How Should a School Leadership Team Evaluate an AI Vendor?

A provider has just finished an impressive demonstration. Teachers can imagine useful classroom applications. IT has been told that setup will be straightforward. Procurement has a price and a timeline. The feature list is long, and the room can see the possibilities.

The harder question is still unanswered: should the school trust this provider enough to keep talking?

Do not start by scoring every feature. Start with three questions:

1. Does it solve a real school need?
2. Can the school stay in control?
3. Can the provider support the work over time?

Move a provider forward only when it can give clear, credible answers to all three. If an important answer is missing, pause. A long feature list should never outweigh a safeguarding, data or governance concern.

This guide helps a school committee decide whether a provider deserves a closer look. It is not legal or procurement advice. Each school still needs to follow its own policies, local law, safeguarding duties and approval routes.

Does the AI solve a real school need?

An AI provider should progress only when the school can name the problem, intended users and educational purpose clearly. The committee should be able to test the proposed use against its own curriculum, learner context and teacher workflows, rather than relying on a broad promise to “support teachers”.

A clearer use might be: “Help teachers create first drafts of low-stakes quiz questions from our curriculum materials, with teacher review before anything reaches learners.”

That one sentence gives the committee something real to examine. It shows who will use the product, what information it may need, where teachers stay involved and what a useful result would look like.

It also separates two decisions that are easy to confuse. A provider may offer a credible product while the school is not yet ready to use it well. Our guide to readiness before scaling AI across schools covers that wider school decision.

Does it fit the school’s curriculum and context?

If a provider says its product aligns with the curriculum, ask it to show you. Which curriculum, subjects, age ranges and languages does it support? Can your academic team test examples using the materials and standards your school actually follows?

International schools should look closely at language, localisation and cultural context. A product built around one national system may still be useful, but the committee needs to understand where teachers will have to adapt or correct it.

UNICEF’s EdTech for Good Framework 2.0 looks at the purpose of a product, the organisation behind it and whether it is suitable for responsible use in education. That is a helpful reminder that educational fit is about more than adding school language to a general AI tool.

Where does teacher judgement remain?

“A human is involved” is not enough. The provider should be able to show where a teacher reviews, changes, approves or stops an AI-supported action.

Ask what happens when the system is uncertain or wrong. Can a teacher correct the output? Can staff see why a suggestion was made? Could anything affect a learner’s support, access or assessment without meaningful human review?

The OECD Digital Education Outlook 2026 makes an important distinction between completing a task and supporting learning. Teacher judgement and sound educational design therefore belong in the buying decision, not in a training plan added after the contract is signed.

What does the evidence really prove?

Different evidence answers different questions. A teacher survey may show that people liked using the product. Usage data may show that they returned to it. Neither proves that learning improved or workload fell.

If a provider makes a claim about learning, time or cost, ask for evidence that measures that claim. Note who produced the evidence, where the product was used, how long the work lasted and what its limits were. A case study can help you understand implementation, but it cannot promise the same result in your school.

Can the school stay in control?

The school should be able to understand and control who uses the service, what information it handles, where teachers remain involved and what happens when something goes wrong. A privacy policy alone does not show that the arrangement will work safely in everyday school life.

For a learner-facing service, ask about age restrictions, filtering, monitoring, staff controls and the route for reporting a serious concern. The provider should be able to explain what happens when the system produces harmful, inappropriate, inaccurate or biased material.

England’s Department for Education generative AI product safety standards give schools useful questions about privacy, security, oversight, age appropriateness and risk. International schools can use those questions as a reference, while still applying the law and duties that govern their own setting.

Where does the information go?

Ask the provider to walk you through one real use from beginning to end:

• What goes in? Student, staff, school, device or usage information.
• Why is it needed? Which parts of the service can work without it.
• Who can see it? The provider, other companies helping to run the service and authorised people in the school.
• Where does it go? Storage, processing and any international transfers.
• How is it used? Including whether inputs or outputs help train or improve a model.
• How long is it kept? Including backups, export and deletion.
• What can the school control? Accounts, permissions, features, monitoring and deletion.
• What happens when the service changes? New models, features, data uses or companies involved.

The DfE’s guidance on procuring educational technology recommends involving data protection and safeguarding expertise early and checking how AI outputs, training data and product changes are handled. Provider documents should support the answers. Ask for the current agreement, data flow, security information, retention rules and admin guidance rather than relying on a verbal assurance that the service is safe.

Who takes responsibility?

A credible provider should know who owns product safety, privacy, security and educational quality. It should also have a clear route for incidents, complaints and important product changes.

Certifications can help, but they do not answer every question. Check which service, period and controls they cover. Then connect the provider’s responsibilities to named owners inside the school. Our guide to school AI governance roles and decision rights shows how to do that without sending every question to a large committee.

Can the provider support the work over time?

A credible provider should show how the school will set up, support, scale and leave the service, not only how the product performs in a demonstration. The committee needs a realistic view of responsibilities, staff time, total cost, evidence, support and exit before it progresses.

What will everyday use require?

Ask what the school and provider each need to do before the first user begins, during a pilot and if the service expands.

Make five things clear:

• Setup: Who leads accounts, access, integrations, settings and testing.
• Staff support: What training, time and guidance each role will need.
• Ongoing help: How support works when the first answer does not solve the problem.
• Growth: What changes when more classes, year groups, languages or schools join.
• Exit: How the school can export its information, close accounts and confirm deletion.

The OECD’s work on AI adoption in education systems shows why policy, staff confidence and practical support matter. A school should not sign on the assumption that staff will somehow fill every gap later.

Look at the whole cost as well as the licence price. Include setup, integrations, training, support, usage limits, optional features, renewal and exit. A low starting price may not stay low if the school has to buy extra services or add specialist capacity to make the product work.

What evidence should the committee ask for?

Use this table to organise the next conversation. Adjust it to the proposed use and the rules that apply to your school.

A pause is not always a rejection. It means the committee needs an answer before the provider moves forward. The provider may be able to close the gap, or the school may narrow the proposed use so that a different level of evidence is reasonable.

How should the committee decide what happens next?

Use three plain outcomes:

1. Pause or stop. A critical answer is missing, contradictory or unacceptable.
2. Progress with conditions. The provider looks credible, but a specific gap must be resolved before a pilot, contract or wider use.
3. Take the evaluation further. The evidence is strong enough for the next stage. This is not automatic approval to buy or deploy.

Scores can help compare providers after these critical questions are settled. They should never allow a large feature list or attractive price to cancel out a safeguarding, data or governance problem.

What should the school record?

Someone who was not in the meeting should still be able to understand what the committee decided and why.

Record:

1. The proposed use: Who will use it, how and for what purpose.
2. The evidence reviewed: Documents, demonstrations, tests and people consulted.
3. The unanswered questions: What is missing, unclear or conditional.
4. The decision: Pause, progress with conditions or take the evaluation further.
5. The owners and dates: Who will resolve each condition and when the decision returns.
6. The reasons to review again: A change to the use, users, model, information, controls, price or contract.

Could TopSchool fit your school’s needs?

TopSchool may suit schools looking for a joined-up, school-aware approach, but it should be tested against the same questions as any other provider. The relevant decision is whether TopSchool can show credible educational fit, school control, implementation support, evidence, cost and exit for the school’s proposed use.

PLAI™ is TopSchool’s Personalised Learning AI layer. It is designed around school context, curriculum, approved resources, policies, different user roles and teacher oversight, rather than starting as a generic AI tool.

That does not remove the need for scrutiny. Ask TopSchool to show how your proposed use would work, what information it needs, who controls access, where teachers stay involved, what implementation support the school receives and how the arrangement changes as use grows.

Whether the provider is TopSchool or another option, keep the same discipline. Start with a real school need. Protect the school’s ability to stay in control. Ask for evidence that matches the promise. Then record why the provider should, or should not, move forward.