43:26 Uh,
43:27 so the journey map for partnership describes
43:29 those elements that are particularly important,
43:31 and I think
43:32 a lot of people will be familiar here with the partnership establishment
43:36 of how you get to that first stage of getting the data
43:39 and then similarly,
43:40 the issues with the data management and the integration.
43:43 But I think what
43:44 maybe the research community doesn't think so much about is then
43:48 the sustainability of these,
43:49 the financial sustainability.
43:51 And then also ultimately
43:53 the impact and usage and how it all comes and
43:56 the fact is you can move between these different elements or
43:58 look at them in a different way as different starting points.
44:02 Um,
44:03 I'll skip part this this very quickly,
44:05 but we use the haves and needs framework quite a lot in partnership establishment
44:09 to really think about where the value's coming from on both sides and to be
44:13 clear
44:14 about that from the start.
44:18 That's frameworks down,
44:19 23,
44:20 COVID response is uh led by our AI for
44:24 Impact initiative which sits outside of Mobile for Development.
44:27 But this is a task force of 21 mobile operators covering 2
44:32 billion people
44:33 um
44:35 that provide that has an advisory panel and and
44:37 works to develop specific initiatives around mobile big data
44:41 and AI,
44:43 um.
44:44 We've seen this
44:45 already with fair to slide,
44:47 and as mentioned,
44:47 I mean
44:48 the big spaces of white where not a lot of work went on.
44:52 Uh.
44:54 So the GSMA worked
44:55 with support from the UK Foreign and Commonwealth and Development Office FCDO
45:00 uh in about 14 countries,
45:03 providing there some advice,
45:05 some technical advice,
45:06 some consulting on
45:08 ethics and and business practice,
45:09 but
45:10 in 4 countries,
45:11 Benin,
45:12 Burkina Faso,
45:13 DRC and Rwanda,
45:14 we worked more intensively
45:16 to facilitate some public-private partnerships and,
45:19 and they're what I'll talk about in a minute.
45:21 Um,
45:22 so I think this is also very familiar to,
45:25 to what we've been talking about today,
45:27 but going back to that first slide and thinking about.
45:31 You know,
45:31 during the process what could have been uh informing these partnerships,
45:35 where different data and particularly from different agencies needed to come in
45:39 to produce different sorts of analytics for different sorts of decisions.
45:43 And this was really
45:44 uh some of the thinking that was
45:48 done at the very start of the process to
45:49 determine what happened in each of those four countries.
45:53 Uh,
45:53 and it's slightly out of place but that's the output for
45:56 DRC in terms of the dashboard on kind of mobility.
46:00 Um,
46:00 as well as,
46:02 um,
46:02 some other analytics.
46:04 Uh,
46:05 so this is revisiting part of that but just to kind of package,
46:09 uh,
46:09 exactly the pro the process that the AI for
46:12 Impact team went through in those four countries.
46:16 Um,
46:16 I'm gonna skip past it to,
46:18 to some of what came out.
46:20 Oh,
46:20 that hasn't scaled very well.
46:22 Um,
46:23 but,
46:24 uh,
46:24 I think,
46:25 you know.
46:26 It was,
46:27 there's 6 things that that we point to that we were proud of within this work.
46:31 I think the first is that you know it was a
46:32 successful use case development for mobile big data products and to
46:37 develop and deliver those in 4 countries
46:40 producing data for decisions in a very short time frame
46:43 uh was an enormous achievement.
46:45 Uh,
46:45 it increased awareness among some of those key stakeholders.
46:48 Many of them were learning that interacting with
46:50 these products or concepts for the first time.
46:52 It promoted a state of privacy and governance,
46:55 so kickstarting a lot of discussions that
46:58 that
46:59 you know perhaps hadn't been considered before,
47:00 especially through the the government agencies that were worked.
47:04 Uh,
47:04 there was a lot of kind of enhancing of technical skills of those key,
47:07 key actors,
47:08 um,
47:09 hopefully laying some of the foundations for,
47:12 for the ecosystem growth,
47:13 uh,
47:14 and,
47:14 and then also our role in,
47:16 in efforts to increase knowledge sharing around this.
47:20 Um
47:22 This maybe you can revisit the slides in your own time cos I'm I'm running out of time.
47:26 er but this this draws on some of the other work
47:29 that we've done around innovative data for urban planning and points to
47:33 you know the checklist of things that that really
47:35 should be considered in in partnership formation and also
47:38 could present stumbling blocks at different times.
47:41 Uh,
47:42 the penultimate slide is just some of the resources that you may
47:46 want to visit and it's it's kind of more in there for.
47:48 Uh,
47:49 revisiting at a later date,
47:50 but to,
47:50 to flag some of the key resources on ethics and and principles for using,
47:55 uh,
47:56 big data AI.
47:57 Importantly,
47:58 and something that hasn't been talked about a lot
47:59 apart from in this lightning talk session is,
48:02 is the business models that go with that.
48:04 Uh,
48:05 especially if you want to,
48:06 to have sustained use and you know where third parties
48:09 come in to apply these analytics to to produce useful products and services.
48:13 Um,
48:14 and
48:15 then there's also some of the utility specific resources there,
48:19 and I think with that it
48:22 ticks off the final point for the day and thank you very much.
48:32 Um,
48:33 all right,
48:33 well,
48:34 thank you so much.
48:35 Um,
48:36 I have a couple of closing
48:38 remarks.
48:39 Um,
48:40 so first of all,
48:41 I wanted to thank you all for being here and it's so,
48:45 um,
48:45 exciting and it's so valuable to be here in person
48:49 and it has been like truly a different
48:52 experience for me because I've only been to maybe
48:54 two conferences in person.
48:57 And the amount of that I've gotten out of today is just
49:00 astronomically different than what I would have gotten out of it,
49:03 you know,
49:04 virtually or in another situation.
49:05 So I hope that that's the case for you and um,
49:09 yeah,
49:09 thank you so much for,
49:10 for,
49:10 for coming here to the World Bank.
49:12 Um,
49:13 another benefit of in person is that we all get to like chat,
49:16 uh,
49:17 and not in a breakout group.
49:18 So,
49:19 um,
49:19 I hope that any questions that you have for
49:22 some of the speakers for the lightning talks,
49:23 for example,
49:24 you're able to ask them,
49:26 uh,
49:26 during the reception.
49:28 Um,
49:29 I just wanna make a note for Berber.
49:31 There was a specific question for you.
49:33 Oh,
49:33 addressed,
49:34 perfect.
49:34 OK,
49:35 great,
49:35 um.
49:37 Next,
49:37 I really want to thank,
49:38 uh,
49:39 some of the co-organizers of this event.
49:41 So some of them are standing right here,
49:43 um,
49:43 they've been
49:44 wanna come out and
49:45 and give a little wave or something,
49:47 um,
49:48 so these are,
49:48 uh,
49:49 the people who helped organize this event and have kept it running,
49:52 and it is truly like,
49:53 uh,
49:54 a large effort on the back end.
49:55 So let's give a round of applause for.
- add-style
- lp-body-content