The UK currently invests over £3 billion a year in apprenticeships. It is a massive commitment to the future of the workforce. Yet by most expert estimates, at least 30% of that investment - nearly £1 billion – doesn’t reach its intended outcome of learner achievement.
That £1 billion isn't just a rounding error; it’s a structural failure at a significant scale. For years, the sector has treated this 30% gap as an inevitable part of the landscape.
It isn't inevitable, but to close that 30% gap we do have to be honest about a limitation that runs through almost every platform and process in the skills sector: our information infrastructure is built to tell us what happened after the fact, rarely in time to change it.
For decades, the dominant model in funded learning has been built around one word: compliance. The focus has been to record what happened, evidence what was done, and prove that the rules were followed.
This model made sense for a long time - when tools were limited to spreadsheets and manual reporting. In that world, the best a provider could do was review the "black box" of the previous month and try to learn for the future. But as John Ingram, CEO of Bud Systems, noted at the AELP National Conference, compliance should be considered the floor, not the ceiling. It’s essential, but insufficient on its own.
Compliance can tell you that a learner withdrew today. It cannot tell you that the withdrawal was predictable six weeks ago. It can show that a gateway was failed, but it cannot show the knowledge gap that was already visible in month three. It can confirm an employer has pulled back, but it cannot tell you that the relationship had been cooling for sixty days before anyone noticed.
Doing that requires us to move beyond simply proving that the correct processes took place and toward identifying the patterns that influence learner outcomes. Continuing to demand rigour around funding and evidence while extracting far more value from the data already being created – and this presents us with a data architecture challenge.
The conversation about AI in the skills sector has moved quickly in the last eighteen months, with most of it focused on capability: everyone is asking what AI can do and how it works. But there is a really important prior question that will determine whether or not an AI deployment can actually deliver on its promise: What data is the AI working from?
AI systems learn patterns from the data they are trained on. In a sector where most platforms have been built for flexibility (highly customisable, adaptable to each provider's own processes and terminology) the data they accumulate tends to be inconsistently structured and difficult to interrogate at scale.
This creates "noisy" data: a "progress note" at one provider means something entirely different at another. An engagement score is calculated differently depending on which fields someone chose to configure. A risk flag was set by one tutor's judgement call and not another's.
When data is inconsistently structured and fragmented, AI inherits those limitations. It can process what it is given, but it cannot compensate for what is missing or incomparable. If your data is a mess, your AI’s insights will be too.
The Bud advantage: eight years of structured data that feeds intelligence
This is where the distinction between "reporting" and "intelligence" becomes clear. Since 2017, Bud has made a deliberate, and sometimes controversial, architectural choice: to build a platform with a defined, consistent structure rather than infinite customisability.
That choice has produced something genuinely rare in the skills sector: a dataset that is clean, deep, and comparable across hundreds of providers and eight years of longitudinal learner journeys. Bud customers benefit from data that is structured consistently, comparable across providers, and tracked across the full arc of a learner's programme, rather than captured as point-in-time snapshots.
|
Noisy & fragmented data |
Longitudinal & clean data |
|
Customizable by design: Data structures vary by provider, tutor, and programme. |
Consistent by design: Standardized structure across 8+ years and 100s of providers. |
|
Reporting focus: Useful for looking back at what happened within one organization. |
Predictive focus: Patterns in one context can be validated against thousands of others. |
|
Fragmented signals: Hard to identify trends across different learner cohorts. |
Clear trajectories: Patterns in engagement turn monitoring into actual prediction. |
Because Bud’s data is comparable, it allows the system to ask a different kind of question. When a learner’s engagement pattern shifts, we don’t just record it. We ask: "Across the thousands of learners who showed this exact pattern at this exact stage of their programme, what was the outcome?"
Frankly, that’s not a question that other platforms can answer.
Predictive risk profiling: Early identification, not late confirmation. So rather than a withdrawal appearing in a report after the fact, we see a risk profile building over weeks: visible, and most importantly actionable before the point of no return.
Cross-programme pattern recognition: The system identifying that a specific combination of signals (e.g. shorter check-in responses, a dip in submission frequency, increased delays in employer feedback) correlates with a specific risk of failure.
Meaningful benchmarks: Tutors can see a learner's trajectory in the context of what "typical" looks like for that specific programme, providing reference points that would typically have taken a human decades of experience to build.
John framed Bud’s ambition at AELP National Conference: "Most platforms in this sector tell you what has already happened. Bud's ambition is to tell you what is about to happen — and give you the tools to act on it."
Bud’s ambition is honest about where it currently sits. The data foundation is real and it is deep. The intelligence layer being built on top of it is in active development. What exists today is the architecture that makes the ambition credible - not a finished product, but a genuine foundation that few, if any, other platforms in this sector are in a position to replicate, because the data that would power it was never structured in the right way.
We are at a convergence point. A regulated, funded sector demanding more accountability for learner outcomes, and AI capability that has crossed the threshold for pattern recognition at scale.
Providers on platforms that support outcomes intelligence; who can intervene earlier and evidence their impact more credibly, won’t be making marginal improvements on current practice. They’ll be closing that 30% gap of apprenticeship investment that currently doesn’t lead to successful outcomes.
The question of how your data is structured isn't a back-office concern anymore. It is the central question that determines whether you can support your learners today, or if you'll only be able to tell them what went wrong tomorrow.