The brief
A complex civil dispute had produced extensive documents: correspondence, records and reports gathered over a long period. The legal team needed to find what mattered, see how the pieces fitted together and be sure that every point they relied on could be traced back to the document it came from.
Reading everything by hand would have taken a very long time. But this was also a setting where a confident-sounding summary that turned out to be wrong would be worse than no summary at all.
My process
Discovery. I began by understanding what the legal team actually needed to answer, and how they'd want to use the results, before touching any technology.
Plan. I agreed one firm rule from the start: nothing would be taken on trust. Every point the tool found had to point back to the exact document it came from, so a person could check it.
Design. I designed a way of working through the documents in which each one's source stays attached to it at every step, so nothing gets separated from where it came from.
Build. I used practical AI tools to help read and summarise the material, always alongside the original documents rather than in place of them.
Testing. I checked the references themselves, not just the summaries. If a point couldn't be traced back to a real document, it wasn't used.
Handover. The results were handed over in a form the legal team could check, question and build on, with the judgement left firmly with them.
What I delivered
- A practical, AI-assisted way of working through extensive documents
- Summaries that point back to their sources
- A checking step, so nothing relies on a summary alone
The outcome
The work did what it was built to do. It made extensive documents manageable while keeping every point traceable to its source, and it met the brief to everyone's satisfaction. Above all, it showed that AI can be a real help in serious, careful work, as long as a person checks every reference.
