AI in finance, legal and professional services: the document is the workflow
Contract review, audit sampling, research memos and the verification-and-audit-trail discipline they all share
The question: In professions whose product is a judgement written down, where does AI create value — and what stops it creating liability?
Finance, law, accounting and consulting are document-heavy, judgement-heavy and regulated — the sectors where cheap competent output collides hardest with accountability. The 2025 deployments show the pattern that works and the one that gets a firm sanctioned.
What the lesson covers
The workflow is the document. Contracts, audit files, research memos, filings, valuations: the professional's work is reading, extracting, comparing, drafting and signing. Multimodal extraction (Lesson 6), retrieval over an authoritative corpus (Lesson 9) and drafting with validation (Lesson 22) map onto it directly — which is why these professions adopted fastest and why Hebbia-class tools (Lesson 21) found their market here.
Where the value is: first-pass contract review against a playbook (flag deviations, draft the redline), audit sampling and anomaly detection across the full population rather than a sample, research memos grounded in the actual authorities with citations that resolve, and the drudgery around filings. The pattern that works is the one from Lesson 26: the model drafts and triages, the professional verifies and signs.
Where the liability is: a research memo citing cases that do not exist, a contract clause mis-summarised, an audit conclusion resting on an extraction nobody checked. Courts sanctioned lawyers in 2023–2025 for filing AI-fabricated citations; the lesson is not "do not use AI" but "verification is the product" (Lesson 38). Every output that a professional signs must be checked against the source, and the check must be recorded.
The audit trail (Lesson 13) is therefore a feature here, not overhead: which model, which version, which sources, what the professional changed, who signed. In regulated professional work it is what makes the capability usable — the firm that can show its work may use a more capable tool than the firm that cannot.
The business model shifts with it (Lesson 38): billable hours on competent output fall, and the premium moves to accountability, framing and the client relationship. The firms that thrive sell judgement and signatures and use AI to make them faster and better evidenced; the ones that sell drafts compete with a subscription.
Key points
- In these professions the document is the workflow: extract, ground, draft, verify, sign.
- Value: playbook contract review, full-population audit testing, grounded research memos, filing drudgery.
- Liability: fabricated citations, mis-summarised clauses, unchecked extractions — verification is the product.
- The audit trail is a feature: it is what makes a more capable tool usable in regulated work.
- The premium moves from drafts to judgement, signatures and the client relationship.
Framework — Extract → Ground → Draft → Verify against source → Sign → Trail
The professional-services loop: the model does the first four with validation; the professional verifies, signs and leaves a trail.
The lab
Run one professional-services workflow end to end with verification and a trail.
Open this lesson, its lab and its quiz
Sources and further reading
- Hebbia: Redefining Productivity for Knowledge Workers Using AI — HBS case 125-075
- Expertise Disrupted? Consulting Confronts Generative AI — Darden UV9488
- Lesson 13 — the audit trail as product — Applied AI Academy