How much does AI integration cost for an SMB?

The cost of an AI project has two parts: development, paid once, and model usage, paid monthly to the provider. The first is predictable and fixed-price. The second has to be measured. Here is my rate card, the cost items people tend to forget, and a method to check that AI brings in more than it costs.

Price ranges

These amounts exclude VAT and come from my public rate card. The quote becomes fixed after scoping.

For a price that fits your tool, the estimator includes a “built-in AI” option among the features you can select.

Model usage costs

AI providers bill by the amount of text processed, measured in “tokens” (word fragments). As an example, OpenAI’s API price list, checked October 27, 2026, quotes prices per million tokens, with separate rates for text sent and text generated, and a cheaper deferred processing mode. The price gap between models in the same lineup is several dozen times.

Three levers drive the bill:

1. The model. A smaller model is often enough for sorting or extraction. This is the strongest lever; how to choose is covered in choosing an AI model.
2. The length of the texts sent. Sending a whole invoice when three fields would do multiplies the cost.
3. The volume. Measured on your actual activity: number of requests, documents or questions per day.

An illustrative example, with made-up amounts to replace with your own: 3,000 documents a month at €0.02 per document comes to €60 a month, or €720 a year. The real unit cost is measured during scoping, on a sample of your documents.

The costs people forget

An honest AI quote includes these items, or states clearly that they are excluded:

• data preparation: gathering examples, cleaning documents, defining the correct answers;
• evaluation: building a test set and measuring quality before going live;
• the review screen: the interface where a person checks and corrects what the AI suggests, often more work to design than the model call itself;
• privacy: pseudonymization, access rights, logging, checking the provider’s contract (see what data to share with AI);
• monitoring: revalidating the use case when the provider changes model versions.

Calculating what AI pays back

Value is calculated as for any business tool: time saved, errors avoided, shorter processing times.

An illustrative example, with made-up amounts to replace with your own:

With development at $9,000 (€8,000)excl. VAT, the investment would pay for itself in under a year in this scenario. The full method is in calculating the ROI of business software. The key point: measure the time actually spent today before starting the project, otherwise no gain can be demonstrated.

When the investment is not worth it

I will be straightforward: AI does not pay off everywhere.

• Volume too low. If the task takes two hours a month, development will never pay for itself.
• Task already covered. If your off-the-shelf software offers the feature, turn it on before funding custom work.
• A simple rule is enough. Sorting by keyword or form field costs less and does not make mistakes.
• Zero tolerance for errors with no way to review. If nobody can check the results, the risk outweighs the gain.

How an AI project works

1. A 30-minute call to identify the use case and check that it pays off.
2. Scoping (1 to 2 weeks): a test set on your data, model comparison, monthly usage cost estimate, fixed quote.
3. Development in batches, with the human review screen in the very first version.
4. Go-live and measurement of time saved over the first weeks.
5. Three months of support included, then maintenance at a fixed fee or by the hour.

The full offer is on the AI integration page. Payment follows the usual rhythm: a 50% deposit at the start, then the balance according to the schedule in the quote.