Lewie Curry - Tech Founder & Thought Leader

I've just come back from the Inspired Media Summit in London - a rare convergence of three major events (CDO Summit, CMO Summit, and CAIO Summit) all happening in the same venue. Spending two days immersed in conversations with Chief Data Officers, Chief Marketing Officers, and Chief AI Officers gave me a unique vantage point across different functions and industries. And one theme cut through every session, every panel, every corridor conversation: bad data is crippling organisations' ability to do anything meaningful with AI.

Data leaders are frustrated. Marketing leaders are making decisions based on incomplete information. AI leaders are trying to build intelligent systems on foundations that simply aren't solid. And as I listened to these challenges, I kept thinking: we have exactly the same problem in corporate learning and development. Maybe worse.

Because in L&D, we've been measuring the wrong things for so long that we've convinced ourselves our data is fine. Completion rates look good on paper. Quiz scores are easy to report. But none of that tells you whether someone can actually do the thing you trained them to do. And now, as everyone rushes to add AI to their learning platforms, we're about to amplify all those data quality problems at scale.

The Bad Data Crisis

The scale of the data quality problem is staggering. According to recent research, 64% of organisations cite poor data quality as their biggest challenge for 2025, and 67% admit they don't completely trust their own data. Think about that for a moment - two-thirds of organisations are making decisions based on information they don't actually trust.

The financial impact is enormous. The average business loses £15 million annually due to poor data quality. Employees spend 27% of their time - more than a quarter of their working hours - correcting bad data instead of doing their actual jobs. And perhaps most concerning for anyone investing in AI: Gartner predicts that 60% of AI projects will be abandoned by 2026 because organisations lack AI-ready data.

Tim Berners-Lee has been saying for years that "bad data is worse than no data at all," and that warning feels more urgent than ever. When you're making decisions in real-time, using AI, and putting faith in the insights, getting the data quality right isn't optional. Bad data leads to bad decisions, which lead to missed opportunities, wasted resources, and in extreme cases, business failure.

In corporate learning specifically, bad data manifests in a particular way. We track completion rates religiously. We report on quiz scores. We celebrate when 95% of employees finish a training module. But what does any of that actually tell us? It tells us that people clicked through the content and answered some questions. It doesn't tell us if they can apply the skill. It doesn't tell us if behaviour changed. It doesn't tell us if there was any business impact whatsoever.

Most LMS platforms are built to track the wrong things. They're designed to prove that learning happened, not to measure whether learning stuck or translated into capability. And that distinction matters enormously when you start trying to build AI on top of that data.

The AI Training Problem

The bad data problem in L&D creates three interconnected issues when organisations try to implement AI:

First, AI models get trained on poor quality learning data. If your underlying data doesn't track actual engagement - just surface-level completion metrics - AI can't personalise effectively. It's working with proxies, not real signals. Unity Technologies learned this lesson the hard way when they lost roughly £110 million in a single quarter because incorrect data from a significant client corrupted the training sets for their machine learning models. The costs included rebuilding ML models, delays in launching revenue-generating features, and the resulting business impact. Their stock price dropped 37%.

In L&D, this plays out when AI systems try to recommend courses based on completion data rather than retention or application data. The AI might see that everyone who completed Course A went on to complete Course B, and start recommending that pathway. But if nobody who took those courses can actually perform the associated skills, the AI is just optimising for clicking through content, not for capability building.

Second, AI systems try to personalise learning with incomplete data. Most learning management systems track basic metrics: did someone log in, did they complete a module, did they pass a quiz. What they don't track is how engaged people actually were, what they retained, whether they can apply it, where they struggled, what made concepts click. AI personalisation without that granular behavioural data is essentially guessing. It's like trying to be a personal trainer when all you know about someone is that they showed up to the gym - you have no idea what exercises they did, how hard they worked, or what results they're seeing.

Third, organisations are using AI to make L&D decisions when their underlying data infrastructure is fundamentally inadequate. They're making investment decisions based on "85% completion rate" without any idea whether those people can actually do the thing they were trained on. They're using AI analytics to identify skills gaps based on which courses people haven't taken, rather than which capabilities they haven't demonstrated. And because AI can process data at scale, it amplifies these problems. Garbage in, garbage out - but now at enterprise scale, affecting thousands of employees and millions in L&D budget.

What Good Data Actually Looks Like

We recently completed a pilot with a large European enterprise client, and one of the most consistent pieces of positive feedback we received was about our data and analytics capabilities. This surprised us initially - we thought they'd be most excited about the teach-back methodology or the retention results. But the data infrastructure turned out to be just as important.

What impressed them was the granularity and usefulness of what we track. Suada captures every single interaction a user takes within the platform. Not just "completed module 3" but hours of video lessons watched, teach-back attempts and revisions, mobile versus web access patterns, whether they logged in via SSO or email and password, time spent on each concept, where they paused or replayed content, how many attempts it took to successfully record a teach-back video.

This granular interaction tracking creates a completely different data foundation. You can see patterns that would be invisible in traditional LMS data. You can identify exactly where learners struggle before they drop out. You can spot cohorts who are racing through content without retention. You can see which concepts require more support or better explanation.

But the real power comes from being able to spin up custom dashboards quickly for different stakeholders. Operations leaders want to see adoption and engagement metrics. Learning teams want to understand content effectiveness. Business leaders want to see impact on business outcomes. With proper data infrastructure, you can serve all these needs without manual report creation or weeks of data analysis.

Most importantly, good data allows you to map learning outcomes to business outcomes. Not just "95% completed safety training" but "incident rates decreased 40% in facilities where staff completed the training versus control group." Not just "sales team finished product knowledge course" but "win rates improved 15% and deal sizes increased 22% among trained reps compared to baseline."

This is what AI-ready data looks like in L&D. It's granular, behavioural, connected to business metrics, and structured in ways that allow for meaningful analysis. Without this foundation, adding AI is just putting a sophisticated engine on a vehicle with square wheels.

What Becomes Possible

When you have a solid data foundation, genuinely transformative things become possible.

AI can actually work the way it's supposed to. Personalisation based on real engagement signals rather than crude proxies. Adaptive learning paths that respond to how someone actually learns, not just what they've clicked through. Content recommendations based on demonstrated knowledge gaps rather than course completion history.

Predictive analytics that matter. Instead of identifying who's at risk of not completing a course, you can identify who's at risk of not applying the learning. Instead of flagging people who haven't logged in recently, you can spot patterns that indicate someone is going through the motions without genuine understanding. These are fundamentally different and far more valuable insights.

Real-time intervention becomes possible. You can support learners before they fail, not after. When you see someone struggling with a particular concept across multiple attempts, you can trigger additional resources or human support. When you spot a cohort showing lower engagement than previous groups, you can investigate and adjust before completion rates suffer.

Genuine skills mapping emerges. You move from tracking what courses people have taken to understanding what people can actually do. This is the foundation for becoming a truly skills-based organisation - something many companies aspire to but struggle with because their data doesn't support it.

Business impact measurement with confidence. You can connect L&D spend to business outcomes not just with correlation but with genuine causal understanding. You can show leadership that specific learning interventions drove specific business results, backed by data that would satisfy any CDO's scrutiny.

The L&D teams that will win in the next few years are the ones who stop measuring completions and start measuring capability. Who build data infrastructure before adding the AI layer. Who track interactions, not just outcomes. Who make data actionable for multiple stakeholders rather than hoarding it in the L&D function.

The Path Forward

The Inspired Summit made one thing abundantly clear: every Chief Data Officer, Chief Marketing Officer, and Chief AI Officer is wrestling with bad data. It's not a learning-specific problem. It's an enterprise-wide challenge that's holding back transformation across every function.

But in L&D, I'd argue we have the problem worse than most functions. We've been measuring the wrong things for so long that we've built entire systems around metrics that don't matter. Completion rates. Time spent in system. Quiz scores. None of these tell you whether someone can do their job better after training than before.

We also have an enormous opportunity. Most of corporate learning infrastructure needs to be rebuilt anyway. The traditional LMS model doesn't serve modern learning needs. As organisations modernise their L&D technology, they have the chance to build proper data foundations from the ground up rather than trying to retrofit them onto legacy systems.

The question isn't whether AI will transform corporate learning. That's inevitable. The real question is whether your data is good enough to let it. Whether you're tracking the right things. Whether your infrastructure can support the kind of granular, behavioural, outcome-connected data that AI needs to be genuinely useful rather than just an expensive way to automate bad decisions.

The organisations that get their learning data right won't just have better training programmes. They'll have the foundation for genuinely intelligent, adaptive learning systems that can prove their business impact and continuously improve based on what actually works. They'll be able to move from hope-based L&D to evidence-based capability building.

That's the future. But it starts with admitting that most of our current data isn't good enough, and committing to building something better. The summit reminded me that this challenge spans every function. The difference is, some functions are further along in addressing it. L&D needs to catch up.