
By Tim, Chief Product and Technology Officer, EduIntelligence
Imagine you’re appointing a new deputy head. Two candidates make the final round.
The first has twenty years in corporate leadership. They’ve read every major book on education, can quote the research on feedback and retrieval practice, and interview brilliantly. The second has spent ten years as a middle leader in schools like yours. They’ve run a department, sat through data drops, chased attendance and written reports for governors.
I’ve asked this question to rooms full of school leaders, and the answer is always the same. You hire the second candidate. Not because the first isn’t clever, but because knowing about education isn’t the same as knowing how a school works.
I think the same question applies to the AI tools our schools now use every day.
I want to be clear about this, because the conversation about AI in schools often turns into a contest. Tools like ChatGPT, Copilot and Gemini are capable and improving fast. Staff use them to draft letters, plan lessons, summarise research and think through problems. That’s a good thing, and I use them myself.
But when the question is about your own school, these tools are in the position of the first candidate. They know a lot about education in general. They don’t know your school, and the way most people use them means someone has to export the data, strip out anything identifying, upload a spreadsheet and explain what every column means. Then they have to do it again next week.
The experienced middle leader brings three things the brilliant outsider doesn’t. Each one matters just as much when the “colleague” is a piece of software.
01
Ask ten schools how they define persistent absence and you’ll get several answers. Your school has its own threshold, its own attendance policy and its own way of counting authorised absence. A general-purpose tool has to work out a definition each time you ask, usually by writing and running a fresh piece of analysis that nobody checks. Ask the same question twice and you may get two slightly different answers. In EduIntelligence, each measure is defined once, in a governed data layer, and calculated the same way whoever asks.
02
When a head asks “how is behaviour?”, they rarely mean the total number of incidents ever recorded. They mean: how has behaviour changed this term, which year groups or students are driving it, and should I be worried? An experienced colleague hears the real question. EduIntelligence is built to do the same. It works out the question behind the question before it goes to the data, then answers the follow-ups you’d probably have asked next.
03
The most dangerous colleague isn’t the one who doesn’t know. It’s the one who is confidently wrong. In a classroom we teach students that it’s fine to say they haven’t understood yet. We should expect the same of the tools we use for decisions about children. If EduIntelligence can’t find the data it needs to answer a question properly, it’s designed to tell you, rather than fill the gap with something plausible.
Watch: EduIntelligence in under two minutes
When I was a head, a good data question could take most of a morning. Pull the report from one system, the attendance from another, check the support register in a spreadsheet, combine it all, work out what it’s saying, and then get interrupted three times by things that couldn’t wait.
How EduIntelligence thinks
We designed EduIntelligence to follow the same thought process an experienced leader would, across the systems a school already uses. It reads your data alongside your own documents, such as policies, improvement plans and support records, and uses your policies as the primary source when it suggests next steps. It’s also aware of who’s asking. An admissions leader and a head of inclusion asking about the same cohort get answers framed for their role, and each only sees the data they’re allowed to see.
The difference is time. A question that once took a morning comes back in minutes rather than hours or days, with the evidence behind the answer, the possible drivers it found and suggested next steps. Those suggestions are exactly that. Deciding what to do is still a leader’s job.
The other thing a good middle leader understands is that student data is not like other data. Most schools rightly tell staff not to paste identifiable student information into general AI tools.
Student data, protected
EduIntelligence was built for exactly this situation. When a request touches student data, names and identifying details are stripped out and replaced with an anonymous identifier before anything reaches the AI model. The model has no route from that identifier back to the student. Names are put back only when the answer is shown to you. No customer data is used to train third-party AI models.
That’s what lets a head of year ask about an individual student’s attendance, attainment and support in one question, which is the kind of question schools most need answered and least want to put into a general tool.
I’m not suggesting anyone stops using general AI. Keep using it for the jobs it does well. But when the question is about your students, your outcomes and your priorities, it’s worth asking which candidate you’d trust with the answer.
You’d hire the person who knows how schools work. It’s reasonable to expect the same of your AI.
Read next: EduIntelligence vs ChatGPT and Copilot: Which Should Your School Use for Data Questions?
See EduIntelligence Answer Your School’s Questions
Book a demo and we’ll show you how EduIntelligence handles the questions your leadership team asks every term.