An Internal AI Assistant in Practice: The Hard Part Was the Data, Not the AI
How Devcity connected company systems to an AI assistant via Make and MCP, why cleaning up data took the most time, and how ISO 27001 and GDPR shaped it.
Our sales team is preparing a public tender bid. They need to know which certifications we hold and which references fit. That used to mean digging through several systems and asking colleagues. Today they simply ask our internal AI assistant.
What we ask the assistant
Everything we used to look up across different systems is connected to the AI assistant and can be queried in plain language. It is used most by back-office, sales and marketing.
- Project overview and who works on what
- Certifications and suitable references for bids
- Website traffic and LinkedIn statistics
- Open positions
How it is built
Data comes from our internal information systems, Make connects them, and the assistant is Claude by Anthropic, connected via MCP (Model Context Protocol), an open standard that lets an AI assistant work with company tools and data in a secure, controlled way.
The hard part was the data, not the AI
Connecting the assistant was fast. What took time was consolidating what lives where and exposing only the data that is genuinely needed. An assistant is only as good as the data it can access.
Security from day one
We are ISO 27001 certified, so information security, GDPR and personal data protection are part of the design from the start, even for internal tools. Not after everything is finished.
You can get the tool in an afternoon. Cleaning up your data takes much longer.
In our view, this is what companies underestimate most about AI. If you are considering an AI assistant, start by asking which data it should see and who owns it.
Working on this yourself?
We are happy to show you how we built ours and talk through where things get stuck for you.
