How to choose an AI training provider
Nine questions to ask a provider before you sign, what a strong answer sounds like next to a weak one, and the three claims worth ignoring entirely.
Ask nine questions and judge the answers. Whose documents the exercises use. What happens in week three. How you will know it worked. Who owns the material afterwards. Who is in the room. What they would refuse to teach. Which tools your team will actually have access to. What they would cut if you halved the budget. And who is delivering it, by name. A provider who answers all nine specifically is rare. A provider who answers with a certificate and a testimonial is selling a different thing.
Every provider sounds the same on their website. The questions below separate them in about twenty minutes.
For each one there is a strong answer and a weak one. The weak ones are often confident, which is why they are worth writing down.
1. Whose documents will the exercises use?
Strong. Yours. They ask for sample documents before the session and build the exercises from them.
Weak. "We tailor the content to your industry." Industry is not your company. A generic retail example is a generic example with the word retail in it.
This is the first question because it predicts most of the others.
2. What happens in week three?
Strong. Something specific. A follow-up session, an office hour, a channel where people ask questions while doing real work.
Weak. "We provide comprehensive materials." Materials are not follow-up. People do not reread a deck.
Training with no follow-up gets forgotten, and providers write about this problem themselves before quoting a single session.
3. How will we know it worked?
Strong. They propose a measurement. Usually a small set of real tasks, timed or scored before and after.
Weak. A feedback form. A completion rate. A certificate.
A satisfaction score tells you people enjoyed the day. It does not tell you anything changed.
4. Who owns the material afterwards?
Strong. You do. Recordings, exercises, and the prompts your team wrote during the session.
Weak. Access to a portal for twelve months.
You should be able to run the session again yourself for a new joiner.
5. Who is in the room, and are they the same room?
Strong. They separate the groups. Engineers get evaluation sets, cost control, and the failure modes that bite in production. Everyone else gets the tasks worth handing to a model and the longer list that is not.
Weak. One session for everyone.
One session for everyone is cheaper and serves neither group.
6. What would you refuse to teach us?
Strong. A real answer. Something like: we will not teach your finance team to trust a model with reconciliation, because that is a rules problem and a model would make it worse.
Weak. Nothing, or a pause.
A provider with no boundaries is a provider who will agree to whatever you ask for. That sounds like service and it is the thing that produces sessions nobody uses.
7. Which tools will our team actually have access to on Monday?
Strong. They ask this before the session and will not run it until the answer is clear.
Weak. They train on their own sandbox.
People trained on a tool they cannot open afterwards forget it within days. This is the most common preventable failure in corporate AI training, and it is an access problem rather than a teaching problem.
8. What would you cut if we halved the budget?
Strong. An immediate answer, with a reason. Usually they cut breadth and keep depth.
Weak. "We could reduce the number of sessions." That is not a choice, it is arithmetic.
9. Who is delivering it, by name?
Strong. A name, and what that person has built. You can look them up.
Weak. "One of our senior facilitators."
The person who sells the session is often not the person who runs it. Ask which you are meeting.
Three claims worth ignoring
Certificates. A certificate proves attendance. Nobody has ever chosen a supplier because their staff hold a certificate in prompting.
Client logos with no story. A logo means somebody bought something once. Ask what changed at that company and listen for whether the answer has a number in it.
Percentage improvements with no source. If a provider claims a figure, ask how it was measured and on what task. A good provider is pleased you asked.
One limit worth stating
We run AI training, so treat this list as written by an interested party and use it on us. The two questions we find hardest are three and nine. Measurement takes real work to set up and it is the first thing a client cuts for time. And the firm is small, so the name is always the same name, which is either reassuring or a concentration risk depending on your view.
What to do this week
Send questions one, three and seven to every provider on your shortlist, in one email.
You will get a reply within a day from the ones building around your work, and a brochure from the ones adapting a deck. That sorts the list faster than any call.
What the market charges is here, and if you are still deciding which format you need, the difference between a workshop, a bootcamp and a course is here. How our version works is on the services page.
Six providers, and what each one actually trains on, are compared here.
If you are sitting on a process that costs more hours than anyone wants to admit, that is the conversation to have.
Book the callWritten by
Radwan Altaf
Radwan runs AISynq. Before that he delivered software inside enterprise programmes at DHL, AT&T, DirecTV and Accenture, which is where the habit of measuring a result against its baseline came from. More about the firm.
Read next
- Best AI training providers for teams, 2026Six ways to train a team on AI, what each one actually teaches on, and why the productivity figures in this market are so hard to compare.
- What is a 20x company, and how do you become one?A 20x company does the work of a firm twenty times its size. The gap between the companies that got there and the 89% that did not is not the tooling.
- AI bootcamp, workshop or course?Three formats, what each one is actually for, and how to pick. Includes the case where the answer is none of them.
Get the next one
New writing in AI training as it goes up, roughly twice a month. One article per email and nothing else in it.