We Built the AI-Powered Training Platform Everyone Wanted. Thirteen Months Later, We Took It Apart.

Nobody likes saying out loud that they got it wrong. It’s worse when the wrong thing cost real money and was, on paper, exactly what everyone had been asking for.

When COVID hit, Learning Dimensions Network didn’t really get a choice. Rooms emptied out overnight. We’d been delivering face-to-face training for almost three decades, and suddenly that had to change fast. It had to look like everything else that was moving online right then: self-paced, click-through-on-your-own-time, built to scale.

So we built it. A pretty sophisticated platform in the end: behavioural simulations, adaptive pathways, the kind of thing that gets an instructional designer genuinely excited once the room isn’t a constraint anymore. It ticked the boxes. It looked like it should work.

The Numbers Were Fine. The Behaviour Wasn’t.

Here’s the part I still find uncomfortable to say. Pilot feedback was good. People finished the modules. If you’d only looked at a dashboard, you’d have called it a success and moved on.

But completion was never actually what we were being paid for. We create safety and leadership training because we want someone doing their job differently on a Monday, not because a certificate is issued. Thirteen months in, with full-time staff working on the project, we still couldn’t ignore what the numbers were actually telling us: when people attend work-sponsored training, it needs to change their skills and knowledge, not just get them through to a certificate while they go back and do the same thing. Beta testers who’d sat through an entire digital safety leadership module were still making the same calls on the floor. Something had gone missing in the move online. It took us longer than it should have to actually go looking for what.

What we found, more or less, was that the platform had turned into the exact thing it was supposed to stop happening. It became a box-ticking exercise. The tech itself wasn’t bad. It was doing a job that used to belong to a person in a room, a conversation, a group of other learners bouncing ideas around, and that job just doesn’t survive the move to a screen in one piece.

Why the Room Was Never the Slow Part

Most of my career has been built around one idea: that people don’t all learn the same way, and that ignoring this is exactly where corporate training quietly falls apart. It’s the whole point of the 4MAT approach our programs are built on, a methodology I learned directly from its creator, Dr Bernice McCarthy, and have spent years refining with our own facilitators. Some people need to understand why something matters at a deep level before they’ll engage with it at all. Others want the facts up front. Some need to physically try the thing. Others want to work it out on their own terms and evolve the concept or skill themselves. A self-paced module has to pick a lane and hope for the best. A good facilitator standing in front of a room reads all of that in real time and shifts accordingly.

That sounds like a nice-to-have until you actually watch what happens without it. Online and self-paced, someone who’s confused has two options: push through anyway, or quietly switch off. Neither one changes what they do at work afterwards. In a room, it’s different. Someone gets stuck, admits it, and half the time somebody else in the group was thinking the exact same thing and just hadn’t said anything yet. The facilitator catches it on the spot and explains it a different way before it becomes a gap nobody notices until it’s too late. That back and forth isn’t decoration on top of the content. For a lot of adult learners, it’s basically how anything actually sinks in.

We didn’t lose sight of that by accident. We lost it because the platform we ended up building was a genuinely good answer to a question we hadn’t pushed back on hard enough: how do you deliver training online? That’s a different question to the one we’d spent thirty years actually answering, which is how do people change what they do at work? Turns out those aren’t the same project at all.

Why This Should Worry Every Training Leader Right Now

I would have left this lesson from COVID in the past, as a temporary distraction, if we weren’t seeing the same risk emerge again today with AI.

AI training tools can do far more than the online platforms we were working with in 2020. They can adapt content, run conversations, respond to learner gaps and create much more individualised learning experiences. That is exactly why organisations need to be clear about what they are trying to achieve before they choose the technology.

The mistake we made was not using technology. It was expecting a platform built for scale and delivery to create behaviour change without first being clear about the behaviour we needed to change.

There is also something a skilled facilitator brings that is harder to build into a platform: judgement about what is happening between people. AI can detect hesitation, adapt a conversation and respond to an individual learner. A facilitator is working with the whole room. They can recognise when the issue is not a gap in knowledge, but disagreement, uncertainty, hierarchy, experience or something happening between the people in the group. Then they can decide what the learning needs are in that moment.

AI does not remove the need to think about any of this. If anything, it can make the problem easier to overlook because the technology is more capable and the experience can look more impressive.

Before choosing a platform, the question should be simple: what do we need people to do differently at work, and will this technology actually help us get there?

Where We Landed

We realised the answer wasn’t to abandon technology completely. The mistake was expecting technology to replace the parts of learning that needed human interaction. Live, facilitated learning, whether that’s in a room or genuinely interactive online, remains the foundation because that’s where behaviour actually shifts. Technology sits under that now instead of replacing it. It handles reinforcement. It tracks progress. It gives a facilitator a heads-up on where a group’s actually struggling before they even walk in the door.

What that thirteen-month detour taught us wasn’t “don’t use technology.” It was that technology has to serve how people actually learn instead of the other way round. Build a training program around what’s easy to roll out at scale rather than what genuinely shifts behaviour, and you’ll hit every metric on your dashboard while missing the only thing that was ever supposed to matter.

If someone’s currently pitching your organisation on AI-driven training as the fix for your capacity problem, ask the harder question before signing off on the budget. Is this actually going to change what your people do, or is it just going to prove, very efficiently, that they logged in?

Melissa Williams is CEO and Lead Instructional Designer at Learning Dimensions Network (LDN), and one of only four people worldwide to hold 4MAT® Master Certification. In 2026 she was named International Learning & Training Professional of the Year at the Stevie Awards for Great Employers, her second consecutive international Gold Stevie.

Website:  https://ldn.com.au/
Linkedin: https://www.linkedin.com/in/melissawilliamsldn/

Image Courtesy of Melissa Williams