AI is rapidly changing the way digital teams work, but knowing how to use it effectively is just as important as deciding when to use it. During a recent discovery for the Nature Restoration Fund, our team set out to become “AI-enabled” from day one. Along the way, we discovered that AI can significantly speed up research, analysis and collaboration, but only when combined with human judgement.
In this blog, I share how we used AI throughout our discovery, what worked well, what didn’t, and the lessons we learned about making AI a practical tool rather than a replacement for expertise.
Background to the Nature Restoration Fund
The Nature Restoration Fund is central to the government’s new planning approach, shifting from site-specific mitigation to landscape-scale ecosystem restoration. Developers can pay into the NRF through a levy, which funds Natural England’s Environmental Delivery Plans. These plans set out how conservation measures will restore nature at scale, with NRF levy funding enabling Natural England to deliver them.
Why I’m writing this blog
I could have used AI to write this blog, but sometimes it is nice to just let the words tumble out of your brain and onto the screen via the keyboard. That’s not to say AI doesn’t have a place in helping to create content. More so, I’ve learned over the past few months when to use AI, and when not to.
How it all began
“We’re an AI-enabled team.”
That was one of the first things we were told as we started to work together on the discovery.
We started off as a small team, myself as Lead Product, Zoe the Delivery Lead, Ali and Paulo as Technical Leads, and finally Paul, the all-singing, all-dancing UCD Lead. Except Paul definitely does not sing nor dance, he does write incredible user research prompts though.
Soon five became four, as we realised one Technical Lead was enough. Supported by the AI Enablement Team, we met weekly to share how we’d been using AI, what we’d learned and where we needed help.
What did we learn?
We discovered there are two main ways to use AI, depending on what you’re trying to achieve.
Using AI to gather information
If you’re looking for AI to answer more advanced questions than you might ask Google, you can use it to provide information and it will return a comprehensive set of results.
For example, I wanted to understand the stages of the wider planning application journey so I could map our new service against the existing process for developers and local planning authorities.
A simple prompt asking Copilot to provide the stages and sub-stages of the Town and Country Planning Act process gave me a list that I could use in Mural to map out the journey.
Being diligent, I checked the sources it referenced and was satisfied that the information was accurate.
Using AI to analyse research
The second approach was to provide detailed prompts using a structured format:
- Context
- Analysis
- Implementation
- Verification
This framework was provided by the AI Enablement Team. Over time, the team learned how to refine the prompts so the results became significantly better.
One prompt was used to analyse our user research findings. Once the interview transcripts had been anonymised, the prompt generated an initial set of findings.
It could have been left there, but we wanted to validate the output. Further prompts were used to analyse the findings and test whether they accurately reflected the research.
We discovered that one finding we thought was consistent across several users had actually only been mentioned by one person. Left unchecked, this could have introduced bias into our conclusions.
By manually reviewing the transcripts—and because the person carrying out the analysis had also conducted the interviews—we were able to understand the context and sentiment behind the responses in a way that AI is not yet able to replicate.
Sharing our work
We presented our ways of working with AI at a co-working day.
To generate the content, we asked Copilot to use the acronym DISCO and map the GDS Service Standard against it.
It suggested:
- D – Define the problem
- I – Investigate the users
- S – Study the landscape
- C – Collaborate and co-create
- O – Outline the opportunities
Using this as the basis for our session, we mapped our prompts and learnings onto a Mural board so others could reuse them when carrying out their own AI-supported discoveries.
And no, Paul did not dance at that event either.
What we learned from this
AI helped the team move faster, but it never replaced human judgement.
The greatest value came when AI-generated insights, supporting evidence and human validation worked together. Rather than replacing expertise, AI strengthened our ability to test ideas, challenge assumptions and make more confident decisions.
That’s probably the biggest lesson from our discovery: AI is at its most powerful when it’s treated as a co-pilot rather than a replacement for experience.
