There is an old line from enterprise technology: “Nobody gets fired for buying IBM.”
It lasted because it captured a truth about institutional decisions. Large organisations rarely buy the most exciting option. They buy the option they can defend: established, legible and unlikely to create an unpleasant surprise at the next board meeting.
Multi-academy trust leaders understand that instinct better than most. A decision made at trust level can touch several schools, thousands of pupils and hundreds of staff. The upside may be shared. The downside will have an owner.
So caution is rational.
But the risk calculation around AI has changed. AI is no longer a proposal waiting in a procurement inbox. It is already entering staff workflows, supplier roadmaps and the world pupils will encounter. Choosing nothing does not preserve the old baseline. It leaves adoption fragmented, invisible and difficult to govern.
That creates the uncomfortable conclusion at the centre of this post:
For a MAT, the riskiest AI strategy may now be to stand still.
The question is not whether to become less careful. It is how to turn caution into controlled learning.
Inaction is still a decision
For years, technology risk was framed around purchase: Will it work? Will people use it? Will it create a data problem? Will the supplier survive?
Those questions still matter. But AI adds a second set of risks that appear when a trust waits:
- Ungoverned adoption risk: staff use general-purpose tools individually, with inconsistent safeguards and no shared standard.
- Visibility risk: leaders cannot see where AI is helping, where practice is changing or where support is needed.
- Consistency risk: individual schools move at different speeds, widening the gap between confident early adopters and colleagues who have had no structured opportunity to learn.
- Implementation risk: the trust eventually makes a rushed decision because events have overtaken its preparation.
- Opportunity cost: workload, professional learning and organisational knowledge improve elsewhere while the trust continues to rely on occasional snapshots.
The UK government’s AI Opportunities Action Plan makes adoption part of the risk calculation. Its proposed rhythm is sensible: scan, pilot, scale. That is not reckless acceleration. It is a way to learn before making a larger commitment.
A risk-conscious MAT should therefore ask two questions of every AI proposal:
- What could go wrong if we adopt this?
- What could go wrong if we do not?
Only the first question appears on many risk registers. The second increasingly belongs there too.
The breakthrough is visibility without surveillance
The most important thing Starlight gives a MAT is not a novelty feature. It is visibility.
Teachers use Starlight to receive private, transcript-grounded feedback on lessons and professional conversations. Leaders do not open those individual reports. Through the Constellation Dashboard, they see anonymised and aggregated patterns across departments, schools and the trust.
That distinction matters.
A trust leader can see themes in curriculum, teaching and achievement; attendance, behaviour and inclusion; personal development and wellbeing; leadership and governance; and staff engagement. The dashboard surfaces shared strengths, priority development areas and strategic actions. It can show how patterns change over time without turning a teacher’s private coaching space into a performance management file.
Starlight makes the organisation more visible while keeping the individual protected.
That is a different model of assurance. It gives leaders earlier signals than an annual survey, richer evidence than a spreadsheet of completed observations and more continuity than a round of school visits can provide on its own.
For a MAT, this can reduce several practical risks at once:
- CPD investment can follow evidence rather than the loudest anecdote.
- Trust-wide priorities can be checked against what is actually appearing in classrooms.
- Schools can share a common language while retaining local context.
- Emerging patterns can prompt support before they become entrenched.
- Board and executive conversations can start with a clearer evidence base.
The point is not to make every leadership decision automatic. It is to make fewer decisions blind.
Deep Field Reports turn a concern into an answerable question
A dashboard gives the wide view. Sometimes leaders need to look closely at one part of the picture.
That is the purpose of Deep Field Reports. They allow a school or trust to define a question and examine the relevant, anonymised evidence across its transcript archive.
A MAT might ask:
- How is a trust-wide focus on questioning appearing across Year 7 lessons?
- What has changed since a professional learning programme began?
- Which inclusive routines appear consistently across schools, and where are the gaps?
- What do coaching conversations reveal about the implementation barriers leaders need to remove?
- Does practice differ by department, phase, session type or time period?
A Deep Field Report focuses the lens on that slice of the evidence. It surfaces patterns and examples without identifying an individual teacher.
This moves the leadership conversation from “we think” to “we can investigate”. It also makes evaluation more useful. Instead of waiting until the end of an initiative to ask whether it worked, leaders can frame a question, look at the evidence, act and return to the question later.
That is risk reduction in its most practical form: a smaller gap between what the trust intends and what it can actually see.
Privacy is an operating principle
Teacher trust is not a communications exercise to complete after implementation. It is part of the product design and the operating model.
Individual Starlight reports remain private to the teacher. Leadership insight is anonymised and aggregated. Recordings and reports are not there to create individual rankings. Starlight’s approach to data, hosting, encryption and deletion is set out more fully in the questions worth asking about the environment and your data.
That separation protects both sides of the system. Teachers need a space where reflection is honest. Leaders need a view of collective strengths and needs. Collapsing those two purposes would weaken the quality of both.
The ICO’s guidance on AI and data protection gives leaders a useful frame for due diligence: lawfulness, fairness, transparency, purpose limitation, data minimisation, accuracy, storage limitation, security and accountability. These are not obstacles to adoption. They are the conditions that make sustained adoption possible.
A MAT considering any AI product should be able to answer, in plain English:
- What data enters the system?
- Who can see the identifiable output?
- What can leaders see?
- Is the data used to train an AI model?
- Where is it stored, and how is it protected?
- How can a user delete it?
- What documentation supports the trust’s DPIA and governance process?
A supplier that cannot make those answers easy is increasing risk, however impressive the demo looks.
Specific. Timely. Actionable. Regular.
Starlight’s standard for useful feedback is simple: it should be specific, timely, actionable and regular. The same four words can test whether leadership intelligence is useful.
- Specific: Does the evidence identify a real pattern, or offer a generic judgement?
- Timely: Does it arrive while leaders can still shape the next decision?
- Actionable: Does it point towards a manageable next step?
- Regular: Can the trust revisit the evidence and see whether the pattern changes?
We have explored why these four words decide whether feedback works. At MAT level, they also distinguish a learning system from a reporting event.
A static dashboard can reassure. A regular evidence loop can improve.
A safer route from interest to scale
The Education Endowment Foundation’s structured implementation process uses four flexible phases: explore, prepare, deliver and sustain. That is a strong model for AI adoption because it treats implementation as a process, not a launch date.
For a MAT, a controlled Starlight adoption could follow this pattern:
- Choose one bounded problem. Start with a real question about feedback, professional learning or visibility that leaders and teachers recognise.
- Set the red lines first. Agree privacy, access, purpose and the explicit separation between coaching and capability processes.
- Pilot with willing participants. Learn from a small number of schools or teams and invite challenge, especially from cautious colleagues.
- Review aggregated evidence. Use Constellation to identify patterns, participation and implementation friction without inspecting individual reports.
- Go deeper where it matters. Use a Deep Field question to test an assumption or evaluate a priority.
- Scale only when the evidence and trust are strong enough. Carry forward what worked and change what did not.
This is what mature risk management looks like in a fast-moving environment. It creates reversibility, learning and clear decision points.
The safest choice is the one that keeps learning
The IBM line belonged to an era when safety came from choosing a familiar supplier and expecting the environment to remain broadly stable.
AI changes that logic. When capability is moving quickly, a static organisation accumulates risk. Skills gaps widen. Unofficial use grows. Evidence disappears into separate tools and private workarounds. By the time the organisation feels ready, it may be making a larger decision with less practical experience.
MAT leaders are right to be risk averse. The mistake is defining safety as immobility.
Starlight offers a more defensible route: private development for teachers; anonymised, aggregated visibility for leaders; focused Deep Field analysis for the questions that matter; and a regular evidence loop that helps the trust learn before it scales.
The breakthrough is not that Starlight uses AI. It is that Starlight can reduce the risk of using AI, and the risk of failing to learn from it.
If you are reviewing AI, professional learning or trust-wide visibility, book a Starlight demo. Bring the difficult questions. They are the right place to start.
Spark Insight with Starlight, and make the next decision with evidence.
The Insight Engine is written by Adam Sturdee, co-founder of Starlight, the UK’s teacher-first AI-powered coaching platform, and a senior leader with responsibility for teaching, learning and coaching. This blog is part of a wider mission to support educators through meaningful reflection, not performance metrics. It documents the journey of building Starlight from the ground up, and explores how AI, when shaped with care, can reduce workload, surface insight and help teachers think more deeply about their practice. Rooted in the belief that growth should be private, professional and purposeful, The Insight Engine offers ideas and stories that put insight, not judgement, at the centre of development.
