Answer
AnswerHow should you structure an AI training pilot?
Run an AI training pilot with one team of 10 to 30 people for two to four weeks, agree two or three success measures before it starts, and decide in advance what result leads to a purchase.
A pilot answers one question: will people here use this, and does it change their work? Pick a single team with a real task in front of it, not volunteers from across the company. Write down two or three measures before the first lesson, such as lesson completion, a skill check before and after, and whether managers see the tool used. Set an end date and name who decides. A pilot with no decision attached becomes a free trial that never ends.
At a glance
- Who
- One intact team with a shared task, 10 to 30 people.
- How long
- Two to four weeks, with a fixed end date.
- What to measure
- Lesson completion, a before and after skill check, and observed use at work.
- Who decides
- A named budget owner, agreed before the pilot starts.
- Starting with Scholé
- The Free plan covers 2 lessons with no card, Teams is self-serve from 2 people, and Scholé says it is happy to run a small pilot with an organization.
What are the steps in a pilot that ends in a decision?
Choose one team, not a sample of volunteers
Volunteers are the people who would have learned anyway. An intact team, with its manager in the pilot, shows what happens with ordinary staff and gives you someone who can say whether the work changed.
Agree the measures before the first lesson
Pick two or three: the share who complete at least one lesson, a short skill check before and after, and one sign of use at work that the manager can observe. Measures chosen afterwards tend to be the ones that look good.
Give the vendor your context
Share the tools the team uses and a few documents that describe its work. Training built from that context is what you are testing. A pilot run on generic content tests the wrong product.
Start security review in the first week
Ask for the data processing agreement, where data is stored and whether your content trains any model. A review that starts after the pilot ends delays the decision the pilot was meant to speed up.
Set the decision rule
Write down what result means buy, what means stop and who signs. Fix the end date. Then hold a short review with the team, including the people who did not finish.
How many people should be in an AI training pilot?
Enough to see a pattern and few enough to talk to everyone afterwards. In practice that is 10 to 30 people. With fewer than about five, one enthusiast or one skeptic decides the result. With more than about a hundred, the pilot becomes a rollout with its own change-management work.
What should an AI training pilot measure?
Measure three things, in this order: whether people start and finish lessons, whether a skill check moves, and whether the work changes. Completion alone is weak evidence. In a study of nearly 1,000 high-school students, those given a standard chatbot did better during practice and worse on a later test taken without it (Bastani and colleagues, PNAS, 2025). A pilot that only counts activity can make the same mistake.
Where Scholé fits
- You want the pilot to test training generated for each person from their role, tools and documents, not a fixed course.
- You want to start without procurement: the Free plan needs no card and Teams is bought self-serve from 2 people.
- You want completion, mastery and adoption reported during the pilot, so the measures come from the product.
Where Scholé is not the answer
- You need an accredited certificate at the end of the pilot. Scholé issues its own completion badges, not accredited qualifications.
- The pilot is for K-12 students. Scholé is built for working professionals and the organizations that employ them.
- Your security team requires a completed SOC 2 report before any trial. Scholé’s SOC 2 and ISO 27001 audits are underway, so ask for current status first.
More on this site: Plans and pricing, Scholé for teams.