---
id: "workshop:data-series-data-collaboration-and-governance"
type: "InnovateUS workshop"
title: "Data Series: Data Collaboration and Governance"
url: "https://innovate-us.org/data-series-data-collaboration-and-governance"
date: "2023-11-06"
duration: "60 minutes"
language: "English"
instructors: ["Stefaan Verhulst"]
recording: "https://www.youtube.com/watch?v=N7yAwnMp7Rg"
transcript: "automatic, from the recording"
source: "InnovateUS (innovate-us.org)"
licence: "CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/), as stated on innovate-us.org"
---

# Data Series: Data Collaboration and Governance

InnovateUS workshop · 6 November 2023 · 60 minutes

In this 60-minute virtual training on data collaboration, participants will learn about different models of data collaboratives, including data pooling, corporate data philanthropy, challenges, and intermediaries, and will understand the advantages and challenges of the different models, with accompanying case studies to highlight each model. Participants will learn about the importance of governance in data collaboration, how to craft data-sharing agreements, and how to address privacy, security, and ethical concerns around data collaboration. The workshop will also provide an overview of the technical requirements and infrastructure for effective data collaboration, including access, transfer, and storage, and introduce tools, platforms, and solutions for facilitating responsible data collaboration.

## Learning goals

- Learn about different models of data collaboratives including data pooling, corporate data philanthropy, challenges, and intermediaries.
- Understand the advantages and challenges of different models with accompanying case studies
- Learn about governance in data collaboration
- Learn how to craft data-sharing agreements
- Learn how to address privacy, security, and ethical concerns around data collaboration.
- Get insight into the technical requirements and infrastructure for effective data collaboration

## Speakers

- **Stefaan Verhulst**, Co-Founder of The GovLab, Co-Founder and Principal Scientific Advisor of The DataTank

Recording: https://www.youtube.com/watch?v=N7yAwnMp7Rg

## Transcript

Automatic transcript of the recording: names and terms can be misheard, and the times can be a few seconds off. Each time links to that moment of the video.

**Unknown speaker**

[0:01](https://youtu.be/N7yAwnMp7Rg?t=1) And spotlight for you, and you can go ahead.

**Stefaan Verhulst**

[0:07](https://youtu.be/N7yAwnMp7Rg?t=7) Great. Thanks so much, Hannah, for this kind introduction and for organizing this series. And again, thanks everyone for joining. This is indeed the third and last session focused on data and how can we use data to improve the way we make decisions, the way we go about governing. And today I will specifically focus on data collaboration. Now, those that have joined us earlier already know what the GovLab is. It's an action research center focused on changing the way we go about making decisions, leveraging new technologies. And specifically, one of the areas that we have focused on is how can we leverage data to change the way we govern, change the way we solve public problems, change the way we design public services. And of course, in order to become data driven, one needs to have data, which is why we also have started working on unlocking data.

[1:10](https://youtu.be/N7yAwnMp7Rg?t=70) Because without having access to data, it's very hard to become data driven. And it turns out that data quite often that you need doesn't reside within your organization. And so what we have been focusing on and what I will be focusing on especially today is how can you actually have access to data that might reside outside of your organization or might not be easily available if you don't have a way to go about it. And this is a particular work that we are doing also within the context of our open data policy lab as well. Now, as I said, we start from the premise that data can help innovate how we solve public problems. But for doing so, we really need to become smarter. And the first session, I really focused on setting the stage. What do we mean by data? What do we mean by datafication? But more importantly, I hope that the key element of the first session was that the real value of data, given its unique characteristic that it's intangible and also non-rivalrous, giving those two specific characteristics, the unique potential and opportunity of data is to reuse data.

[2:23](https://youtu.be/N7yAwnMp7Rg?t=143) Meaning reusing data that was collected for one purpose, for another purpose. And that that requires a distinctive approach, a distinctive kind of new sophistication in order to really start thinking about unlocking data for reuse. And the second session was that if you want to start unlock data or use data for decision making, don't start from the data, but start from the problem. And more specifically, start from the question. And it turns out that we are actually quite often challenged in how we go about defining and identifying the questions that matter. And so last week, I spent about 60 minutes trying to go through this graphic by really emphasizing that we need to define the problem before we start leveraging data. But more importantly, that we need a new science of questions that can complement the data science environment as well.

[3:25](https://youtu.be/N7yAwnMp7Rg?t=205) And so that was last week. What I would like to start today is with the asymmetry that we already addressed in the first week. By which I mean is that on the one hand, we have a massive increase of the supply side of data. And as we've seen, there's a whole range of new kinds of data sources that have emerged just as a result of digitalization and advances in computation. But unfortunately, that the supply is quite often not matched with the demand. And so as a result, we have seen that we have a whole set of asymmetries, divides between the demand and the supply. And so one of the key challenges in the current environment, if we want to start leveraging data for public policymaking or for governance, one of the key challenges is then how do we overcome those asymmetries that we have in society where the supply of data is hard to get access to in order to meet the demand, i.e. the demand that can help solve public problems as well.

[4:34](https://youtu.be/N7yAwnMp7Rg?t=274) And so that's where we're going to start off. And so for the last few years, we have tried to really become smarter about how do we deal with those asymmetries, especially asymmetries between the private sector and the public sector. And how do we go about matching demand and supply? And so the first premise that I will start with is that we have tried to understand to what extent can we actually establish new public-private partnerships around data, which I have called data collaboratives. And so data collaboratives can actually help bridge data asymmetries. And that has been a project for the last five years at the GovLab. If you go to datacollaboratives.org, there you will find about 200, if not more, examples of how the public sector and the private sector have come together in order to establish this kind of new kind of public-private partnership, which we call data collaboration, around data, where you bring data to the table, but quite often also expertise, and in some cases also lived experience.

[5:43](https://youtu.be/N7yAwnMp7Rg?t=343) So it's quite often a public-private people partnership that really enables access to data for public interest purposes. And so what I will reflect on during the remaining kind of 40 minutes, or I will try to do less so that we have Q&A as well, is what have we learned by, on the one hand, studying this emerging phenomenon of data collaboratives, but also what have we learned by actually establishing a few of those data collaboratives? Because as an action research center, we not only have examined and developed a repository of examples so that we could study them, but we also have actually actively engaged with a whole range of sectors, including the private sector, but also the public sector at the national, at the subnational, at even the international level, on establishing a set of data collaboratives. And the first element that we've learned is that there is no one way to do this.

[6:45](https://youtu.be/N7yAwnMp7Rg?t=405) We really have to start thinking about what is fit for purpose as it relates to data collaboratives. And some of you might have heard that if you want to establish a public-private partnership, you should start thinking about a data trust. Others have started suggesting, no, it should be a data commons or a data cooperative. Well, it turns out that all of these might be feasible and appropriate, but they might not be aligned with the purpose. And so we really have to start thinking about what is the purpose and how do we go about understanding the purpose. And quite often we really have to think about, given the purpose, for instance, we want to have access to personally identifiable information that is anonymized in order to understand advances in particular kind of public health interventions. Well, for that purpose, you may want to have to start thinking about, well, how open can the data be or how open should the data be versus how closed, right?

[7:48](https://youtu.be/N7yAwnMp7Rg?t=468) So we need to start thinking about the spectrum of openness to determine what kind of data collaborative we need. But we also have to start thinking about what kind of involvement do you want to have with regard to the data holder versus the data user? Do you want to have the data holder to actually actively co-design the project? Or do you just want to have the data holder provide access to the data and the data user develops the data collaborative as such? And so we typically use this kind of matrix in order to determine what is fit for purpose by first looking at the spectrum of openness and then the spectrum of conditionality and the spectrum of engagement so that we have this matrix of conditionality. Because the lesson learned is that data quite often is hardly ever made accessible without conditions. And those conditions can be more flexible, more relaxed versus more stringent. And of course, some of those conditions might also relate to the level of engagement that you want to have.

[8:53](https://youtu.be/N7yAwnMp7Rg?t=533) Now, in addition to that, there are other criteria that you can use in order to determine what is fit for purpose. I already mentioned data accessibility, but it could also be data attributes. Do we want to have raw versus pre-processed data? Or is insights good enough? Is it one data provider or is it multiple data providers that need to provide access to data? Is it one data set or is it multiple data sets? How do we go about collaboration dynamics? How do we go about scope and so on? And so by combining all of those, you can already see that there are multiple versions of data collaboration feasible. And as I said, there is no one perfect unless, of course, it is fit for purpose. And I think that's what one really has to start figuring out. Now, having said that, despite the fact that you have all those possible variations, what we have identified is that there are actually a set of models emerging. And what we have tried to do is to start thinking about, well, what are some kind of models that one can start thinking about when developing a data collaborative without automatically going through all those possible variables that might determine a data collaborative?

[10:09](https://youtu.be/N7yAwnMp7Rg?t=609) And we have identified six kind of models. And, of course, anyway, if you start thinking about it, some of those models might be hybrids. They could combine some of those models. Some of them might be more extreme. Some of them might be one-offs to a large extent. But we have identified kind of six models that one can start thinking about when considering establishing a data collaborative. And for those that are interested in many types of those different models, you can actually have a look at the report that we published kind of two years ago or so, where we identified those six models, but then within those six models also variations of those models as well. I'm going to go quickly through those models just to give you a flavor of what they are. Again, I have a lot to cover today, so I'm not going to go into detail for each example. I will share those slides as always. And also, if you have questions, we can pick those up in the Q&A.

[11:12](https://youtu.be/N7yAwnMp7Rg?t=672) But the first kind of model is kind of the public interface model, where it's kind of fully open and also more relaxed with regard to the type of collaboration. And so one example is, for instance, Google Earth outreach, where you have access to the data as an interface, a public interface. And so the context here is that it's one data holder. It's Google and Google Earth. And then you have many data users that actually can have access to the data set. And here we have typically two types. One is kind of using an API. The other one is kind of a data platform. Again, I'm not going to go through here. I will share these in slide. But suffice to say is that here, of course, the processing and the sharing is where Google Earth provides access to a public data catalog.

[12:14](https://youtu.be/N7yAwnMp7Rg?t=734) And that's kind of one type of data collaborative. And, of course, anyway, there is still a certain level of conditionality on how you actually use Google Earth determined by the API. And you also have data platforms. Another example here is Uber movement that provides access to Uber data for those that are registered. And that other example of a public interface where there is one data provider, multiple possible users, the flow using the API or the data platform. The other example, the second kind of model is the data pooling model where there are some data holders, more than one quite often. And there are some data users also more than one quite often. And here we have a variety of data pooling examples.

[13:15](https://youtu.be/N7yAwnMp7Rg?t=795) One is, for instance, the public or the private data pool or the private data pool is, for instance, a California data collaborative where in California the water utilities have decided to develop a data pool to access water data that they share within a private data pool, which means it's kind of a club good model. Only those that are a member of the club have access to the data. And, of course, that's based upon a membership, in this case, all the water utilities as well. Now, the great advance of this kind of California or this kind of private data pool, the great advances, of course, is that it's kind of a reciprocity. You share something and you get something back, in this case, other data. Example of public data pool is, for instance, Global Fishing Watch, where different types of data is being provided and actually everyone has access to this kind of pool as well in order to understand to what extent this is, for instance, illegal fishing.

[14:18](https://youtu.be/N7yAwnMp7Rg?t=858) And quite often this is done in order to facilitate actually crowdsourcing of phenomenon, such as illegal fishing or where ships might go into territories that they are not, that way they have no authority or no license for as well. The third type of data collaborative is actually one that is, again, more based upon kind of crowdsourcing, where a set of one or more data providers actually provide access to a data set, one or more, through prizes and challenges. So where they say, well, look, you can do this. have access to the data. And then anyway, you can come up with an example on how you can leverage the data for a particular kind of purpose. And then we will give you a price in some cases, that might mean that you actually have access to the full data set as a price. And so here, we have, for instance, selective innovation challenges. This was with Turkey, for instance, where the telecom organization or the data for refugees challenge, where the Turkish telecom operator provided limited access to data sets for kind of registered researchers. And then they could make access and leverage that data in order to show what models they could come up with in order to understand the flows of refugees using telco data as well. And these are, for instance, open innovation challenges. In this case, there was a whole set of competitions, for instance, that are being provided that driven data as well. The fourth kind of data collaborative model is where you actually have a trusted intermediary, right? So in order to match demand and supply, you actually have an intermediary in the middle that tries to figure out how to do so in the best

[16:18](https://youtu.be/N7yAwnMp7Rg?t=978) way. So typically, that intermediary has some kind of authority by the data provider in order to figure out who can actually access to data as well. And so here, you can have data brokerage, or you can have third party analytics. Data brokerage is, for instance, the example here where POPSTAT, which is an initiative where also the Columbia Climate School and ESRI, for instance, are a part of. And so POPSTAT is basically deciding who has access to the data that is being provided by ESRI, by Facebook and others, in order to understand population movements, for instance, and the impact on climate as well. And then, for instance, we have third party analytics, where an intermediary actually is also the person or the entity that does the analytics. And in this case, for instance, this is an example where Dahlberg, which is a consultancy firm, was the intermediary to have access to certain kinds of data as well, but then also works with a public sector in order to analyse their questions. And then the last kind of data collaborative is kind of at the other spectrum. Remember, we started off with public interfaces, where anyway, data was almost provided via an API, or kind of a public interface for having access to the data.

[17:51](https://youtu.be/N7yAwnMp7Rg?t=1071) Now, the other extreme is, of course, where no data is shared, but insights are shared, intelligence is shared. And this is kind of the last kind of model that we see, actually, quite often increasing in popularity, because it's for the data holder, it's quite often safer to share insights from the data than the data itself. And one can argue whether that's a fully data collaborative, but obviously, they still share something from the data and towards that extent, I still feel like that might actually, for instance, cities and states might actually be good enough to actually have access to the insights from the data, not just the data as well. And here an example, the Spanish-backed BBVA, for instance, developed an urban discovery, where they provide access to insights that they have, as it relates to urban opportunities and urban well-being. And there's another example, of course, here in the US, which is the JPMorgan Chase Institute, which is a really interesting institute, it's kind of perceived as kind of a new kind of think tank, but a think tank that leverages only JPMorgan Chase data. And as you know, JPMorgan Chase is the largest bank in the US. And so they have massive amount of data. And so what they do is to actually leverage that data in order to provide economic statistics or insights, for instance, on what's the impact of COVID-19 on economic security. And this has proven to be extremely valuable for policymakers, they have no access to the data, but they have access to the insights as such. And this is also a kind of a data collaborative, because it's actually collaboration around the questions that policymakers care about, as such. So these, and then of course, sorry, I've got one research and analysis partnerships, which is kind of the traditional kind of data

[19:53](https://youtu.be/N7yAwnMp7Rg?t=1193) collaborative that most of you probably already know, where actually you work with researchers, typically that could be due to data transfer, for instance, the food, UN Food Agency, has, for instance, had access to data from Telefonica, in order to understand, for instance, food security, in a particular reason, as it was during a drought, or in some cases, you can have data fellowships, where researchers have access to the data. In this case, for instance, google.org has a data fellowship where they have access to Google data, and other data sets during a limited period of time. And so this is kind of a research partnership. And we see a lot of those happening as well, and could be valuable for also public officials to actually work with those researchers, public officials might not have access to it. But the researchers might have access to it and could address the questions that public officials are interested in, as well. So these are kind of the spectrum of data collaboratives. And, and as I said, there is no one way, right. And there are multiple hybrids of this, and it goes from fully open to fully closed and only insights to more collaborative, where you actually have multiple actors collaborating to ultimately a one on one kind of customized collaboration. And all of these are experiments in order to match demand and supply, especially matching private sector, as you've seen private sector data to inform public interest objectives. Now, having done this work, and especially having studied what is happening in the space for the last few years, one of the real challenges, however, of data collaboratives, is that it's actually very hard to make it systematic, sustainable and responsible. So many of those examples that I've showed you

[21:56](https://youtu.be/N7yAwnMp7Rg?t=1316) are examples that are actually kind of the exemption quite often, right, because what we see is a lot of those data collaboratives get started, but they never go beyond a pilot phase, they never become systematic. And they are quite often very hard to sustain, because no funding for it, especially if this is a public good kind of initiative, very hard to get funding to sustain data collaboratives. And of course, you also need to make it responsible. And so for the last, again, few years, we really not only have worked on actually understanding what types of data collaboratives exist and how and what is fit for purpose, but also we've tried to understand how do we actually make data collaboratives more systematic, sustainable and responsible. And so what I will share is a few lessons. And again, this would almost take like several days to actually really unpack all of this. But I will in the 60 minutes that we have today, I will just feature a few foundational areas that I feel could make a difference if you focus on those as well. And the first one is that it's really about determining the value proposition. And by this, I mean, especially if you work with the private sector, it's not good enough to really think about what's the societal value, if I would have access to the data, you can, it's very hard to make a data collaborative that involves the private sector, but also the nonprofit sector, sustainable and systematic, if you also don't figure out what is actually the business case for data we use, what's the business case for a data collaborative, in order to make their data accessible. And again, some of you might already have had that kind of reaction, like anyway, why would a private sector actor share their data?

[24:01](https://youtu.be/N7yAwnMp7Rg?t=1441) And that's what we have been trying to figure out is that what's not only the societal case, but also what's the business case. And what I will share with you is what we have identified by working with the private sector on what are some of the reasons and what are some of the incentives for the private sector to provide access to the data. And so if you develop a data collaborative, it's not only important to figure out what's the question, what's the data that we need? How do we match the question with the data and vice versa. But it's also super important to think about what's the value proposition we seek to establish, not only in terms of the value proposition as it relates to the public interest, but also the value proposition as it relates to the business case. And so we've come up with this kind of model, which we call the nine Rs model, because as you can see, it happens that all those incentives starts with an R. And, and I've done this mainly so that I can remember it. So that I need these days a mnemonic to actually remember all those things. But I hope it's also useful for you guys to have this kind of mnemonic to think about more than just one incentive. And so the first incentive, why corporations share the data is quite often for what I call outsourcing R&D in a very cheap way. Meaning if you have data, and you don't really know what could be the other value of my data, well, one way to figure this out is to actually provide access to your data. And so this is what we see is that a lot of corporations are actually interested in sharing the data as part of kind of an R&D strategy so that they provide access to, for instance, the public sector to then actually understand, well, what can my data do? And so that's kind of a first example here, which is we've worked with an

[26:02](https://youtu.be/N7yAwnMp7Rg?t=1562) organization called new lab, which is a an organization innovation lab based in New York, mainly focused on urban innovation. And one of their corporations is called that they are incubating is called numina. And numina is a startup that has sensors everywhere in New York City, especially in Brooklyn, but also in many other cities now. And they planted those sensors in our Internet of Things, in order to advance autonomous vehicles, so that autonomous vehicles would know what's happening in a particular kind of place and that they actually could drive autonomous autonomously. Now, it turns out that by providing access to the data, they started realizing that actually the data had a lot more value for other purposes, including for informing, for instance, Brooklyn City Council on a variety of urban mobility challenges, but also to inform, for instance, if we have construction in the street, what is then the impact of construction on food traffic?

[27:16](https://youtu.be/N7yAwnMp7Rg?t=1636) And what's then the impact of lower food traffic on small and medium sized enterprises. And so they started to realize that the data that they initially collected for one purpose, had a lot more value if they started sharing it with, for instance, public authorities. And so they now have developed kind of an additional kind of business model, in addition to providing access to still many facturers. They still now also on the one hand, public authorities, and some of this for freedom of it, they also have figured out, well, so that's the first reason. Second reason is reciprocity, by which I mean is then I get something back. And quite often, that's a real. Yes.

**Audience**

[28:12](https://youtu.be/N7yAwnMp7Rg?t=1692) Yes, you, you cut out for a few moments, when you were explaining the first reason, could you go back and do that again?

**Stefaan Verhulst**

[28:22](https://youtu.be/N7yAwnMp7Rg?t=1702) Right, and I will stop my video. So the first reason is accessing data, so that corporations can outsource research and development in kind of a easy, flexible and often cheap way. And so the example that I gave was to, for instance, example that we worked with in New York City, which is a company, a startup called numina, which is based at new lab, which is an innovation lab focused on urban innovation. And numina is a corporation that collects data using sensors across the city, in this case, across Brooklyn, in order to facilitate autonomous vehicles. Now, it turns out that the data can also be repurposed by the public authorities, in order to understand, for instance, what's the impact of construction on food traffic? And what's the impact then of lower food traffic on smaller, medium sized enterprises, for instance. And so this gives a clear example. Now, this was very important for numina to know that data had other use cases, which meant that they also now have kind of a new line of business to facilitate those that have an interest in, for instance, food traffic as well. And so they provide their data, on the one hand for free for, for instance, public uses, but also have now an additional kind of model that they can monetize, in addition to what they were initially thinking as well. And so research and insights, which quite often means developing new intellectual property rights, by actually sharing data is kind of a key business case.

[30:06](https://youtu.be/N7yAwnMp7Rg?t=1806) Very quickly, another one is reciprocity, where you get something, you give something, and you get something. And that quite often might mean I give access to my data, and then I have access to your data, which of course, anyway, for instance, in the context of a data pool, as I already explained, or a club model, which is not to be underestimated. And the World Bank has developed that. They have developed something called the Development Data Partnership, where I think about 30 plus companies are now sharing data with the World Bank. But what they get in return is they get access to expertise from the World Bank, and in some cases, also access to advanced insights that they otherwise would not have. And so this is kind of the reciprocity aspect as well. The other reason why sometimes it makes sense from a business perspective, to share data is to actually know, what's the quality of my data? Is my data actually any good? Because if you care about data, and using data for your own purpose, you actually do want to have, for instance, some extra eyes to know whether the quality of your data is up to standards. And so quite often, by providing access to the data, you can also then see whether there are some errors, and then that can be rectified. And then subsequently reused again.

[31:26](https://youtu.be/N7yAwnMp7Rg?t=1886) And so improving data quality is actually an important reason to share data. Now, obviously, this means that you have to be somewhat vulnerable in providing access to your data. But the reward is that by the end of the day, your data will be better. And we see this also, by the way, with open government data, is that by actually putting your data on a portal, you actually see what the value is of your data. And by the way, you can quite often get forced before you put it out to really start thinking about how do I actually clean it? How do I make it ready for reuse? Reproducibility, this is especially important in places where you want to make sure that, for instance, your analysis actually makes sense. And so, for instance, Microsoft is sharing data, they have this Microsoft Research Open Data Portal, where they share the data in order to actually make sure that any analysis that they have done can be reproduced as well, which is, of course, the fundamental standards of research. And so that's another reason quite often why it makes sense to share data. Now, reputation, obviously, is another purpose, another incentive. Brand equity is super important when you want to establish a business or when you want to convince consumers, or you want to convince, for instance, public authorities that they give you a license to operate. And so quite often we have seen, especially during COVID, for instance, that many corporations really not only, obviously, for social value, but also it strengthened their reputation. And like, for instance, cell phone companies were really important during COVID-19, and it also established a kind of credibility on how they manage and steward their data. Other one is corporate social responsibility, and I've worked on that more and more, is that quite often providing access to your data could become part of the ESGs or part

[33:27](https://youtu.be/N7yAwnMp7Rg?t=2007) of the corporate social responsibility. And in some cases, this could also be kind of the new philanthropy, right? And so Mastercard, for instance, in the early days, they started providing data philanthropy as kind of a new way of going about corporate social responsibility. And as I mentioned, I've written about this also extensively, that actually providing access to data through data collaboratives is actually the new corporate social responsibility of the data age as well. And then retaining and recruiting talent can also be a business case, which I mean is that quite often corporations do a lot of work and spend a lot of investment in attracting data talent. Now, if that data talent quite often only is used for, for instance, optimising advertisement, that talent quite often gets quickly burned out. And more importantly, it might leave the corporation to do more public interest, socially valuable work. And so by having data collaboratives, what we have seen is that it actually also incentivises data talent, data scientists to stay within a corporation. And so it is actually a human resources strategy to also establish collaboratives with, for instance, public interest organisations or the public sector as well. Here, an example of, of course, AI for Good, Data for Good by Microsoft.

[34:57](https://youtu.be/N7yAwnMp7Rg?t=2097) But also we've seen that, for instance, Facebook and Meta had a huge morale crisis and that they are now also doing a lot more work on actually public interest data analysis. And then, of course, regulatory compliance, that doesn't require any explanation that might, and we see this, for instance, more and more. Some of you might know that in Europe, they are about to agree on the Data Act. And the key element of the Data Act is going to be a mandate to facilitate business to government data sharing, which means that this will require, this will give government an opportunity to actually request data from businesses at times of extraordinary need. And so you see that this can actually be, anyway, a business case is to actually comply with those kinds of regulations as well. And as I mentioned, the Data Act is something to, to look for because they will establish kind of a precedent on how legislation can actually force data collaboration as well. And then last but not least, revenue generation. And again, typically when we start thinking about data collaboration, quite often people assume that all of this needs to be free.

[36:14](https://youtu.be/N7yAwnMp7Rg?t=2174) It doesn't have to be free. In some cases, actually cheaper to acquire and to work with data holders to have access to the data at a certain cost than actually doing the collection yourself. And so for that reason, it might actually be fair to pay a certain price. And of course, this could be elastic as well, but revenue generation is quite often another, of course, business case. And so it's interesting, for instance, if you look at Wikipedia, they have developed something called Wikipedia Enterprise, where they still provide access to Wikipedia data in general. But if you are a corporation and you want to scrape Wikipedia data for commercial gains or for training of your large language models, for instance, you have to pay. And so you see, this is, for instance, an example where you can actually have a tiered pricing free for everyone. And then, of course, a price if there is a commercial purpose as well. Now, what we typically do with corporations and with those that want to establish a data collaborative is actually go through each of those incentives and then really trying to figure out which of those makes more sense, which one has the stronger weight. And so this is, for instance, a spider chart that you can develop in order to really understand what is actually the business case for data collaboration to make it more sustainable. You also, and of course, have to really understand what are the costs. And I don't have time today to actually go through the accounting, but it's important to think about that data collaboratives do have a cost and you actually do need to understand what are the elements that are recurrent and what are the elements that are kind of one-off investment costs, and how do you then match that with the benefits, but more importantly, who is actually going to fund all of this so that you're not surprised

[38:16](https://youtu.be/N7yAwnMp7Rg?t=2296) once you launch a data collaborative that actually you need a budget moving forward as well. The second element, and I'm going to go quickly here, is that in order to make the data collaboratives not only systematic and sustainable, you also have to think about how do we make it responsible. And that means that you actually have to start thinking about the governance structure of a data collaborative. Too often we see too much attention to the technical architecture without actually paying a lot of time and energy on establishing a governance structure that provides for legitimacy as well. And so we have worked on establishing governance structures for data collaboratives in a variety of settings, and the lesson learned from that is that if you start thinking about a data collaborative, whether this is kind of a standalone initiative, and again even in some cases a new non-profit, or whether it's a project of a particular kind of organization, either way you need to start thinking about what are the kinds of what's the kind of governance that we need to have. And now when we look in governance there are typically four elements that you need to consider, and this can be used as kind of a checklist. The first element is that, and again this goes back to what I discussed last week, in order to justify a data collaborative and especially develop a collaborative governance structure that is legitimate, you really have to specify the purpose why do you have the collaborative in the first place, because without specifications it's going to be very hard to communicate and let alone to actually establish for instance bylaws. In addition to that you need a set of principles that will guide the decisions, both at the level of the organization, at the level of the data collaborative, how will you make decisions for instance, and who has access to it, but also at the level of the data. Then you need a set of processes in order to leverage the principles

[40:20](https://youtu.be/N7yAwnMp7Rg?t=2420) so that you can make decisions aligned with the principles to meet the purpose, and that requires quite often processes such as a board or a trustee group of trustees or a set of steering committee members for instance and so on, so that you actually have a way to make decisions with regard to a data collaborative that is aligned with the principles to meet the purpose. And of course in order to do so you need a set of practices and tools, and one tool which I will focus on specifically is of course the data sharing agreements that quite often you need to establish in order to determine what are the conditions under which one has access to data or to also de-risk the data as well. Now there are all kinds of purposes as I said, I'm not going to go through this, actually we discussed this at length last week when we were thinking about the problem definition and about the questions, but it's important when we start looking into the principles it's important to make a distinction between organizational principles and data principles, and organizational principles are principles that you will use to make decisions such as for instance transparency, such as for instance accountability, such as participation, representation, who will be represented when you make decisions about data, will the data subjects be represented for instance, will they have a voice, and you also need then a set of processes. Now in order to establish those processes it might also be important to think about who actually already makes decisions, because typically a data collaborative of course is not starting from scratch and so towards that end you may want to think about a decision provenance tool, i.e. who actually makes decisions already as it relates to collection, as it relates to processing, as it relates to analysis, so that you then have an understanding of existing

[42:22](https://youtu.be/N7yAwnMp7Rg?t=2542) processes and whether you need additional ones to actually complement those as it relates to data collaboratives. As I said sometimes you might also want to think about providing a stronger representation of the data subjects and establish a social license, because we're talking about reusing data that was collected for one purpose, especially if this is personally identified information, it means that you got content for that one purpose, but quite often it's very blurred when you actually got content for other purposes, and so that's where you need a social license, and that means that you need an additional way of engaging society, and so one way might be by establishing a citizens assembly, that's kind of a process as part of the governance, but then you need practices, and as I said one practice is establishing data sharing agreements, and for those who are interested here is our tool for establishing data sharing agreements, again I will share all of this afterwards as well, but basically what we've tried to do here is to establish the contractual wheel of data collaboration, by the way this week we will release an update of that contractual wheel, but suffice to say is that developing a data sharing agreement is not rocket science, it basically needs to think about why do you need to have the data, what data will be shared, who will have access to the data, and under what conditions, how will the data be shared, when will it be shared, will this be sunset at a certain point in time, or is this kind of accessible forever, and under what jurisdiction is this going to happen, and so once you have figured this out, you can actually develop a data sharing agreement in a pretty straightforward fashion as well. I'm going to skip the technical infrastructure because that would go in too much kind of detail, but the last thing I would say in order to make data collaboratives

[44:22](https://youtu.be/N7yAwnMp7Rg?t=2662) systematic, sustainable, and responsible, one also needs to start thinking about a new function, and that's where data stewards comes into play, because what we have realized is that you can have the best technological infrastructure, you can have the best kind of governance approach, you can have a data operational model that is fit for purpose, but if you don't have a human infrastructure that actually can facilitate data collaboration, it will be very hard to do so, and let me explain briefly by going through some kind of a use case where you seek to match the demand and supply. And so imagine the demand side is kind of an NGO, the supply side is kind of a private sector that needs to figure out how to match, and whether at all, to provide access to the data as well. And what I'm going to do here is to just show the complexity of actually figuring out the demand and supply matching ecosystem.

[45:24](https://youtu.be/N7yAwnMp7Rg?t=2724) Because the first thing one needs to do is, of course, the demand side needs to formulate its demand. You need to formulate a question. This is what we've learned last week. We need to try to understand what's the minimum viable data, also something that we discussed last week. But from the supply side, they really need to start thinking about, do we actually have the data? Do we have a data catalogue that I can use to understand whether we can even respond to that solicitation? And why should I do it? What's the value proposition? Other questions are, of course, is this actually legal? Are there confidentiality concerns? How should I prepare my data? Do I need to de-identify it? And similarly, from the demand side, we have to start thinking about what are the possible risks as well. Other questions are then questions with regard to the variables that I set out at the beginning of this lecture or seminar. How open should the data be? Should it be off-site, online? Is this about data that, how should I make it reusable and so on?

[46:32](https://youtu.be/N7yAwnMp7Rg?t=2792) So that's kind of the first set of questions that we have. And the second set of questions are then about the partnership itself, right? The first set of questions are about how do I actually respond with data? How do I do the data audit? The second type of questions are then linked to it. Okay, so we have data, but how do I structure the partnership? And how do I engage with the community? How do I know that the partner is trustworthy? Who else should be part of the data collaborative to make sure that I not only have access to data, but also can make sense of the data? So these are partnership questions. Then the third kind of questions are kind of questions related to internal coordination and staff engagement. Quite often, the biggest challenge of a data collaborative is actually getting internal sign off, both from the supply side, but also quite often from the demand side to even touch the data.

[47:34](https://youtu.be/N7yAwnMp7Rg?t=2854) And so a key function of data stewardship is actually internal coordination, but is also figuring out who else needs to be part of co-designing the data collaborative. Is it legal? Is it business? Is it IT that needs to be part of actually discussions? And so that's a key function that quite often is ignored. And that means that data collaboratives are not designed in a systematic way because you don't have the internal support. Then you need to start thinking about what we'll discuss and how we'll affect it. And then lastly, of course, it's not declaring victory when you have access to data or when your data collaborative is established. It's really about how do we actually make impact? Is anyone going to use it? And so what we see in today's society, which is why it's so hard to have data collaborations in a systematic, sustainable and responsible way, is that quite often those functions are actually not filled by a designated data stewardship function.

[48:40](https://youtu.be/N7yAwnMp7Rg?t=2920) And as a result, a lot of those data collaboratives fail or halt because we have not invested in the human infrastructure that is needed. And that's why we've been calling for data stewards. And so that's, from my point of view, a third key element of actually making data collaboratives more systematic is to invest in that kind of function. This could be a position, it could be a team, it doesn't have to be a position, but it has to be at a minimum kind of a designated authority to respond to solicitations or a designated authority or mandate to actually request as well. And again, if you're interested, we have a whole report about that as well. And as I mentioned, these are the five, which is just a reiteration of what I've mentioned. The first function is really about scoping and iterating, figuring out what is fit for purpose. Second is really about stewarding relationships, right? The first one is about stewarding data assets.

[49:43](https://youtu.be/N7yAwnMp7Rg?t=2983) Data stewardship is also stewarding relationships. It's also stewarding internal resources, expertise, and authorities. And it's also stewarding sustainability. How do I make the business case? What we discussed today as well, but also how do I get funding? And then lastly, of course, it's also about stewarding insights. How do I communicate what I actually have learned from the sharing of the data? And so to conclude, if you're interested, because this is, anyway, unfortunately only three sessions where I could introduce you to a few concepts and to a few lessons that we've learned. If you're interested, we have a much longer series that actually unpacks the five functions of data stewardship, which through our Data Stewards Academy, this is kind of an eight week series. And for those that want a shorter version and really delve into data stewardship, I also have a LinkedIn learning course, which is called Global Data Stewardship.

[50:46](https://youtu.be/N7yAwnMp7Rg?t=3046) You can send me a note afterwards or via Hannah perhaps, and I can provide you free access to that course if you send me a note, if you have a desire to take this 90 minutes course to go into detail about data stewardship. So to summarize today's session, data collaboratives is a way to actually bridge the symmetries that we have, where the demand and supply quite often are not matched. When you think about data collaboratives, don't think about just one model, but really start thinking about what is fit for purpose? What are the variables that will determine what data collaborative you actually need? Is it an intermediary or should you start with a prize and challenge? And then later, based upon what you've learned, move towards a data pool. There are different data collaborative models that are effective in different contexts. And I've given you kind of six kind of operational models that we have seen more pronounced. And then of course, it's very important to think about the business case, but also about the governance structures.

[51:52](https://youtu.be/N7yAwnMp7Rg?t=3112) And more importantly, about what is actually the human infrastructure that I need in order to make all of this work so that data collaboratives are becoming systematic, sustainable, and responsible as well. So I'm going to stop here, Hannah, and I know this was a whirlwind, but at least I wanted to make sure that the last session was worth everyone's time and that they got what they were hoping to at least get a sense on how to go about data collaboration as well. So eager to hear whether anyone has any questions.

**Unknown speaker**

[52:29](https://youtu.be/N7yAwnMp7Rg?t=3149) Thank you so much, Stéphane. And yes, we already have quite a few questions lined up. For those who are interested in the courses that Stéphane mentioned, I put my email in the chat. If you could just email me and let me know you're interested and I'll get the info from Stéphane. Alternatively, I will be sending you all an email after this with the slides and the recording. So you can just respond to that email without having to collect my email address right now. I've also put in the meeting chat a link to our one-minute feedback survey that we really hope you can take part in or fill out after this session. It's very, very brief, but it's important to us to hear how we're doing. I'm just going to come off of recording.

* * *

Original page: https://innovate-us.org/data-series-data-collaboration-and-governance · Licence: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/), as stated on innovate-us.org