OpenAI · GPT
OpenAI integration in Danish business software
OpenAI integration means your backend calls OpenAI’s API — GPT, embeddings, optionally DALL·E — with your data as context, and the answer lands in a field a human can edit. This is not “we have AI”. It is a feature with an owner, a prompt contract and an audit trail. In Danish SMEs it pays off on draft case notes, classifying documents and searching your own files. It does not pay off as a marketing-homepage chatbot or automatic send-out without a human on anything that costs money or law. GDPR starts with minimisation: CPR numbers, health and payroll must not leave the building.
Build the feature into the system you already work in — not a separate “AI room”. How we put AI into custom software. Models change. Keep the engine behind a thin layer so you can swap GPT, Claude or Gemini without rewriting the screen.
Jobs that pay off in an SME
- Drafts for case notes, quote text, rejection mail
- Classification: document type, topic, priority
- Search in your own cases with embeddings
- Extract: “what did we promise in this thread?”
What saves hours
- A field with a draft a human edits
- A fixed prompt and a known output shape
- A log of prompt, answer and user
What wastes budget
- A homepage chatbot without your data
- “AI that runs the business”
- Automatic send-out with no human
Prompt
Menneske godkender før udsendelse
Operations that belong in the price
Minimise
Strip CPR, health, payroll and needless names before the call.
Call
Key on the server, a spend cap, timeout and retry.
Human
Edit and approve before anything leaves the building.
Log
Prompt, answer, model and user — or you cannot debug.
Speech-to-text is a different socket: Whisper. Images sit with DALL·E or Stable Diffusion. In a Ceptiv package OpenAI is one integration. See pricing.
What to do now
Point to one queue where you spend five hours a week writing the same thing. That is the brief. Not “can you just put ChatGPT on it”. Write to us.
FAQ about OpenAI integration
May we send customer data to OpenAI?
Only with a legal basis, a processor agreement and minimisation. Many flows can run on anonymised text. That is a legal choice, not a model choice.
Should the customer see that it is GPT?
No. The customer sees “Draft”. You pick the model per feature and can swap later.
What if the API is down?
Your app needs a fallback: empty field, queue or “try again”. It must not fall over because Dallas has an outage.
Is a homepage chatbot an integration?
That is a widget. An integration is when your case data is the context and the answer lands in your database.