A plain-language breakdown of what drives AI project cost up or down, without a single made-up number attached to it.
Key takeaways
The AI model itself is rarely the largest line item in a project's cost. What drives the total is how many systems it has to connect to, how much of the existing process has to be mapped and redesigned, and how much human review the workflow needs to stay trustworthy.
Two projects that both say “AI for customer enquiries” can cost very differently depending on whether one touches a single inbox and the other has to reconcile data across four disconnected systems first.
It is tempting to scope a first project broadly, on the theory that a bigger rollout is more efficient to build once. In practice, a narrow pilot on one workflow is both cheaper and more informative, because it surfaces the real integration problems on a small scale before you have committed budget to solving them everywhere at once.
A pilot that only ever touches one workflow can also be shut down cleanly if it does not work, which is a form of cost control a broad rollout rarely allows for.
A quote for building a tool is not the same as a quote for running one. Ongoing costs, monitoring output quality, reviewing edge cases, updating the workflow as your business changes, are genuine and recurring, and a proposal that never mentions them is usually missing part of the real total.
Ask, specifically, who reviews the tool's output after launch and how often, and what that review actually costs in hours. That answer tells you more about the true price than the build quote alone.
The single most expensive outcome is not an overpriced project, it is a finished one that the team quietly stops using because it never fit how the work actually got done. That cost rarely appears on an invoice, but it is the largest one most businesses actually pay.
Everything about scoping a project narrowly, testing it with the people who will use it, and building in a real review step exists to avoid that specific outcome, because it is far cheaper to prevent than to fix.
A 30-minute call is enough to scope what a first project would actually cost here.
