Ask any L&D leader what they want from AI and personalization comes up fast. The vision is clear: an employee gets the content they need, when they need it, shaped around their role, their gaps, their pace. No more one-size-fits-all courses. No more irrelevant modules sitting between a learner and the thing they actually need to know.
It’s a reasonable goal. AI can, in principle, deliver it. The issue is that most L&D teams are trying to get there before the groundwork exists to support it.
The problem isn’t the AI. It’s what you’re feeding it.
The structure problem nobody talks about
Most organizations sitting on a library of 200 courses think they have the raw material for personalization. They don’t, at least not yet.
Long, linear courses built as standalone experiences are hard for AI to work with. They weren’t designed to be pulled apart and recombined. There’s no clean separation between concepts, no consistent metadata, no modular logic. Asking AI to personalize from that foundation is like asking someone to rearrange a book by cutting out individual sentences.
Modular content works differently. Each unit is self-contained, built around a single concept or skill. Think of it as the difference between a stack of index cards and a bound book. The index cards can be shuffled, skipped, or reorganized based on what a learner needs. The book can’t. Twenty modular blocks are far more useful for personalization than 200 courses, even if those courses contain more total content.
The content structure isn’t a technical detail. It determines whether personalization is possible at all.
What has to come first
Most L&D teams think of AI adoption as adding tools to their existing workflow. It’s more useful to think of it as building a connected system, where each stage feeds the next and the whole thing can run at scale without constant manual intervention. That’s what orchestration means.
Before personalization, three parts of that system need to be working.
- Getting content ready for AI.This means putting your content into a state where AI can reliably find and use it. That includes consistent metadata, structured tagging, and version control. Without it, AI pulls from conflicting sources and produces inconsistent outputs. Garbage in, garbage out.
- A coordinated production workflow.Most L&D teams use several AI tools that don’t talk to each other. One for text generation, another for video, another for translation. Each produces output in its own format, with its own context, and someone manually bridges the gaps. A connected production workflow replaces that with automated steps from source content through to finished output.
- Quality and human oversight built into the process. AI-generated content needs traceability, not just review. That means knowing which source material each output came from, where the AI was uncertain, and where a human needs to make the call. Being able to show how and why content was generated is what makes it trustworthy enough to publish at volume.
Personalization only works once those three things are in place.
Why most organizations skip straight to the end
The pressure to show AI results is real. Leadership wants speed, and personalization is a compelling thing to promise. So teams try to get there before the foundation is ready, and then wonder why the outputs aren’t good enough to use, or why it works in a pilot but falls apart at scale.
The pattern comes up repeatedly: AI tools get adopted faster than the production system gets redesigned around them. The tools work. The system around them doesn’t.
What the shift actually looks like
When these foundations are in place, the efficiency gains follow. In our work with a pharma client on their clinical operations training, consolidating and structuring source content – a task that previously took a subject matter expert two days – became a fraction of that effort. That time shifted to work that actually required human judgment.
But the more important shift is structural. It means moving from producing training as a series of finished, standalone objects to building modular content that can be recombined, updated, and personalized as needs change.
That’s a different way of thinking about what L&D produces. And it matters more than which AI tools you choose.
We cover this in detail – including a full breakdown of how the system fits together – in this on-demand webinar.Click Here to Watch the full session on YouTube
