A financial planning firm put Claude to work without exposing client data

Clear rules for when to use AI

We built the firm an AI governance framework with three levels. Green means the team can use AI for the task. Yellow means ask a manager or the compliance officer first. Red means stop: tax information and Social Security numbers never go into AI. Each workflow gets a task card that lists the data that goes in, what comes out, who reviews it, and what records to keep.

The firm adopted the framework as its baseline while its compliance attorney reviews it alongside the firm's draft AI policy.

Nine skills, set up the way the firm works

We loaded 9 custom skills into the Claude desktop app, chose which local folders Claude can trust, tightened its security settings, and wrote custom system prompts so answers come back the way the team needs them.

Personal data stays on the firm's computers

The scrubber encrypts and hashes sensitive fields locally, so client details never reach an AI model in raw form. It matches the same client across files, so records still connect. The newest version finds sensitive fields on its own.

A second reviewer for financial plans

A QA tool now reviews financial plans automatically. We later moved it to a lower-cost model, and the firm confirmed the output quality held up.

What it added up to

• AI usage runs inside the firm's current subscriptions and promotional credits.
• A governance framework the firm can show its compliance counsel and extend to new workflows.
• Client details are scrubbed on the firm's own computers before any AI model sees them.

Next on the list

An asset allocation skill is in testing, and a daily download automation is planned for November.