Coaching, not compliance: what I told BusinessCloud about the feedback problem in schools

BusinessCloud has published a profile of the work we are doing at STAR21 as part of its EdTech Founders series. You can read it here: https://businesscloud.co.uk/news/meet-the-assistant-head-building-a-coaching-platform-for-teachers/

Interviews are a useful discipline. A journalist asks you to explain, in plain terms and to a general audience, what you are doing and why, and you discover quite quickly which parts of your thinking are solid and which you have been getting away with. Most of what I said will be familiar to anyone who has read this blog for a while, but a few threads are worth pulling on properly, and there are one or two things the interview did not have room for.

The problem is structural, not personal

Every school leader I know wants to give teachers better feedback. Every teacher I know wants to receive it. And yet, in most schools, a teacher will teach the best part of 800 lessons in a year and have a colleague in the room for perhaps three of them, with a conversation afterwards if the timetable allows.

That is not a failure of will. It is a failure of scale. Observation and coaching are among the most labour-intensive things a school does, and the people who do them are the same people running departments, year groups and the timetable. I have sat on both sides of this: as the leader responsible for teaching and learning, trying to reach every teacher with something specific and useful, and as the teacher whose developmental feedback arrived once or twice a year, if that. We have known for a long time that feedback which is specific, timely, actionable and regular is the engine of professional growth. The difficulty has never been knowing it. It has been doing it, for everyone, every week.

That is the gap Starlight was built to close.

Coaching, not compliance

The first thing most teachers ask when they hear AI and lesson recording in the same sentence is whether this is surveillance. It is the right question, and it deserves a proper answer rather than reassurance.

Our answer is built into the design rather than into the terms of service. The teacher records their own lesson, on their own device, and chooses what to upload. The transcript and the coaching report that come back are private to them. Leaders see anonymised, aggregated trends across the school, which is exactly what they need to plan professional learning, and never a named individual report, which is exactly what they do not need and should not have. School data is never used to train foundation AI models, and we do not train models of our own. The line we hold ourselves to is nine words long: coaching, not compliance; insight, not surveillance; growth, not grading.

I would add something the interview did not have space for. The surveillance worry is not really about technology. Teachers have been observed for decades; what they distrust is judgement arriving from above, based on a snapshot, with consequences attached. Remove the judgement and the consequences, give the teacher ownership of the evidence, and the same activity becomes something quite different: reflection with a very good memory.

Why the whole lesson matters

A twenty-minute observation is a snapshot, and a snapshot can mislead. Whole-lesson audio shows how an explanation lands, whether questioning deepens or stalls, how routines hold as the hour wears on, and who is actually participating. Starlight turns the recording into a transcript, reads it against the focus the teacher has chosen and the school's professional learning priorities, and returns a structured report within minutes: strengths evidenced in the transcript, a small number of high-leverage next steps, and prompts for reflection.

Two things about that matter to me as a school leader. First, the evidence is the teacher's own words and their pupils' words, not an observer's impression of them. Second, "within minutes" changes the rhythm. Feedback that arrives while the lesson is still fresh gets acted on. Feedback that arrives three weeks later gets filed.

What other systems are teaching me

Starlight is now used by more than 700 teachers in more than 30 schools across the UK, Switzerland and China, with distribution partners in Hong Kong, the Netherlands and West Africa. A Catholic secondary school in Cardiff moved from a small pilot to whole-school rollout in June. An international school in Shanghai runs a single recording through two languages. At my school, it supports a regular, teacher-led coaching rhythm alongside the human coaching we have developed over several years.

The UK system is the one I know best and the only one I have taught in. But looking closely at others has been instructive. What stands out abroad is not better technology but a different posture towards it: less fear, less bureaucracy standing between a good idea and a decision to try it, and a level of optimism about AI in Asia that is markedly higher than here. I do not think that optimism is naive. I think it is a mindset we could borrow, carefully, in the service of better outcomes for children.

Why I have not stopped being a teacher

I am asked, reasonably often, how an assistant headteacher builds a company. The short answer is that the two roles keep each other honest. School gives Starlight a constant reality check: I see what teachers will trust, what leaders can implement on a wet Tuesday in November, and what reduces workload rather than adds to it. That feedback loop shapes what we build and, just as importantly, what we decide not to build. Being a serving school leader is not something I work around; it is a large part of why the product remains grounded.

Where this is going

Our working ambition is to grow from 700 teachers to at least 3,000 by July 2027, and we are far more interested in active use, renewal and demonstrated value than in a headline school count. The larger ambition is for transcript-based reflection to become a normal part of professional learning: as familiar as lesson observation, but more frequent, more private and owned by the teacher. Beyond the UK, I think this technology has exceptional potential where great teaching is scarcest, in emerging markets and across the global South.

If you lead teaching and learning and any of this sounds like your school, the simplest way to test the idea is to try it. Starlight's Launchpad trial is free: five teachers, four weeks, and all anyone needs is something to record audio on. You can start at https://starlightmentor.com/launchpad or, if you would rather see it first, you can book a demo at https://starlightmentor.com/demo-request and we will walk you through it. We will also be at the Schools and Academies Show at the NEC in November, and I would be glad to talk it through in person.

And do read the BusinessCloud piece. It ends, as most conversations with me eventually do, on a boat. As I told them, the wind and the tide remain completely unimpressed by software, which is a useful thing for anyone building an AI company to remember.