AI · 2026-08-04 · 9 min read · by Juan Carlos Zuloaga
Nobody publishes prices, so buyers arrive with no reference point and get anchored by the first proposal they see.
This article gives ranges, explains what moves them, and shows how to structure payment so that stopping is cheap.
All figures in euros, based on the European mid-market. Treat them as orientation, not quotes.
Diagnosis. Understanding your processes and data. Typically 10–20% of a first engagement.
Build. The largest component. Scales with system integrations, not with model choice.
Adoption. Training, documentation, and the weeks of small corrections after launch. Consistently underestimated — budget 20% and you will be roughly right.
A proposal where build is a minority of the cost is a research project in disguise.
Focused sprint (2–4 weeks, one process): roughly €8,000–€20,000.
Quarter engagement (one cluster of use cases plus internal capability): roughly €30,000–€80,000.
Ongoing sparring retainer: roughly €1,500–€5,000 per month depending on cadence.
Large strategy houses sit well above these ranges and small freelancers below. The spread reflects overhead and seniority far more than delivery quality.
- Integrations with legacy systems that lack an API.
- Regulated data, which adds review cycles rather than build time.
- Multiple stakeholders with veto rights.
- Undefined success criteria, which turn the project into an open loop.
- Multi-language or multi-country rollouts.
Notice that only the first is technical. Cost in this category is mostly organisational friction, priced.
- A single decision owner.
- A narrow first scope.
- Access to systems arranged before kickoff, not during.
- Willingness to change a process rather than replicate it exactly.
- Reusing an existing pattern rather than designing from scratch.
That fourth point is worth the most. Automating a bad process faithfully is the most expensive way to get a small improvement.
Split any engagement into stages with a decision gate between each. Pay per stage.
After stage one you should have enough evidence to continue or stop without penalty. If a contract makes stopping expensive, the incentive to tell you bad news early disappears.
We structure sprints exactly this way, in euros, with a fixed scope per stage — because the alternative rewards the wrong behaviour on both sides.
Use the baseline you took before starting. Compare time per unit, error rate, and cycle time.
Then apply a conservative discount — assume half the improvement is attributable to the project and half to attention. If it still pays back within a year, it was a good decision.
Companies that skip this calculation cannot defend their AI budget in the next planning round, regardless of how well the project went.
Write one page: the process, the current numbers, the outcome you want, the systems involved, and who decides.
Send that same page to everyone you are evaluating. You will get comparable proposals instead of five different interpretations, and the quality of the questions you get back will tell you who to shortlist.
If you want that page reviewed before you send it, book a call: https://calendly.com/inspiralgrowth/discovery-call