From “Learning on the Job” to AI-Driven Performance Support  

My proposition is simple, yet far-reaching:

At its core, this means elevating learning on the job by embedding AI-powered performance support directly into people’s workflows.

Think about how employees currently look for help and support

• they search Google

• they browse YouTube

• they dig through SharePoint pages

• they scan internal knowledge bases or LMS catalogues

When you think of it, this process is very very inefficient and time consuming without any guarantee to actually find what you are looking for. People rarely need an entire 20-minute video. They need the two minutes between 16:40 and 18:20. They don’t need a full e-learning module. They need one paragraph, one example, or one visual. Most of the time, far more effort goes into finding and consuming information than into actually applying it.

This is where AI changes everything. AI can dramatically reduce this friction by:

• guiding employees to the right answer immediately

• summarizing and contextualizing information

• filtering out what is irrelevant

• delivering support at the moment of need

From Reactive Support to Proactive Learning

The real power of AI, however, goes beyond efficiency. AI has the potential to move learning from reactive to proactive. Imagine an AI system integrated into a sales platform like Salesforce. Instead of waiting for someone to ask for help, it could detect patterns:

• delays in certain process steps

• repeated errors

• unusual behaviors compared to peers

Based on that signal, the AI might intervene and say:

“It looks like this step is taking longer than usual. Would you like some guidance or examples?”

We are not fully there yet — but this is clearly the direction of travel. Learning becomes embedded, contextual, and almost invisible. It happens because of work, not around work.

AI Takes Over the Full Learning Cycle

What makes this truly different from earlier interpretations of the 70% is that I believe AI will eventually take on the full learning and development cycle — at an individual level. I already made this prediction at the Learning Technologies Conference in London back in 2018 (!).

That includes:

• identifying learning needs, person by person

• designing the appropriate intervention

• generating or curating content

• deploying it through the most effective channel (email, chat, voice, virtual assistant)

• analyzing whether performance actually improved afterward

This is a significant shift. In the traditional view of the 70%, L&D still plays a central role in:

• designing performance support systems

• creating content

• managing structures and processes

In the AI-driven version of the 70%, much of this becomes fully or largely automated.

That does not mean L&D disappears — but it does mean that this part of the learning model will no longer be the primary space where L&D differentiates itself. And that brings us to the remaining 20% and 10%. Because this is where L&D’s future relevance, value, and strategic position truly come into play.

The 20%: AI-Enhanced Social Learning

Let’s continue with the 20%: AI-Enhanced Social Learning

In my view, this part of the model remains remarkably close to the original interpretation: social learning. Humans are, by nature, social learners. We learn constantly from others — often without realizing it. In meetings, in informal conversations, through observation and collaboration. I see this very clearly with my own children: they learn new skills while playing video games with friends, which for them is a deeply social activity. Social learning does not need to be invented. It already exists. What will change, however, is how visible, intentional, and supported it becomes.

AI-Enhanced Social Learning

I believe the next evolution of the 20% will be AI-enhanced social learning. Much of today’s social interaction at work already happens in digital environments:

• online meetings

• one-on-one conversations in Teams

• collaborative work on documents, presentations, and deliverables

Because these interactions are digital, they generate data. We already see this today through automatic transcriptions, summaries, and action lists in tools like Microsoft Teams. This opens up a fascinating opportunity. Imagine finishing a meeting and receiving personal, private feedback from AI:

• What did I do well in this conversation?

• Where could I improve?

• How effective was my communication?

• How inclusive was my behavior?

• What could I try differently next time?

Suddenly, social learning becomes conscious learning. Reflection is no longer dependent on memory or intuition — it is supported by evidence and feedback.

However, introducing AI in social learning as analyst, facilitator, administrator and even coach does mean that like in the ‘70%’ much of the work currently performed by L&D will be taken over. My expectation is that only the best facilitators and coaches (those who demonstrate their added value) will stick around, but in reduced numbers.

Walking the Thin Line: Data, Trust, and Ethics

Now, let me be very clear: this is a sensitive area. AI-enhanced social learning only works if it is built on trust. It requires individuals to explicitly consent to how their data is used. It requires absolute clarity that:

• confidential conversations are not analyzed for performance appraisal

• personal discussions are not used to predict attrition or risk

• insights are used to support the individual, not to control them

If I have a confidential conversation with a manager or colleague, the content of that conversation should never resurface in a way that could harm me. That boundary must be non-negotiable. At the same time, there is room for aggregated, anonymized insights — just as we already use aggregated people data today. When done properly, those insights can help L&D, HR, and leaders make better decisions without compromising individual privacy.

The key message here is simple, if AI-enhanced social learning is part of your strategy, data privacy and security must be one of your explicitly managed risks.

Handled poorly, it will fail immediately. Handled transparently and ethically, it can be incredibly powerful.

The 10%: From Formal Learning to Boutique Learning

That leaves us with the 10%.

This is the part I deliberately no longer call formal learning. Instead, I refer to it as boutique learning. For me, boutique learning is a massive upgrade from traditional course based training. I strongly believe that much of what we currently label as formal learning will actually move into the 70% and be automated through AI. Especially knowledge-heavy generic programs. Take compliance learning as an example. I don’t believe we will have traditional compliance courses in the future. Instead:

• you might receive a prompt at the moment of risk

• you might be presented with a short scenario and asked to respond

• you might receive contextual guidance when handling sensitive data

Compliance learning becomes embedded in daily work — personalized, relevant, and unavoidable. Exactly where AI excels. And if we are honest, a very large portion of today’s learning catalogue — sometimes 70–80% — consists of generic knowledge programs. That generic nature is one of the main reasons why learning impact is often disappointingly low. All of that content belongs in the AI-driven 70%.

Introducing Boutique Learning

So what is boutique learning? I’ve written about this concept before, and I deliberately use the word boutique because of the image it evokes. I’m thinking of the small French bakery competing with large supermarkets. The bread in the supermarket is cheaper. It’s convenient. It’s everywhere. And yet, the boutique bakery continues to exist — and thrive. Why?

• the quality is exceptional

• it fulfills something fundamental

• it is part of a social and cultural ritual

• it has character, history, and craftsmanship

• it offers an experience you simply cannot get in a supermarket

Boutique learning is exactly that. It is:

• high-quality

• deeply contextual

• highly personalized

• focused on strategic and critical skills

• designed as a memorable, often transformational experience

The closest equivalent today is probably top-tier leadership development programs — the kind where senior leaders walk away feeling they’ve had a genuinely life-changing experience.

Why Boutique Learning Is L&D’s Strategic Future

Boutique learning represents a unique opportunity for L&D. It brings the human dimension in an increasingly AI-driven world. It addresses skills that are too nuanced, too strategic, or too rare to ‘teach’ the AI. Sometimes simply because there is insufficient data available. Or the topic is so classified that the organization does not want to take any risks exposing it to an internal large language model (LLM).

Boutique Learning allows L&D excel in what is does best. To clearly articulate L&D’s value where AI cannot replace human judgment, experience, and interaction. I firmly believe that boutique learning will become one of the core businesses of L&D going forward. But there is a condition.

• Boutique learning is more expensive.

• It requires significant investment.

• And therefore, quality must be exceptional.

That also means L&D must be very explicit about:

• what participants gain

• what the organization gains

• why the investment is justified

And that is exactly where evidence comes in. This is where learning analytics becomes not a “nice-to-have”, but an absolute necessity.

Tips for Getting Started

You don’t need to wait for perfect AI, new org charts, or a shiny strategy deck. There are concrete steps you can take today to prepare for the AI-driven 70–20–10.

1. Clean up your learning catalogue (brutally)

If 70–80% of your catalogue is generic knowledge, you’re already carrying technical debt. And you are spending time, effort and money on things that most employees might already be asking ChatGPT. So why waste it?

Why not do the following:

• Label your catalogue honestly:

  • “AI-suitable & generic”

  • “Contextual but reusable”

  • “Strategic & boutique”

• Start sunsetting or freezing low-impact content instead of updating it endlessly.

• Ask one uncomfortable question per program: if AI could answer this better at the moment of need, should this still exist?

2. Make AI-enhanced social learning visible

Social learning already happens. And AI is already supporting it! The challenge is to surface it, make it visible and explicit without breaking trust.

Here’s some of the examples you could look at:

• Start small and ethical: use AI summaries for personal reflection.

• Experiment in area’s where AI might already be involved:

  • AI-supported meeting reflections

  • feedback on communication patterns

  • coaching prompts after collaboration moments

• Reflect on what you consider to be clear red lines, like:

  • no performance appraisal

  • no hidden monitoring

  • no individual data reuse without consent

I cannot stress this enough: trust is the operating system of AI-enhanced social learning.

3. Invest deliberately in Boutique Learning (the new 10%)

Stop spreading your budget thin. Boutique learning only works if it is exceptional.

And you can already start moving towards Boutique Learning tomorrow:

• Identify skills that are strategic, rare, high-risk, deeply contextual. These could be possible topics for Boutique Learning programs

• Reflect on your criteria for Boutique Learning experiences, should they be fully

  human-to-human? Immersive? Challenging? Unforgettable? Or all of these?

• Define success before you start to design your programs:

  • what should change in behavior?

  • what decisions should improve?

  • what performance must move?

👉 Boutique learning is not “premium content”. It’s strategic capability building.

4. Build analytics into design — not after delivery

If you don’t design for data, you limit what AI and analytics can ever do to support. That is why we are always strongly advocating to ‘design for data’. If you want to be able to show evidence that your Boutique Learning programs deliver, you must decide beforehand:

• “What should improve if this works?”

• “How will we see and prove it?”

This will allow you to design the experience in such a way that it’s not just delivering impact, but also generates the data that you need to do the analytics that you want.

👉 Good analytics starts at design, not in dashboards.