AI in Childcare Management: Where It Helps and Where It Shouldn't

AI is useful in a nursery where the work is writing, searching or retyping: turning rough notes into an observation, finding the answer buried in your own records, or getting data from one system into another. It is not a substitute for a practitioner's judgement about a child, and nothing it writes should reach a family or a record without a person reading and approving it.

That is the line this article draws: where AI earns its place in early years settings today, where it shouldn't be used, and the questions to ask any supplier offering it.

Where AI genuinely helps

Writing up observations

Practitioners see far more than they have time to write up. Jotting a few words, or saying them aloud, and getting back a draft observation in the setting's usual style means more of what happens in the room gets recorded. The practitioner still decides what the observation says, edits the draft, and links it to the areas of learning; the manager still approves it before a parent sees it.

Checking writing against your house style

Most settings have rules about how observations are written: use the child's first name, describe what you saw rather than labelling it, avoid jargon in reports to parents. AI can check a draft against those rules as it is written and suggest a fix, which is quicker than a manager rewriting it later. These checks should be advisory: a reminder, not a block.

Reading a term of observations for the end-of-term review

An end-of-term review draws on observations that already exist, but reading a whole term of them for every child takes hours. AI can do that reading and lay out what it found with the evidence: what each area shows, which next steps look achieved, and what might come next. The key person still makes every judgement; they just start from the evidence rather than a blank page.

Answering questions about your own setting

“Which invoices are unpaid?”, “who's booked in on Friday afternoon?”, “what does our policy say about late collection?” are questions with answers in your records. Asking in plain English is quicker than building a report, as long as the answer shows where it came from so you can check it.

Getting data in

Moving to a new system usually means a spreadsheet export in someone else's format. Matching columns automatically, or reading a child's details off the old system's page, saves days of retyping. Every imported record should still be shown to you before it is saved.

Where AI should stay out of the way

• Judgements about a child's development. Whether a child is on track, or needs extra support, is a professional judgement made by people who know the child.
• Safeguarding decisions. AI can help summarise a long concern history for the designated safeguarding lead, but the DSL remains responsible for the record and for every decision that follows.
• Anything sent without review. No observation, report or message should go to a parent automatically because a model wrote it.
• Children's photos. Drafting a written observation doesn't need the AI to see a child's image, so there is no reason to send it.

Questions to ask a supplier

1. Which features use AI, exactly, and can each one be turned off?
2. What data is sent to the AI for each feature? Are photos or videos ever included?
3. Is our data used to train models? Is it ever visible to another setting?
4. Is everything the AI produces a draft that a person must approve?
5. How does the assistant show where an answer came from?
6. How does the system handle questions about child protection?

If a supplier can't answer these plainly, that tells you something. The ICO's guidance on AI and data protection is a useful reference for your own data protection impact assessment.

How Pigeonhole uses AI

In Pigeonhole, AI drafts observations from notes or dictation, checks staff writing against rules you write in plain English, prepares termly reviews from published observations (suggesting levels and next steps, and drafting the family's term summary), answers questions about your setting's own data, maps CSV columns on import, and drafts safeguarding summaries for the DSL to edit. Four rules apply to all of it: AI output is always a draft you review; writing-rule checks are advisory; photos and videos are never sent to the AI; and every approval step your setting uses still applies. Your data isn't used to train models, and observation drafting and termly review preparation can each be switched off.

See AI writing and AI insights for the detail, or nursery management software for how it fits with the rest of the system.