6-week due diligence in 8 days. 2.5× more deals delivered per year.
A document-to-board-pack pipeline for a UK mid-market M&A advisory. 2,400 documents per deal extracted, scored, and synthesised; analysts spend their time on judgement, not data entry.
DD throughput capped at one deal per analyst per six weeks — and the bottleneck was reading, not judgement.
Norton Burr is a UK corporate finance and M&A advisory firm working primarily on £5M–£75M mid-market transactions. The firm's DD process was thorough and respected — but slow. Each deal needed 5–6 weeks of analyst work before the first draft of the IM or board pack landed in a partner's lap.
Most of that 5–6 weeks was reading. A typical data room contained 1,500–3,500 documents — contracts, financial statements, employee records, IP filings, customer lists, supplier agreements, tax filings, regulatory correspondence. The analysts spent the first 3 weeks extracting structured information into Excel before any real analysis began.
Comparable-deal retrieval was harder than it should have been. The firm had completed 200+ engagements over a decade. Their precedents were sitting in shared drives organised by year, not by deal characteristic. When an analyst needed comparable transactions in the same sector with similar deal mechanics, the only way to find them was to ask the partner who remembered them.
The firm wanted to deliver more deals per partner. They didn't want to dilute the methodology — every DD process produced the same quality of analysis. They needed analysts to spend their time on judgement, not mechanical extraction.
A Claude-orchestrated pipeline that does the reading so analysts do the analysis.
Every document uploaded to the deal data room flows through a structured extraction step. Claude reads the document, identifies what type it is (MSA, employment contract, audited account, etc.), extracts the schedule the firm needs for that document type, and writes findings into the firm's house templates. Risk-bearing clauses are flagged with cited extracts; missing-document checks run automatically against the deal's expected dossier.
Financial models are auto-populated from extracted statements — historical P&L, working capital cycles, capex profiles, debt schedules. Analyst time on model setup dropped from days to hours. Partners review a populated model with anomalies pre-flagged, not a blank template.
The precedent retrieval system indexes every prior engagement (200+ deals over 11 years) by sector, deal mechanics, valuation multiple, integration outcome, and the firm's own assessment notes. An analyst working on a healthcare services roll-up can pull the 7 most-comparable prior deals in 90 seconds, with the partner notes attached.
Board packs are assembled from the firm's templates — populated with extracted financials, scored risk register, comparable deals, valuation range, and the analyst's commentary. The pack lands on the partner's desk as a 60-page document, not as fragments across a shared drive.
Nothing about the firm's analytical methodology changed. The judgement, the framing, the recommendations — all still from the team. We removed the 3 weeks of reading and let the team start at the point where their judgement matters.
Intralinks, Datasite, SharePoint — pulls documents into the structured pipeline with full audit trail
Per-document-type extraction into firm house templates, with cited source extracts for every finding
Change-of-control, MAC clauses, IP indemnities, restrictive covenants — surfaced with original-document quotes
Historical P&L, balance sheet, cash flow into the firm's template with anomaly flags
11 years × 200+ deals indexed by sector, mechanics, valuation; analyst queries return citations
Auto-ranked match candidates with partner-notes attached; analyst chooses 5–10 for the comp set
60-page document assembled from templates, extracted data, risk register, valuation, comp set, commentary
Deal status, document coverage, risk hot-spots, open analyst questions — single screen
Every extraction cited, every model input sourced, every partner edit timestamped
Deal data stays in AWS London; SOC 2-aligned controls; data deletion on engagement close
DD cycle time compressed from 5–6 weeks to 8 days on the first 12 deals through the pipeline. The compression came almost entirely from removing the reading + extraction phase — partner-level analysis and synthesis time barely changed.
The firm delivered 18 mandates in the 14 months following deployment, against a baseline of 7–8 in the prior comparable period. Partner time per deal dropped roughly 25% — small relative to the analyst time saved, but enough to materially increase the partner's capacity to take on new mandates.
Analyst time recovered: roughly £840k/yr based on the firm's blended analyst rate × hours saved across the 18-deal volume. The firm reinvested most of that into business development and a new sector practice rather than reducing headcount.
The precedent retrieval system has surfaced 4 specific instances where a comparable-deal lesson changed the firm's recommendation on a live mandate — exactly the kind of compounding institutional memory the partners had been losing as people moved on.
9 weeks from kickoff to first deal through the pipeline.
Week 1–2: Discovery, methodology audit, data room sample analysis. Week 3–5: Extraction pipeline, risk clause flagging, financial model population. Week 6–7: Precedent indexing (11 years of deal history), retrieval system, comparable scoring. Week 8: Board pack generator, partner dashboard. Week 9: Pilot on a live engagement with partner shadowing the AI output, production deploy.
More systems like this
How many partner-hours could you reclaim from data extraction?
Book a 30-minute call to talk through where the highest-leverage compression points are in your DD process.