SUBMIT A SUPPORT REQUEST
When was the last time your SLT recorded what it decided and why?
Schools make dozens of significant decisions every week. Some are minor. Many are not. Exclusions, changes to SEND provision, staffing adjustments, curriculum pivots, responses to safeguarding concerns. These decisions happen in meetings, in corridors, in quick exchanges between a headteacher and a deputy. And in the vast majority of cases, what gets written down, if anything does, is what was decided, not the thinking that led there.
That gap matters more than most school leaders realise. Governance audits, Ofsted inspections, and parental complaints all have a habit of asking questions that only a well-maintained decision record can answer. And as remote and hybrid working has become more common for SLT meetings, particularly across multi-academy trusts, the informal paper trail that once existed has become thinner still.
What research on AI-assisted decision-making is finding
A growing body of academic research, including recent work published in Science, has been examining how AI can genuinely support human decision-making. Not by replacing judgement, but by helping people structure their thinking, surface assumptions, and document their rationale more consistently. One consistent finding across studies is that when AI assists with the process of decision-making rather than the decision itself, the quality of the outcome improves. Not because the AI is smarter, but because the structured prompting encourages the people involved to articulate what they actually think before they commit to a course of action.
A 2024 systematic review published in Group Decision and Negotiation identified four emerging patterns in how AI and humans collaborate on decisions. The most useful for organisations like schools is what researchers call the interpretive analytical model, where AI helps a team interpret evidence and evaluate options, while humans retain full ownership of the final call. That distinction matters in education, where decisions carry ethical weight and professional accountability that no algorithm should absorb.
Why this matters to primary schools right now
Meeting bots have been growing rapidly. Tools like Microsoft Copilot in Teams, Fireflies, and Otter.ai are increasingly common in organisations that hold remote or hybrid meetings. A significant number of school SLT teams are already using Teams or Zoom for some of their collaboration, and some are recording those meetings, often without a clear plan for what happens to the transcript afterwards.
This is where the opportunity sits. A raw transcript from a meeting bot is not particularly useful. But a transcript that has been processed into a decision log, capturing the options considered, the concerns raised, the rationale for the outcome, and the actions agreed, is genuinely valuable. It creates an institutional memory that survives staff turnover. It gives governors something meaningful to review. And it means that when a parent challenges a decision six months later, the school has a clear account of how it was reached.
Research consistently finds that when AI supports the process of decision-making rather than the decision itself, the quality of thinking improves. Not because the AI is smarter, but because structured prompting encourages people to articulate what they actually think.
What this could look like in practice
The workflow does not need to be complicated. An SLT meeting is recorded via the existing video conferencing tool. The transcript is passed, manually or automatically, to a structured AI prompt that asks specific questions: What decision was reached? What alternatives were considered? What concerns were raised and how were they addressed? Who was present and who holds accountability for the outcome? What is the review date?
The output is not a set of minutes in the traditional sense. It is a decision record: short, structured, and searchable. Stored consistently over time, it becomes a genuine audit trail of SLT thinking. And because it is generated from what was actually said rather than reconstructed afterwards by whoever drew the short straw for minute-taking, it tends to be more accurate and less filtered than traditional minutes.
There are practical constraints worth acknowledging. Any recording of staff meetings needs appropriate consent and clear data handling policies. The transcript should be treated as a working document, not a verbatim record for external use. And the AI output should always be reviewed by a human before it is filed. The process supports good practice; it does not replace it.
Where to start
The simplest first step is to take the transcript from your next SLT meeting, whether generated by a meeting bot or typed up manually, and pass it to Claude with a prompt asking for a structured decision log. Specify the format you want: decision reached, options considered, concerns raised, actions agreed, accountability, review date. Review the output against your own recollection of the meeting. If it captures the session more usefully than your current minutes, you have the beginning of a process worth building on. If it does not, adjust the prompt and try again.
The investment is small. The institutional value, accumulated over months and years of consistent use, is considerably larger. And it is the kind of quiet, unglamorous improvement that makes a real difference when it matters most.
