How much does an AI project cost for an SME? The factors that matter

"How much will it cost?" is the question we hear most often about AI projects, and the honest answer is the least satisfying one: it depends. But "it depends" doesn't mean "impossible to say". It means the number is driven by a handful of specific factors — and knowing them lets you read a quote critically, and cut spending where it adds nothing.

## What drives the cost

- **Scope.** An assistant that covers one process costs a fraction of a platform that touches ten. The tighter the problem, the smaller the bill. - **Data readiness.** If your documents and records are already organised and accessible, work starts immediately. If they need to be collected, cleaned and structured first, that's often the biggest line item in the whole project. - **Integrations.** Connecting to a modern system with an API is routine; connecting to a closed, fifteen-year-old ERP means custom work. - **Buy vs build.** Where an off-the-shelf product covers the case, adapting it costs less than building from scratch. Custom development earns its price where the process is genuinely yours. - **Requirements.** Privacy, security and data-residency constraints add work — legitimate work, but it belongs in the estimate from day one.

## Indicative ranges, not quotes

With all due caution: a pilot built on existing tools can sit in the low thousands of euros; a solution configured on top of a ready-made product, with light integrations, lands in a middle band; a custom integration with your core systems typically starts in the tens of thousands and grows with scope. These are orders of magnitude, not prices — your number depends on the factors above.

Then there's the running cost, which many forget: calls to AI models are billed by usage, on top of hosting, monitoring and maintenance. The figures are usually modest, but they belong in the budget from the start, because they scale with adoption.

## Keeping the budget under control

Start with a pilot and a single success metric — hours saved, response times, cases handled — and decide in advance what number justifies the next step. Buy what's standard; build only what sets you apart. And be wary of both the inflated quote "because it's AI" and the tiny one that ignores data and integrations: the truth usually lives in the details that are missing.

A serious estimate starts from your process, not from the technology. If someone gives you a price before understanding how you work, that number isn't worth much.