ChatGPT is where most commercial teams already sit. A custom GPT wired to your company data API turns account research from a tab-switching exercise into a single question.
Build the action first
Actions are described with an OpenAPI schema. Expose only the endpoints a researcher needs: company search, company detail, and recent signals. Give every parameter a description — the model chooses arguments from those descriptions, not from your docs.
Use an API key or OAuth per user rather than one shared key, so usage and cost trace back to a person.
Write instructions that force sourcing
Tell the GPT to call search before answering, to state the register and last-updated date, and to say 'not filed' instead of estimating. Add an explicit rule that Dutch company data is out of scope on drimble.com.
Include two or three worked examples in the instructions. Examples shape behaviour far more reliably than adjectives like 'accurate'.
What people actually ask it
In practice the top requests are: summarise this account before my call, find similar companies in this region and size band, and tell me what changed since last quarter.
Design your tools around those three jobs and adoption follows without training sessions.
Frequently asked questions
Do I need ChatGPT Enterprise?
No, custom GPTs work on paid consumer plans too. Enterprise or Team mainly adds admin control, sharing scope and data-handling guarantees.
Can ChatGPT read a Drimble company report directly?
Yes — the public report pages are readable, but a proper action gives structured fields, which is far more reliable for scoring and list building.
