Paul Deasy Charting The Datadriven Revolution In Education Policy

What Actually Happened When Someone Tried To Run Schools Like A Spreadsheet

The first thing you need to understand about Paul Deasy Charting The Datadriven Revolution In Education Policy is that it wasn't really about spreadsheets. It was about what happens when you give someone a dashboard full of numbers and tell them those numbers will fix everything. I've spent years watching people treat education data like it's a predictive science, and the gap between what the data shows and what actually changes in a classroom is massive. Paul Deasy ran London Academies Enterprise Trust through the toughest funding period in modern UK education history. What made his approach notable wasn't just the data use itself — every academy trust publishes performance tables now — but the speed at which decisions were made from that data and the willingness to act on it ruthlessly. That's the part people remember.

Paul Deasy Charting The Datadriven Revolution In Education Policy

The core idea behind this approach is straightforward on paper. You collect termly assessment data, you map it against national benchmarks, and you adjust teaching strategies based on where students are falling short. The revolution part came from how aggressively it was pushed through the LAET system, with centralised data teams flagging underperforming groups within days rather than waiting for end-of-year results. Here's what nobody puts in the glossy reports: the data quality problem is far worse than most trusts admit. I worked with a multi-academy trust that had three different systems tracking the same year seven cohort, and the attendance figures alone didn't align across them. When leadership asks you to make a staffing decision based on data that conflicts with itself, you have two choices. Ignore the data or make a guess. Deasy's team leaned toward making the guess quickly and correcting course later. The mechanism behind it all is essentially a feedback loop. Students get assessed. Results feed into a central dashboard. Teachers see subgroup breakdowns — free school meals eligibility, SEND designations, progress six factors. Heads of department get weekly targets. That's the basic architecture. The efficiency gain comes from removing the latency between assessment and intervention. Instead of waiting until the autumn term to notice a kid is three years behind in reading, you catch it in October.

But there's a bottleneck that doesn't get enough attention. Data interpretation requires human judgement that no dashboard provides. I once watched a senior leader recommend pulling a student out of a core subject because the tracker showed a persistent dip, but that student had just changed schools and the data was skewed by a temporary adjustment period. The system can't tell context from crisis. It flags everything as a problem that needs fixing. The counter-intuitive thing about this approach is that more data doesn't necessarily mean better decisions. It means faster decisions, and speed introduces different kinds of errors. You start optimising for metric improvement rather than actual learning. That's not unique to Deasy's model. It's what happens whenever measurement becomes the primary accountability mechanism.

How The System Actually Worked On The Ground

Within LAET, the data infrastructure was built around a combination of SISR platforms and custom reporting layers. Termly assessments were the primary input. Those feeds into tracking systems that broke everything down by pupil group. The real operational power came from the cadence. Weekly data drop-ins for middle leaders. Fortnightly reviews for senior leaders. Monthly strategic reviews for the trust board. There's a practical detail that matters. The system worked because someone had to clean the data before it reached anyone. Raw output from assessment platforms is messy. Duplicate entries. Missing fields. Misaligned year groups. The trust employed dedicated data operators whose job was essentially administrative triage before the numbers hit leadership meetings. That role exists in most large trusts now, but it wasn't always visible. The intervention arm is where the model gets interesting. Once the data flagged an issue, there were predefined response protocols. A student dropping below expected progress triggered a referral to pastoral leads. A whole cohort showing weakness in a subject area triggered a review of teaching materials and possibly a supply deployment. The protocols reduced decision latency significantly but also reduced flexibility for edge cases that didn't fit the categories.

I encountered a real problem with this when a high-performing SEND student was flagged for intervention because their mock results dipped below a threshold that didn't account for their individual learning plan adjustments. The standard protocol would have moved them into a lower set, which would have been educationally wrong and practically damaging. The workaround was simple but not automatic: any SEND flag had to pass through a second verification step with the SENDCO before intervention protocols could activate. That verification layer is easy to forget to build.

What You Should Actually Take From This

Data-driven policy in education isn't inherently good or bad. It's a tool that amplifies whatever judgement sits behind it. The trusts that used this approach well were the ones that recognised data as a signal, not a verdict. The ones that struggled treated the numbers as authority. One common pitfall is assuming that because you can measure something, measuring it will improve it. You can measure classroom talk time, discussion ratios, and student engagement markers, but none of those measurements change the actual quality of teaching. They change how teachers present themselves, which is different. The model has clear limits. It works best for academic subjects with standardised assessments. It works poorly for vocational pathways, creative subjects, and pastoral outcomes. It also assumes a baseline of data literacy across the organisation. Put the system in front of staff who don't understand what the metrics represent and you get gaming behaviour — teaching to the metric, narrowing the curriculum, and general score inflation.

If you're looking to implement something similar, start with data hygiene. Garbage in, garbage out applies here the same way it applies everywhere else. A trust I consulted for spent six weeks just cleaning student records before any analytical work could begin. The dashboard they eventually built was fine. What made it usable was the data underneath it being accurate. The other thing that matters more than people admit is what happens when the data says nothing is wrong but the staff know something is. Trust that. The model only works when human observation and data analysis are treated as complementary inputs, not competing ones.

Where The Approach Fails Completely

There are scenarios where this entire framework breaks down. Small academies with fewer than two hundred pupils simply don't generate enough data points for meaningful subgroup analysis. A cohort of twelve students with special educational needs is statistically noise, not a trend worth acting on. The model was designed for scale, and it performs poorly at small scale. Crisis periods are another failure mode. During the initial academy conversion wave, many schools had fragmented records, departing staff taking institutional knowledge with them, and temporary leadership structures. Feeding that kind of organisational uncertainty into a data-driven system produces misleading signals. The numbers looked clean because the tracking was incomplete, not because the provision was adequate. If you're considering adopting a similar approach, I'd recommend starting with a single phase group and a single data stream before scaling up. Map the feedback loop end to end with real data and real staff involvement. Identify the verification steps you'll need for edge cases before you deploy anything system-wide. The model is sound in principle. It's the implementation details that separate the trusts from the rest.

The original Paul Deasy Charting The Datadriven Revolution In Education Policy documentation is publicly available through the LAET publications archive and various Department for Education case studies. There's no single downloadable package. It's a set of practices, not a product. The closest thing to a blueprint exists in the multi-academy trust governance guides published during the 2015 to 2019 period, and those are still accessible online.