Developing a use case to test whether a federated approach to digital collaboration could unlock the value of data in the charitable sector

In 2020, research and design consultancy Etic Lab undertook a discovery project to to test the appeal and potential of ‘data federations’ in the charitable sector. Etic Lab defines data federations as data sharing collaborations built on federated learning technologies, which enable organisations to create shared value from data without sharing or standardisation. It wanted to explore if technologies such as federated learning could enable organisations to attempt projects that would otherwise have been unachievable.

In 2020–21, we set up a stimulus fund as part of the fourth year of the ODI’s R&D programme – funded by Innovate UK – and our broader programme of work on data institutions. The fund aimed to help explore approaches that enable trustworthy and ethical data sharing to help citizens and businesses lower their impact on the environment, improve public services and save lives.

The stimulus fund provided us the space to experiment with a radical new approach to digital collaboration, whilst putting us in touch with a supportive community of like-minded colleagues

– Richard Woodall, Research Officer, Etic Lab

Facts and Figures

  • Etic Lab developed a test case that presents data federations as a viable approach for organisations seeking to establish a data access initiative or institution while managing sensitive data.
  • The test case shows that techniques like federated learning can provide a platform for collaborators to productively address issues such as data privacy and resource constraints. However, in order to capitalise on this potential, it is necessary to develop a collaborative structure in which all participants are able to fully engage with the design of the project.
  • Etic Lab outlined learnings and highlighted a range of challenges that must be addressed to successfully implement a data federation model in a real-world setting.
  • The research suggests that the data federation model can be a viable proposition for local government bodies, legal and financial services, unions and cooperatives seeking to unlock the value of data without needing to hold that data in a centralised location or collect it in a standard format.

What was the challenge?

Generating valuable insights based on sensitive data can be challenging. For many organisations who manage sensitive data, machine learning techniques and predictive modelling can be a great way to do this. However, these techniques require large amounts of data to provide their best return. This creates a barrier that is difficult to overcome for smaller organisations, who want to make use of the insights machine learning techniques can offer, but don’t have access to large datasets.

Privacy-preserving techniques, such as federated learning (which allows predictive models to be trained on separate datasets without any of the data being shared or exposed), could address this challenge. These technologies have the potential to open up ways to collaborate with others that manage similar data. Building coalitions based around the use of privacy-preserving technologies could enable smaller organisations to access the benefits of machine learning techniques without requiring access to large datasets. However, collaborating with others comes with its own challenges. Previous ODI research has shown that investment in artificial intelligence (AI) is characterised in part by a preference for business models that prioritise the collection and siloing of data, often to secure a competitive advantage when building, implementing, and operating AI systems.

In a digital economy, value is not derived primarily from the ownership of data assets, but the capacity to do things with them

– Richard Woodall, Research Officer, Etic Lab

The discovery project led by Etic Lab aimed to explore the social, regulatory and political dynamics of collaboration around privacy-preserving technologies in the charitable sector. Understanding the social and organisational challenges that prevent or enable successful implementation of such technologies could provide valuable insights that could be applicable across a range of use cases.

Technical fixes in themselves will achieve nothing unless we are able to develop forms of collective organisation necessary for their successful management and implementation

– Richard Woodall, Research Officer, Etic Lab

How is Etic Lab solving the problem?

Before the project began, Etic Lab had developed instances of federated digital technologies and a working theory as to how they might be used as the basis for a collaborative project. The purpose was to test this theory by exposing it to a group of real-world collaborators, with the aim of ideally allowing them to attempt a digital project which would otherwise have been beyond their means.

1. Finding a test case

Etic Lab engaged in sector-wide conversations among charities working to improve access to justice around how organisations might best share resources and knowledge, presenting the data federations model to a range of different audiences.

In the process, it learned the importance of long-term relationships and sector-specific knowledge for building consensus around digital projects, and the value of engaging key decision makers who can help raise awareness of and lend credibility to new ideas.

2. Building a test case – engagement

Etic Lab found two organisations committed to using shared data to develop insights which could be used to improve decision- and policy-making within their given sectors.

Both organisations were members of existing data collaborations. They were drawn to the data federation model as a way to create shared value without requiring their partners to fully commit to the process of building and implementing a universal data standard.

3. Design

Together, Etic Lab and its partners were able to scope a project which would use the possibilities afforded by privacy-preserving technologies to generate shared insights from distributed datasets. This established the feasibility of the data federations model in principle, and demonstrated its appeal to organisations in the process of attempting data access initiatives.

However, the proper social and organisational structures needed to support this kind of collaboration are still to be defined. In particular, the project demonstrated the importance of learning more about how to build consensus and buy-in for projects based on the use of novel technologies, as well as how to create an environment in which partners feel both comfortable and empowered enough to participate in the governance and design of such an endeavour.

What was the impact of taking this approach?

Etic Lab has opened up a conversation with organisations in the charitable sector about the benefits that privacy-preserving machine learning techniques like federated learning can offer to them. These benefits include lowering the up-front costs of collaboration while providing a more direct path to the generation of shared value, and providing a set of tools which allow collaborators to proactively engage with questions of data ethics, privacy and security. However, more work needs to be done to see this approach not simply as a means of mitigating barriers to collaboration, but as a way for organisations to expand their scope of possible action.

The issue of how to express your value proposition to non-specialist stakeholders is common to all data projects, and is perhaps even more pressing in the case of data federations. A successful data federation must be founded in a common purpose and shared goals. It follows from this that all partners must have a good understanding of the value that is to be created, even if this is not exactly the same for everybody

– Richard Woodall, Research Officer, Etic Lab

While this discovery project was based primarily in the charitable sector, the barriers Etic Lab sought to address are common to any circumstance where organisations with limited means attempt to work collaboratively with the data they manage. These barriers include:

  • A lack of time, resources and expertise to dedicate to digital projects
  • A resistance to adopting these systems due to the regulatory context and perceived risk associated with sharing personal data
  • An unwillingness to share commercially-sensitive information with competitors or regulatory bodies
  • A lack of experience in successfully delivering a digital project

This project has demonstrated the potential for data federations to help overcome these barriers, provided that the appropriate governance and organisational frameworks can be articulated. The ideas developed in this project will therefore be applicable to a range of other sectors, including local government, legal and financial services, unions and cooperatives.

Lessons learned

  1. The initial challenge faced by this project was to find a group of collaborators to work with – a process which was made slower by the Covid-19 pandemic. Etic Lab demonstrated that sector-specific knowledge, long-term relationships and a large pool of contacts are key assets for organisations interested in facilitating digital collaborations.
  2. Etic Lab learned that the foundation of a successful data federation will include the formation of shared values and goals. These cannot be assumed or taken for granted, but must be surfaced and explored as a crucial stage in the formation of a data collaboration.
  3. Perhaps the greatest inherent risk of the data federation approach is that it is framed as a solution to a set of narrowly-defined problems (such as privacy or non-standard data), when it requires a holistic methodology and has its own specific challenges and priorities that affect every aspect of a project’s design.

What next?

Etic Lab will continue to pursue the question of how to create collaborative structures that empower people to engage with the creative possibilities of privacy-preserving digital technologies. It plans to expand its research into the formal accounting of the governance and decision-making mechanisms available for data federations. One of its upcoming projects will be to develop an open standard to support organisations seeking to design and implement projects based around the use of federated learning and other similar techniques.

If you are interested in Etic Lab’s research, please read the report: ‘Data Federations: Digital Collaborations Without Data Sharing’ and listen to the ODI Fridays lunchtime lecture: ‘Data federations – collaborate without sharing’.

Find out more

If you are interested in the research or work ODI is doing into data institutions and data access initiatives please get in touch with a member of the team.

Other recipients of the data access stimulus fund award were Open Climate Fix, Your Dsposal, DNV, Collections Trust, Open Data Manchester and ODI Leeds. Find out more about the stimulus fund and what we learned here.