This article is an interview between Tom Orrell and Andrea Ulrich, two Deputy Directors of Programs, and Mihai Postelnicu, Deputy Director of Software Development at data and digital organisation, Development Gateway.

  • The problem: Generative AI can be a great tool for organisations looking to improve their productivity, particularly when it comes to simplifying complex analytics tasks, but many don’t know how to use it in a safe and ethical way.
  • Why it matters: At Development Gateway (DG), the development and implementation of an AI policy was seen as crucial for navigating the complexities of generative AI, addressing ethical concerns and ensuring the responsible and effective use of these tools in enhancing organisational workflows.
  • The solution: DG management collaborated in the creation of an action-oriented and adaptable AI policy, supported by continuous communication, an internal committee and screening processes, to guide staff in safely and effectively utilising generative AI tools in their day-to-day work.

“Picture an AI that truly speaks your language – and not just your words and syntax. Imagine an AI that understands context, nuance and even humour.

_ _

_This is no longer just a futuristic concept – it’s the reality of ChatGPT.” _

– Bernard Marr, 2023, Forbes


Generative AI is having a moment. The functionality of tools such as Open AI’s ChatGPT exponentially improved over the course of 2023. As a result, generative AI’s applicability to complex analytical tasks has exploded. Generative AI tools are now so advanced that they can be used across business environments to support programme developers to improve their code lines and software, help analysts brainstorm ideas, support employee training and education and advise staff on the tone and language of draft reports and emails, among many other tasks. However, they remain far from perfect tools. Questions and uncertainties relating to data ethics, privacy and copyright are just some of the challenges that have to be navigated when using generative AI tools in the workplace.

Be prepared to update the policy document continuously to keep pace with developments in this space.

Within the sustainable development and public policy spaces, many organisations are considering using generative AI tools in their work. However, many struggle to discern what types of use cases are effective, appropriate and safe. Development Gateway, an IREX Venture, is an early adopter of generative AI technologies and is now safely using several different tools in its day-to-day work. A key enabler of this for Development Gateway has been the production of an internal ‘AI policy’.

In this interview with Andrea Ulrich, Deputy Director of Programs and Mihai Postelnicu, Deputy Director of Software Development at Development Gateway, Tom Orrell, Managing Director at DataReady, explores why and how Development Gateway produced its AI policy, how it is helping staff effectively and safely use generative AI tools and what advice they have for other organisations and institutions seeking to follow suit.

Tom: To start us off, please could you tell us about the work that Development Gateway does?

Andrea: Development Gateway: an IREX Venture is a data and digital organisation. We create digital tools and design processes that help collect, visualise and use data for more accountable, effective institutions. We work with ministries of health, education and agriculture in countries around the world to improve how they use data. We also support improved learning and data use for donor institutions like the UK’s Foreign, Commonwealth and Development Office, and Global Affairs Canada.

Tom: Given your work, why was it important for you to develop an AI policy?

Mihai: Development Gateway has long been involved in digital and data project implementation. We have strong technical development and programme teams. Given our interest in technology generally, when generative AI tools started to make a splash in 2022, our technical teams explored how they could be applied to our work. Within the tech team, we were lucky enough to have the skills and knowledge to be able to informally experiment with, and test, the applications and limitations of new generative AI tools. We discovered that there was substantial scope for their application in both our technical and programmatic work. However, we were aware that many colleagues across the organisation may struggle to understand how to best use the tools, what their limitations might be and what risks they might pose if improperly used. This is where the need for an AI policy first arose.

Andrea: I’d also add that over time we also became aware of a lot of broader challenges and risks associated with generative AI, for instance, the potential for biased, racist, or sexist outputs, and issues surrounding copyright over data used to train generative AI models. At the same time, staff members were already starting to use generative AI tools to brainstorm ideas and even do budget calculations. Sometimes, the tools were getting these tasks wrong. This combination of factors – knowledge of the external risks involved, and a pragmatic need to move quickly to support staff – really drove home the need for an AI policy at Development Gateway.

💬 Let the authors know what you think about the risks and opportunities of AI by leaving a comment below

Tom: What are some of the benefits and risks of using generative AI at Development Gateway?

Mihai: Generative AI tools are an incredible technology that has the potential to transform office productivity. If used properly and safely, they can turbocharge the production of any kind of document, the writing of source code for software and other things. Our developer teams have already experienced a boost in productivity by using the tools to edit code, learn new coding techniques and even learn new coding languages.

Andrea: On the Programs side, we have used generative AI tools to summarise large quantities of qualitative interview data – after having carefully removed any personal and sensitive data beforehand. I’m also aware of staff members for whom English is not their native language using generative AI tools to check the tone of emails to ensure they are clear and not inadvertently rude or offensive. As a global organisation, we work across countries in multiple languages, so a really helpful use case of generative AI for us has been in translation and editing.

Mihai: In terms of risks, one of the big risks that generative AI tools pose is that of ‘hallucinations’. Generative AI tools are creative by design, and this means that they sometimes tend to drift into more creative interpretations of a problem that you might expect. They can sometimes simply invent information to respond to a problem or misinterpret it. This means that – at least with the current generation of generative AI tools – it is really important to ensure that humans are kept in the loop at every stage of data input and output. Other risks we must be attuned to include ensuring that client data is not misused in any way and ensuring that we are aware of any biases that might arise as data is processed.

Tom: How did you develop your AI policy? What was the process?

Andrea: I’m a big fan of clarity and structure, especially when it comes to guidance for team members. I believe that once staff have clear guidelines and they know the limitations of the sandbox, then they can be free to be creative and experiment with new innovations within those boundaries. This thinking, on the one hand granting staff the space to experiment, while on the other ensuring that the necessary guardrails were in place, drove our AI policy production process.

We started the process when I invited the management team to join a task force to drive the development of the policy. Once a team of four had been established, which represented both the programmes and tech team, we then brainstormed what the steps of the process needed to be:

  • We first undertook a literature review of AI policies but found that there was very little already developed.
  • We then worked on compiling a database of all the generative AI tools already being used by staff within the organisation.
  • Once we had this information, we started to draft a policy structured around two components: i) guidance for staff on how to screen generative AI tools; and, ii) guidance on how to use approved AI tools.
  • Finally, we undertook two rounds of feedback on the draft; first with senior management, and then with eight colleagues from across the organisation. We then amended the policy based on the feedback.

The feedback rounds at the end of our drafting process were really important. By ensuring that staff across all levels of the organisation were able to feed into the process, we ensured that it really responded to their needs and was practically oriented.

Tom: Can you briefly summarise the policy?

Andrea: Our AI policy is very action-oriented and adaptable. It simply covers why the policy is needed, who it applies to, how to screen generative AI tools to ensure they are effective and safe and who to contact if further information or support is required. The policy is supplemented by an ever-growing database of use cases and audited tools.

Guidance on screening generative AI tools from Development Gateway’s AI policy
Before you commence using a new tool, please take the following steps:
  1. Determine if this tool has been approved by referencing the “Tools” section of this policy as well as the “Use Case” section. Is your use case approved? Is the tool you want to be approved in the spreadsheet?
  • If yes to both questions: Great! You’re done. Be sure to abide by the use case guidelines.
  • If the tool is approved, but not for how you plan to use the tool, go to step 2.
  • No, this is a new tool. Go to step 2.
  1. Define how you expect to use the tool (to the best of your knowledge)
  • What data or information will you feed into the tool?
  • Is this for internal use (DG eyes only) or external use?
  • Are there any privacy concerns or constraints related to the information you are inputting?
  1. Conduct an assessment of how the tool uses your data.
  • Check the terms and conditions. Do they cover if/how your data is shared with external parties?
  • Do they use your input data for training data? If so, is there a way to disable that feature?
  1. Based on the information from Steps 2 and 3, do you have any concerns with using the tool?
  • No. Great! Inform us and begin using the tool for your use case.
  • Yes. Do not use the tool! Please alert us so that we can further review it.
  • You’re not sure, or you need more advice. Please let us know so that a member of the committee can help you assess further.

I’d emphasise that one of our primary objectives in developing this policy was not just a policy but a process. This process enables constant reflection, discussion, screening of tools and experimentation. This is really important to us because the generative AI space is continuing to evolve so rapidly. To this end, we have two layers of support that surround the policy. We have an internal committee that reviews questions from staff and supports policy interpretation and use. We also have a live Slack channel, as well as a dedicated email address, that all staff can access and use to contact us with any queries they might have.

Mihai: In practice, the policy envisages a three-stage process in screening generative AI tools’ appropriateness for use within Development Gateway. First, staff can check the existing list of vetted tools to see whether the tool they want to use has already been approved or excluded. Second, if the tool they want to use has not been approved, they can screen it themselves, using the guidelines from the policy as a benchmark. Finally, they then notify the AI committee about the research they have undertaken and the outcome of the screening exercise.

Excerpt from DG’s AI Policy

Even for approved tools and use cases, you must abide by the following guidance:
1. You must follow the guidance laid out in the use cases and tools documents.
2. Be Transparent about AI Tool Usage Internally and Externally
  • If you are working on a project, you must ask the client for consent before using generative AI tools.
  • You must also disclose to your project team when you have used a generative AI tool for internal or client work.
3. Don’t input sensitive or client information
  • Do not enter any sensitive client or internal DG information or personal information from DG team members or stakeholders to projects (e.g. key informant interview names) into generative AI tools, unless there are data safety guidelines in place. Any violation of this process is considered a breach of DG’s data privacy policy.
4. Review is Required
  • If the client and team do agree to use generative AI tools, you must review all outputs from generative AI. Often the data produced is outdated or incorrect, so your thorough review is required. This is often called “human-in-the-loop” and is a mandatory expectation of all DG consultants and staff.

Tom: Finally, what advice would you give to other social purpose organisations and public institutions seeking to develop their own AI policies?

Andrea: My top three reflections are:

  • Make sure you include a wide range of individuals and functions in the development of the policy. Our policy was greatly improved by going through a feedback cycle with colleagues who were already using generative AI tools.
  • Don’t discount yourself from the process of developing an AI policy just because you are not sure about how generative AI works. There is a tendency for people who come across something technical to say, “this is something that should be dealt with by IT” but because of the nature of these tools, it's really important that a range of voices are included in the policy development process.
  • It is important to OVER-emphasise the importance of communication, in particular two-way communication that allows for internal discussions and reflections to drive organisation-wide learning.

Mihai: My top three reflections are:

  • Don’t wait to start a conversation about how these tools can be used in your work. Even if you don’t think they are relevant, it may be that some other coworkers are already using generative AI tools. It is very important to start the conversation as soon as possible and start a process to discern what appropriate and safe use of generative AI tools looks like in your context.
  • Be prepared to update the policy document continuously to keep pace with developments in this space.
  • There are benefits to being early adopters of generative AI tech. These tools can really work in your favour and give you and your organisation a competitive edge. They are relatively safe to use in a lot of contexts so long as guardrails are put in place and there is a clear process to screen tools and use cases.

More information on Development Gateway’s AI Policy can be found here: If you have any questions about Development Gateway’s work or its AI policy, please contact aulrich@developmentgateway.org and mpostelnicu@developmentgateway.org.

Done reading? Make sure to share your own thoughts on AI policy by leaving a comment below ⬇️

(Image credit: Pexels)