Automating document-heavy workflows for law firms.

A typical mid-sized practice loses fee-earner hours to four predictable workflows: contract review, document drafting, client intake, and matter summarisation. The work is repetitive enough to feel wasteful, important enough that you can't just outsource it, and structured enough that it's a perfect fit for AI implementation.

What it looks like in practice: a senior associate spending 90 minutes pulling key terms out of a contract before annotating it. A trainee retyping client details from an intake form into the practice management system, then again into the engagement letter template. A partner reviewing a bundle of NDAs that all say roughly the same thing, looking for the one that doesn't.

None of this work is hard. All of it costs real money, fee-earner time at fee-earner rates, doing administrative tasks at administrative speed.

1. Contract data extraction, pulling parties, dates, terms, governing law, liability caps and unusual clauses into a structured summary. The kind of one-page key-terms document associates currently produce manually. Lives in the data extraction lane.

2. First-pass contract review, reading a draft against a checklist of red flags and producing a review note with quoted clauses and proposed redrafts. Reduces a 90-minute review to a 10-minute approval. Lives in the AI agents lane.

3. Client intake automation, the chain from initial enquiry to matter open. Acknowledgement reply, conflict check, ID collection, engagement letter, matter setup. Eight steps becomes two manual touchpoints. Lives in the workflow automation lane.

4. Matter summarisation, turning a folder of correspondence into a structured chronology, key issues and outstanding actions. A matter you haven't looked at in six months can be re-grasped in five minutes instead of 45.

The audit ranks these by what's actually costing your firm the most, usually intake and contract review come out on top.

Every system I build for a legal practice follows the same rule: AI produces drafts, a fee-earner approves what goes out. The model never sends an email, files a document, or makes a substantive legal judgement unsupervised. What it removes is the reading, the data entry, and the boilerplate drafting, not the lawyer.

The architecture is the same across builds: a structured input (form, email, document upload) flows into a Claude-powered pipeline that classifies, extracts, summarises or drafts as needed. The output is structured, sections you expect, clauses quoted in full, ambiguity flagged rather than guessed, and lands in the practice management system or back in the fee-earner's inbox for review.

Source data stays inside your existing infrastructure. The AI processes what it needs, nothing more. Everything is logged.

A worked example: a 12-fee-earner commercial firm with a manual client intake. Before: eight steps from enquiry to matter open, three hours of someone's week, lead-to-reply averaging 24–48 hours, two leads a quarter dropped because nobody followed up. After: two manual touchpoints, 30 minutes a week, lead-to-reply 90 minutes, zero drops in the first quarter.

Another: a contract review pipeline. Before: 90 minutes per contract review for an associate, partners receiving an annotated PDF on day two. After: 10-minute approval of a structured review note, partners receiving it the same morning. Stack: Claude Sonnet, Node.js, integration with iManage. Build time: 4 weeks.

Numbers vary by firm. Shape doesn't.

A good fit if:

• You have at least one workflow that runs 10+ times a month (intake, contract review, matter summarisation).

• Fee-earners are spending visible chunks of time on tasks that don't require their judgement.

• You're comfortable with the AI drafting outputs that a fee-earner reviews before they ship.

• Your firm has 5–50 fee-earners, the sweet spot where the pain is real but the build stays proportionate.

Probably not the right move yet if:

• You're a sole practitioner, the volume usually doesn't justify the build cost.

• You're a Magic Circle / Silver Circle firm, your scale needs a different conversation (enterprise procurement, multi-office rollout).

• You're looking for the AI to make legal decisions unsupervised. That's not what this is.

This page covers the law-firm angle. The broader practice is AI implementation — the audit, scoping, build and measure process I follow across every engagement, regardless of sector.

Worth reading: What AI Implementation Actually Means for a Law Firm.