Case Study: An Internal Knowledge Agent That Freed Up a CEO's Time

Client details are anonymized. The problem, the build, and the results are real.

The problem

The CEO of a marketing agency in Switzerland was scaling the team. Lots of new hires and contractors coming in, and during onboarding, new people kept coming to him with the same questions. He didn't have time to answer them all, but he had to, because that was the only way work moved forward without getting blocked.

The bottleneck wasn't knowledge. The answers to all those questions already existed inside the company: SOPs, documented notes, internal messages. All of it valuable, all of it scattered.

The approach

The first questions on any project like this: can AI solve this? How? And to what degree?

Here, the shape of the solution was clear once the problem was framed properly. Don't generate answers from thin air. Collect the knowledge that already exists into a single source of truth, and put an agent in front of it.

The build

• One database as the single source of truth, fed from the company's existing SOPs, notes, and internal documentation.
• An AI agent that uses that knowledge base to answer any new hire's question in seconds, through a simple internal chat interface.
• Built as a web app: every employee has their own account with their own access level. The admin adds and manages sources, creates accounts, and controls who sees what.

No exotic tooling, no platform lock-in, and documentation that lets someone new pick it up and keep it running.

The result

• The CEO no longer answers the same repeating questions over and over. His time went back to the high-priority parts of the business.
• New hires get answers in seconds instead of waiting on one busy person, so they start delivering value sooner.

A major bottleneck in the company was solved, and that directly served the thing that matters: the company growing faster.

Working with me

Every project follows the same three principles: pick the right problem, keep it simple, make it outlast anyone. If you have a bottleneck that looks like this one, let's talk.