Document automation for a trust attorney

The problem

A large trust project required separate transaction documents across several families: each with its own names, numbers, and clauses. The data arrived in spreadsheets, hand-filled by external collaborators working late under a deadline. Every typo was a legal risk. The attorney is personally accountable for every document's validity. Done manually, the work wasn't achievable in the time available.

The approach

We deliberately opted for the simplest possible pipeline, one the attorney could follow step by step. A single Word template holds all the legal language, with tags for the details that change and clauses that switch on and off based on the data. The collaborators kept working in the spreadsheets where they were comfortable. From there, we focused on two steps: 1) validate and 2) generate. The validator checks everything before any document exists and points to the exact cell when something is off. Only a clean spreadsheet becomes documents, each with a line in an audit log.

We had a working prototype up and running within a day or two. Getting it completely right took dozens of runs, but the structure held even as requirements changed mid-project.

The outcome

• We delivered. More than 100 documents, a week ahead of the deadline, and $135,000 of billing enabled.
• Nobody lost. The attorney billed more profitable hours, not fewer. Her clients saved money on work that would otherwise have gone to outside specialists. And the clients came back wanting more of her services.
• The capability persists. The attorney, who started with essentially no AI experience, runs and extends the system herself. She can now confidently take on the next phase of her engagement with triple the number of required documents.

What she says it changed

Working at the periphery of her expertise. With the system in place and expert review where it counts, work she'd previously have sent out came in-house: a matter that would have meant roughly 30 hours billed to a corporate attorney took two. "I get paid more, while the specialists get paid less, and I get more powerful as I learn the tools."

Quality control is the craft. "You obviously can't take the results at face value." As one example, tax treatment differed across states and needed careful human judgment. The system was built for exactly that: verification you can see, designed for an accountable professional who must answer for every choice.

Why it worked

• Legible beats impressive. The system was shaped like something the attorney already understood, so she could follow every step. That is what enabled her to trust it, learn it, and take it over.
• More verified, not less. The checks ran deeper than anything she would have done by hand, and every one was visible in the logs. The usual AI fear, reversed.
• A concrete first pass to react to, and a pipeline that absorbed mid-stream changes without derailing.
• Understanding as a deliverable. The attorney remains accountable to her clients, so the system was built to be explained, spot-checked, and signed off, not to replace her judgment. She owns it because she understands it.