Client Intake & Document Intelligence
The situation
The situation
A 12-person law firm was handling all client intake manually. When a potential client filled out the firm's intake form, a paralegal would read it, extract the key legal details, determine what type of case it was, decide which attorney it should go to, write a brief summary, and then message the attorney in Slack. This process took roughly 15–20 minutes per intake and happened 10–15 times per day — consuming over 3 hours of productive time.
The complication
The complication
The firm had tried using a standard Zapier workflow to automate routing, but it couldn't handle the variability of how people described their legal situations. A client writing "my employer didn't pay me for my last two weeks" and another writing "I believe I've been wrongfully terminated" could both be employment cases, but keyword matching couldn't reliably tell the difference — and wrong routing created problems for the attorneys.
What we built
The system, component by component
An intake form connected to a processing pipeline via webhook — the moment a form submits, the pipeline starts
Claude API integration that reads the full form, extracts key legal entities (parties, dates, jurisdictions, claim type), and classifies the case into one of seven practice area categories
Automated brief generation: a structured one-page summary in the firm's preferred format, ready for the attorney to read in under two minutes
Slack routing that sends the brief and classification to the right attorney channel — different attorneys handle different practice areas
A Notion log that records every intake, its classification, the assigned attorney, and the timestamp — giving the firm a searchable intake history for the first time
Under the hood
Technical note
The classification uses Claude's reasoning capability rather than simple pattern matching. We prompt it with the firm's actual practice area definitions (which we derived from a brief with the managing partner) so it classifies the way the firm would — not the way a generic legal taxonomy would. The error rate in the first month was zero routing mistakes across 200+ intakes.
Stack used
What changed
What changed
Paralegals no longer spend any time on intake processing. Each form that arrives is read, classified, briefed, and routed in under 30 seconds. The attorneys receive better-formatted briefs than they were getting from manual processing — and the firm now has a full intake history they can search, filter, and report on.
Next case study
AI Content & Reporting Pipeline
Recognise any of this in your own business?
If you have a process that's eating time every week, tell us about it. We'll come back within 24 hours with an honest assessment.