The honest reason a price is hard to publish is that "AI automation" describes work that ranges from a two-week integration to a six-month programme. The same three words cover connecting two systems that already speak JSON and rebuilding a claims process across four departments. Any number attached to the phrase itself is marketing.
What can be published is the shape of the cost and what moves it, which is what the rest of this page does. If you want a number for your own case, the mini-audit exists precisely to produce one without either side committing money first.
We split the work into stages so that the expensive commitment comes after the cheap information, rather than before it.
Mini-audit — free. A few days. You bring one process; we take it apart and tell you whether automating it pays, roughly how big the prize is, and what a full audit would cost and cover. No fee, no obligation to continue.
Full audit — $1,000–5,000, fixed. One to two weeks. Produces a process and data map, an ROI projection with the success metrics named, and a fixed-price proposal for the build. Credited in full against the build, so it is paid for once. The findings are yours whether or not you proceed.
Production build — fixed quote. Priced against the scope the audit produced, not against hours. The number is agreed before work starts and does not move unless the scope does, in writing.
Support — monthly retainer. Tuning, new scenarios and monthly reporting. Cancellable at any time; it is not a lock-in mechanism.
The reason the audit is credited is structural, not generous: it removes the incentive to sell a build that the audit should have talked you out of.
Almost none of the cost variance comes from the AI part. It comes from the state of the systems and the clarity of the decisions around it.
Number of systems in the chain. Each additional system is an integration, its own auth, its own failure modes and its own owner to coordinate with. Two systems is a project; six is a programme.
Whether the APIs exist. A system with a documented API and a sandbox is straightforward. A system with no API and no supported export can double a build, and sometimes we will tell you not to do it at all.
How clear the rules are. If the decision logic has to be discovered and agreed during the build, that discovery is the build. Clients who arrive with written rules routinely pay less.
Exception handling and compliance. Four-eyes approvals, audit trails, retention rules and data residency are real engineering, not configuration. In regulated sectors they are also non-negotiable.
Data quality. Inconsistent master data does not stop an automation; it makes the automation faithfully reproduce the inconsistency at speed.
A build price is not the whole cost of ownership, and a proposal that ignores the rest is not doing you a favour. Budget separately for model and API usage, which scales with volume and is usually the smallest line; for any licences the flow needs; and for the internal time the project requires — realistically a few hours a week from the person who owns the process, and more during the pilot.
The last one is the most often underestimated and the most decisive. Projects that stall almost never stall on budget; they stall because the one person who knows how the process really works had no time allocated to the project.
Sometimes the arithmetic does not work, and the mini-audit exists to find that out at no cost to you. Low volume, rules nobody can state, a process about to be redesigned, or an existing product that already covers the case are all reasons we will say so rather than scope a build.
We would rather lose a project at the mini-audit stage than deliver one that shows up in nobody budget as a saving. That is also the commercial logic: this business runs on referrals, and a build that did not pay for itself does not produce any.