Due diligence is a business of thoroughness under deadline. Our client, a firm that investigates acquisition targets, counterparties, and key individuals on behalf of investors, faced the trade every research business faces: the deeper the investigation, the longer it takes, and the deadline does not move.
The problem
Each engagement meant analysts working through corporate registries, litigation records, news archives, sanctions lists, and the target's own footprint, then reconciling it all into a report a client would pay for. The work was expert but repetitive: the same sources, the same cross-checks, the same assembly, engagement after engagement. Senior analysts spent most of their week gathering rather than judging.
The firm did not want AI-generated conclusions. In this business, a wrong claim in a report is worse than a slow report. What it wanted was the gathering done faster, with every finding traceable back to its source.
What we built
We built a research pipeline of specialized agents that mirrors how the firm's own analysts work:
- Collection agents work the sources in parallel: registries, filings, court records, press coverage, and the target's public footprint, each agent scoped to the source type it knows.
- A cross-referencing agent reconciles what comes back, connecting entities across name variations, jurisdictions, and corporate structures, and flagging where two sources disagree rather than silently choosing one.
- A summarization agent assembles the findings into the firm's own report structure, with every statement carrying a citation to the underlying source, so an analyst can verify any line in one click.
The output is not a finished report. It is a fully sourced draft dossier that lands on an analyst's desk in a fraction of the time the gathering used to take. The judgment, the narrative, and the conclusions stay human, because that is what the firm's clients are paying for.
The outcome
Research that took weeks now takes days. The firm quotes shorter turnarounds without thinning its coverage, and analysts spend their time on the part of the work that justifies their rate: weighing evidence and standing behind conclusions.
Coverage got deeper as it got faster. Sources that were previously sampled under time pressure are now worked exhaustively, because the marginal cost of another source is minutes of agent time rather than hours of analyst time.
Why it worked
The line between machine work and human work was drawn where the client's liability lives. Agents gather and organize; analysts judge and sign. Every claim is traceable to a source, which made the system auditable enough for a firm whose product is trust.
The engagement started as a single scoped workstream on one report type, proved itself in production, and grew from there. That is the pattern we recommend to every research-heavy business: pick the report you produce most often, and take the gathering off your experts' desks first.