How to choose an AI training provider: eight questions to ask before you sign

Why does the choice of provider matter more than the choice of course?

Most organisations do not have an access problem; they have an application problem. Nearly nine in ten organisations now use AI regularly, yet only 37% attribute any earnings impact to it (McKinsey, The state of AI in 2026; Indicative). The tools are in place. What is missing is people who can apply them to real work, safely.

Training has not closed that gap. Only 36% of employees feel they have received sufficient AI training (BCG, AI at Work 2026; Indicative). In a 47-country study, fewer than half of employees (47%) had received any AI training, while 66% said they rely on AI output without checking its accuracy (University of Melbourne and KPMG, 2025; Indicative).

The demand will not ease. Employers expect 59 in every 100 workers to need training by 2030 (World Economic Forum, Future of Jobs Report 2025; Indicative). So the question is not whether to train, but which provider will change how work gets done.

What should you decide before you go to market?

Four decisions make every proposal easier to judge:

• The result you need. Decisions funded, workflows changed, agents built, or all three in sequence.
• Who it is for. Leaders who decide, teams who apply AI to their work, or individuals who will build.
• Your rules. Which AI tools are approved, and what data may and may not be used.
• How you will judge success. A measure you can take before and after, not a satisfaction score.

The eight questions

1. What will our people have produced by the end?

A strong answer names the output: a prioritised list of AI ideas with a decision brief, a workflow tested on real inputs, a working agent with a handover plan. A weak answer names a certificate, a number of hours or a satisfaction score.

2. Whose problems will they work on?

Ask whether the provider runs discovery before the program and builds exercises on your processes. Generic cases teach the tool; your own cases change the work. The organisations that get most from AI are three times as likely to have redesigned their workflows (3 in 4 high performers, against 1 in 4 of the rest; McKinsey, 2026; Indicative). Training that never touches a workflow cannot contribute to that.

3. How will you measure the change?

Look for a baseline taken before the program and a report against it afterwards: time and quality on named tasks, workflows in use, decisions made. If the only measure is how participants felt about the day, the provider is measuring the session, not the result.

4. Who delivers, and what have they built?

Ask for the facilitators by name and for what each has built or run outside a classroom. Practitioners who have deployed AI teach judgment, failure modes and trade-offs. Presenters teach features.

5. Which models will we use, and where does our data go?

A credible provider trains on the tools you have approved, or on models that can run in your own environment, with sample data that mirrors your records. Ask where prompts and files go during the program, and who can see them.

6. Can you deliver in Arabic and English?

In the Gulf, many participants work across both languages. Ask for bilingual facilitators and materials written in each language, not translated slides, and for examples that use Arabic documents and requests.

7. Where does a person approve what the AI does?

Every workflow or agent built in training should have a defined point where a named person signs off before anything binding happens. If governance is a separate module at the end, it will not shape what people build.

8. What happens after the first program?

A single workshop rarely changes an organisation. Ask how the provider moves from leadership decisions to team workflows to live pilots, and whether the stages can be contracted once, so each step starts without a new procurement round.

Which red flags should end the conversation?

• The output is a certificate or a count of training hours.
• The same deck is used for every organisation, with your logo added.
• No baseline, and no plan to measure anything after the program.
• Facilitators described as "AI experts" with no named work behind the title.
• No clear answer on where your data goes.
• Training built around one vendor's product, delivered by that vendor.

How do you write a scope that makes answers comparable?

Ask every bidder to answer the same five fields for each audience: the output, how it will be measured, the named facilitators, how data is handled, and the language of delivery and materials. State formats in days, not hours, and ask for one anonymised example of a past output. Proposals written to the same fields can be compared in an afternoon.

Measure where your own team stands: take the free Readiness Snapshot →