How much does a custom AI project cost? What really drives the price

"How much does it cost?" is the most legitimate question there is, and the hardest to answer in one line. Not because providers hide their prices, but because two projects with the same name — "a WhatsApp assistant", "invoice automation" — can have almost nothing in common once you look at what they actually have to do.

Rather than a price list that would be wrong for half of our readers, here is what drives the price. With it, you can estimate the order of magnitude of your project and, above all, understand a quote.

The seven factors behind the price

1. The number of processes

This is the first multiplier. Automating one process — supplier invoice entry, for instance — is a project. Automating five at once is not five times the work: it is often more, because processes intersect and their interactions have to be handled. Hence our constant advice: start with one.

2. Integrations

Every tool to connect — ERP, email, spreadsheet, CRM, WhatsApp — means work. A modern tool with a documented programming interface connects quickly. Old software with no planned access, or an Excel file that three people edit by hand, takes far more effort. Before any quote, list the tools involved: that is often where most of the cost hides.

3. The state of your data

AI is only as good as the data it is given. Invoices all in the same format, a clean history, well-named files: the project moves fast. Crooked scans, three contradictory versions of the same procedure, duplicates in the customer file: order has to be restored first, and that work is billed.

4. Languages and formats

Text in French is the simple case. Messages mixing Darija, French and Arabic, voice notes, phone photos of documents: every extra format requires testing, tuning, sometimes a different model.

5. The reliability required

An AI that suggests a draft for a human to review can occasionally be wrong without consequence. An AI that acts alone — sends the invoice, answers the customer, approves the order — must be far more reliable, and that reliability is built: control rules, edge cases, tests on hundreds of real examples. It is often the difference between a prototype and a production system.

6. Hosting

Two broad options, with very different costs. A model reached through a cloud API: little upfront investment, but a bill that grows with usage. A model installed on your own servers: an upfront hardware investment, then a near-zero marginal cost. We detail this choice in our comparison of local AI and cloud APIs.

7. Volume

Processing fifty documents a month or fifty thousand does not require the same resources, nor the same attention to performance. Volume also strongly affects running costs in the cloud.

The three cost lines not to forget

A development quote only tells part of the story. The full cost of an AI solution comes in three lines:

• Building it. Design, development, integrations, testing, go-live, training. This is what the quote shows.
• Running it. Hosting, calls to AI models if you use an API, any subscriptions to connected tools. It is a monthly expense that depends on your volume.
• Maintaining it. The tools around the solution change, and so do your needs. An ERP update, a new type of document, a business rule that changes: someone has to follow up. Either the provider, or your own team if the solution is well documented.

A project whose quote only covers the first line is not cheaper: it is simply incomplete.

How to lower the bill without lowering the result

Start with a single process, the most frequent one. The time it saves often pays for what comes next.

Insist on a prototype on your real data before committing to everything. A few weeks is usually enough to know whether the approach holds.

Keep a human in the loop at first. An AI that suggests and a human who approves is cheaper to build than an autonomous AI — and you move to autonomy once trust is established, on the cases where it is justified.

Reuse your tools. A solution that fits into your email and spreadsheets costs less than an entirely new application, and your teams adopt it faster. We discuss this in custom-built or off-the-shelf.

Reading a quote: five questions to ask

1. What exactly is included? Integrations, testing, training, documentation: every line should be named.
2. What will it cost to run each month, at my current volume and at twice that volume?
3. What happens when the AI gets it wrong? Who notices, how, and who corrects it?
4. Who owns the solution? The code, the settings, the data: can you change provider without rebuilding everything?
5. Who maintains it, and at what price? A support contract, a flat fee, on demand: what matters is that it is written down.

A serious provider answers these five questions easily. If they hesitate, that tells you something.

The right order

Don't start with "how much does it cost", but with "how much is this costing me today". The time spent every week on the process, multiplied by the cost of the people doing it, over a year: that is your reference budget. If the solution costs clearly less than the problem costs you, the project makes sense.

Describe the process and what it costs you today: we'll tell you what a solution would change, and the order of magnitude of the investment.

Articles on regulatory topics are technical and documentary analysis, not legal advice. Excerpts from legal texts are reproduced from official publications; our interpretations are flagged as such.