Work

Starting point

Bankers were trying AI tools on their own accounts. Client files sat in shared folders with no rules for AI use, and the partners had a long list of workflows they wanted.

What was set up

A plan for a firm-owned Microsoft setup covering accounts, client-file permissions, records settings and four shared skills, plus a scope for the first banker workflows.

Safeguards

• Client-file access tested case by case, including files the AI creates
• Every banker workflow scoped and tested on its own before it is built
• Clear limits on accuracy, records and support promises, agreed before work starts

Result

A scope the partners could approve, with a fixed setup, each workflow priced on its own, and no open-ended promises.

What came next

Setup of the shared environment, then the first banker workflow chosen by the partners.

Starting point

One person on the client-success team kept onboarding plans, health checks and follow-ups across notes, email and spreadsheets, by hand.

What was set up

Eight workflow skills, including a morning list, call prep, call close-out, drafts and chasers, and client health reviews, on a records service the firm controls.

Safeguards

• Nothing is saved until they say so, and nothing is ever sent
• Health ratings come from tested business rules, never worked out by the AI
• Every fact in an answer says where it came from
• Daily copies of the records with a written recovery procedure

Result

Each saved change has a draft, an approval and a source. The rules behind every rating are written down, tested and agreed with them.

What came next

More skills added through the same review and test process as the team asks for them.