Due-diligence risk analysis agent
Turning a folder of 200 documents into a risk register in minutes.
- Client
- A large property investment and development business
- Sector
- Property and development
- Status
- Delivered
Business problem
What problem are we actually solving?
Development managers assessing land for acquisition must surface every risk that could delay the deal, blow the budget or conflict with existing agreements: legal title, planning and permitting, utilities and infrastructure, ground and environmental conditions, and regulatory compliance. The information is spread across dozens of documents of varying formats, and gaps are as dangerous as findings.
Target users
Who is this for, and what does their day look like?
Development managers and acquisition teams within the development business.
Current process
How does it work today, and where does it hurt?
Manual read-through of the full due-diligence data room, with risks tracked in spreadsheets and email threads. Coverage depended on the individual's experience and the time available before the deal deadline. Missing documents were often only noticed late.
Desired outcomes
What does better look like?
- A structured risk table produced from the data room without manual reading.
- Explicit identification of anything that could delay acquisition, extend the programme or exceed budget.
- Conflicts flagged against existing deal agreements, financing terms, design standards and local requirements.
- Missing information called out explicitly, with a description of what still needs to be obtained.
Constraints
What limits, systems, data or rules apply?
- All analysis had to be grounded strictly in the documents provided, with no external assumptions or hallucinated findings.
- Documents lived in the organisation's existing document management system; the agent had to work with that repository as it was.
- Output had to be a consistent table format the team could drop straight into their review workflow.
- Governance: approved enterprise AI platform only, with no data leaving the organisation's environment.
Success metrics
How will we know it worked?
- Review time per site, before and after.
- Risk categories covered on every review, against the full taxonomy, where manual coverage varied by reviewer.
- Documentation gaps identified before the deal deadline rather than after it.
What was built
A document-grounded agent with a structured prompt defining the reviewer persona, the full risk taxonomy, the repository scope, and a strict rule to cite or flag as not found. Output is a fixed-schema risk table with a gap list.