AI Assistant
for Knowledge Workers

Situation

A professional services firm with around four hundred knowledge workers gave everyone access to an AI assistant. After three months, 11% used it weekly. License costs kept running, and the promised productivity gain was nowhere to be found in the numbers.

Challenge

The tool worked fine. The problem was that "you can use it for anything" means, for most people, that they do not know where to start. An empty input box is not a workflow. And the people who did use it did not share how, so no habit spread through the organization.

Approach

We moved away from "assistant for everything." In working sessions with four teams we identified the tasks that actually consumed time: searching through files, drafting reports, rewriting quotes to a client’s context, and finding internal knowledge. For each of those four scenarios we built a ready-made starting point, embedded in the tool where the work already happened.

Decision

Two requested features were deliberately not built: automatic client correspondence and support for performance reviews in personnel files. The first because of reputational risk, the second because of employment-law sensitivity. That choice was documented along with the reasoning, so it would not resurface every quarter.

Result

Weekly active use grew from 11% to 68%. A time measurement across two representative teams showed an average of three hours per person per week recovered, concentrated in the four scenarios. That time was not cut but redirected to billable work, which is what finally made the business case add up.