Anyone who wants to use a language model in Salesforce sooner or later faces the same question: do we build our own OpenAI integration, or do we take a package that already does it? The cost of an OpenAI integration in Salesforce rarely sits in the first version. It sits in everything that comes after. This article puts both routes side by side, on cost and on speed, and says honestly when building it yourself is better.
What a custom integration involves
A custom connection to OpenAI usually consists of a named credential for the API key, Apex code that makes the call and processes the response, and an invocable action so admins can use the call in a Flow. For a developer who knows Salesforce, that is perfectly doable.
The real work is in everything around it:
- Error handling. What does the Flow do when the model does not respond, times out or returns an unexpected answer?
- Limits. Salesforce caps the number of callouts per transaction. A Record-Triggered Flow that makes a call for every record during a bulk import hits that ceiling sooner than you expect.
- Testing and maintenance. Apex needs test coverage before you can deploy it, and the code has to keep working when the provider changes its API.
- A second model. If you later want to try Anthropic or Gemini as well, that is a new integration with its own API format.
Where the costs are
When you build it yourself, you pay for two things: development time and the calls. The model provider charges for the calls based on usage. Development time is a larger one-off cost, but it does not stop at delivery. Every change, new model or outage lands with a developer again. Those ongoing costs are often missing from the budget.
With the AI Prompt Builder the costs are known upfront. You pay a fixed fee of €150 per month, including 1,500 callouts. Beyond that you pay from €0.10 per call, with volume discounts for larger usage. If you connect your own LLM, the per-call costs disappear and only the fixed platform fee remains. Updates and maintenance of the package are included.
To work it out for your own situation, start with volume. How many records pass through the Flow each month, and how many of them get an AI step? That number drives most of the bill on either route.
How fast you go live
A custom integration follows a development track: build, test, deploy through a sandbox, and only then the first Flow. The AI Prompt Builder is a Managed Package. Resolve IT handles installation, configuration and training, after which an admin with Flow experience builds prompt templates with merge fields. No Apex or API knowledge is required. Our guide to ChatGPT in Salesforce Flows shows how that works inside a Flow.
You notice the difference most with the second and third use case. A new prompt is then a configuration change rather than a new development request, and switching between OpenAI, Anthropic, Google Gemini and Azure Co-Pilot needs no new code.
When building it yourself fits better
A package is not always the right answer. Building your own makes more sense if:
- you have an in-house development team that will keep maintaining the code in the long run;
- your use case goes well beyond a prompt and a response, such as a complex chain of calls with its own logic;
- you have processing requirements that a standard package cannot meet.
For use cases such as summarising, classifying or drafting text from a Flow, as with AI-driven lead qualification, a package is usually faster and more predictable in cost.
Not sure which route fits your org and your volume? Get in touch and we will work through both options with you.

