AI · 2026-08-04 · 10 min read · by Juan Carlos Zuloaga
Every agency added "AI" to its homepage in the last eighteen months. Very few added the ability to ship it.
The hard part of choosing an AI consulting company is that the proposals all look similar. Same discovery phase, same maturity model, same roadmap deliverable.
This is the framework we would use if we were the buyer.
Strategy houses. Excellent at framing, expensive, and they hand off the build to someone else. Useful if your bottleneck is genuinely executive alignment.
Implementation shops. They build what you specify. Fast and cheap when you know exactly what you want; dangerous when you do not, because they will happily build the wrong thing.
Partner-operators. A small team that diagnoses and ships, and stays close enough to see whether it worked.
Most mid-size companies think they need the first and actually need the third. Decide which you are buying before you read a single proposal.
Case studies are easy to write. Working systems are not.
Ask for one reference where the work is at least twelve months old and still in use. Then ask the reference a single question: what broke, and who fixed it?
Every real project has an answer to that. A firm whose references have no failure stories has either done very little or is coaching them.
Who will be in the room after the contract is signed? In many firms the senior person you meet in the pitch appears once a month afterwards.
For a mid-size engagement, you want a small team where the person diagnosing is close to the person building. Distance between those two roles is where requirements get lost.
At Inspiral Growth, engagements are led by the person doing the work. That constrains how many clients we take. It also removes the translation layer.
Day rate. Transparent, but rewards slow work.
Fixed scope. Predictable, but every change becomes a negotiation.
Retainer. Good for ongoing capability, bad if there is no clear deliverable.
Outcome-linked. Rare, and only honest when the outcome is genuinely attributable.
We price in euros, fixed-scope for sprints and retainer for ongoing sparring, because those two match how the work actually behaves. Ask any firm to explain why their model matches the work rather than their cash flow.
- A proposal that names a model or vendor before naming your bottleneck.
- "AI strategy" as a standalone six-week deliverable with no build.
- No mention of data access, permissions, or GDPR.
- Unwillingness to state what will not be included.
- Headcount-reduction promises made before seeing your processes.
That last one is the loudest signal. Nobody can promise you a savings number from the outside.
Do not sign a year. Buy a small, sharply scoped piece of work with a defined artefact at the end.
A good pilot has: one process, one owner, one metric, a fixed price, and a date. If a firm cannot shape that, they will not manage a larger engagement well either.
At the end you learn two things that no reference call reveals — how they behave when something goes wrong, and whether your own team engages with what they built.
Referrals from companies your size in adjacent industries beat directories. So does looking at who built the systems you admire and asking directly.
When you shortlist, look at their own work first. A firm that cannot make its own site fast, accessible, and findable is telling you something about its standards.
If you want to test us against that standard, our case studies are the honest version: https://inspiralgrowth.com/case-studies