An AI product can have an impressive demo and still be the wrong choice for young children.
Early-years settings have to ask a different set of questions from ordinary software buyers. Does the tool support development rather than replace human interaction? What data does it collect about children? Can educators understand and override it? What evidence shows benefit for this age group?
UNICEF’s Guidance on AI and Children 3.0 sets out a child-centred approach covering safety, privacy, fairness, transparency, inclusion, well-being and accountability. Those principles can be translated into practical procurement decisions.
This article complements, rather than repeats, broad guidance on AI in early childhood by focusing specifically on the buying decision.
1. What child-development outcome is the tool meant to support?
Reject vague claims such as “personalised learning” until the vendor defines the intended outcome. Is the product supporting language practice, educator planning, accessibility or another bounded task?
If the organisation cannot explain the benefit without mentioning AI, the use case may not be mature enough to procure.
The AI for Early Childhood Education course helps educators and leaders understand how smart technology can be evaluated within real early-years practice.
2. Why is AI necessary?
Compare the product with a simpler non-AI alternative. AI may add adaptive interaction or help educators organise information, but complexity should earn its place through measurable value.
3. What data are collected about children?
Inventory identity data, voice, images, behaviour, progress, device data and inferred characteristics. Ask whether the vendor collects more than the educational purpose requires.
Find out where data are stored, who accesses them, how long they are kept, whether they are used to improve models and how deletion works. Privacy expertise such as the Certified AI Data Protection Officer may be needed for higher-risk processing.
4. Is the system appropriate for this age and context?
Evidence from adults or older students does not automatically transfer to young children. Request evaluation with the intended age group, language, setting and accessibility needs.
Look beyond engagement. A product that holds attention is not necessarily producing a beneficial developmental outcome.
5. How does it protect human relationships?
Young children learn through relationships, play and interaction. Ask which activities remain educator-led and how screen or AI interaction fits within the setting’s pedagogical approach.
AI should support professional judgement, not become a substitute adult.
6. What happens when the AI is wrong?
Test incorrect, confusing and inappropriate outputs. Can the educator see why something was recommended? Can they correct it? Does the system learn or persist with the error?
For generative tools, assume fluent content can still be inaccurate. Build review into the workflow before material reaches children or families.
7. How are bias and inclusion tested?
Ask how the product performs across language, disability, culture, skin tone where computer vision is used, and other characteristics relevant to the function.
UNICEF’s child-centred guidance emphasises fairness and non-discrimination. Procurement evidence should therefore include groups who could be underserved, not only an average performance claim.
8. What can educators control?
Teachers and early-years practitioners need clear settings, override routes and the ability to stop the system. Ask whether automated recommendations can be disabled and whether educators can inspect or correct records.
Leadership capability matters here. The Artificial Intelligence for Educational Leadership course can support leaders creating institution-wide rules and governance around AI use.
9. What will parents and children be told?
Transparency should be age-appropriate and understandable. Parents or carers need to know what the tool does, what information it uses and how to ask questions or exercise relevant rights.
Do not rely on a long vendor privacy policy as the only explanation.
10. What evidence will trigger renewal—or exit?
Set success and safety measures before signing. Examples might include reduced educator admin time without increased errors, improved access for a defined learner need, or a measurable learning outcome supported by appropriate evidence.
Also define stop conditions: unacceptable content, data incidents, harmful bias, poor adoption or a vendor change that materially alters the product.
Use a child-centred procurement gate
Before purchase, require five PASS decisions:
- Benefit: a clear developmental or educator outcome.
- Evidence: support for the intended age group and context.
- Rights: proportionate data use, safety, fairness and inclusion.
- Human role: educators retain meaningful judgement and control.
- Lifecycle: monitoring, change notification and exit are defined.
If any gate depends only on vendor assurance, request evidence before award.
A practical example: an AI reading companion
Suppose a nursery is considering a conversational reading tool. The demo is engaging and adapts questions to the child.
The procurement review discovers that voice recordings are retained to improve the service and that the vendor’s performance evidence comes from older primary pupils. The setting pauses deployment, requests a data-minimised configuration and age-relevant evidence, and pilots the tool only with educator supervision.
The result may still be a purchase. The difference is that the decision is child-centred rather than technology-led.
Final takeaway
Buying AI for early years requires more than comparing features. Define the developmental purpose, minimise child data, test evidence for the intended population, protect human relationships and give educators real control.
The strongest product is not the one that looks most intelligent. It is the one that can demonstrate appropriate value for children within a safe, transparent and professionally led learning environment.
Responses