AI-upskilling

I’m writing this from what I half-jokingly call my AI sabbatical, an eye-opening deep dive into just how quickly this technology is moving and the capabilities it can unlock.

Since leaving Sanofi in late 2024, I’ve been building phoque.ai, a new venture, and in the process I’ve gone hands-on with AI in a way that would have been almost unimaginable even two years ago. I’ve been writing about what I’m learning as I go in my blog. And I’m working more hours than I ever did in corporate life, entirely by choice, because the speed at which things are moving is genuinely breathtaking.

What I’ve come to believe is this: the greatest value from AI does not come from the technology on its own. It comes from the overlap between deep subject matter expertise, built over years, and AI expertise, which can now be built in months. That overlap is where the real value sits. And it is where I would urge every organisation to focus.

At a simple level, AI gives us two new superpowers when it is used well.

The first is capacity: the ability to do far more, far faster, with AI as a sounding board, research assistant, thought partner, and friction-remover. It reduces the admin burden that clogs up so much of modern work.
The second is capability: access to expert-level information, analysis, pattern recognition, and synthesis that most people would not normally be able to find, produce, or connect quickly on their own.

With the caveat, always, that it has to be verified. AI hallucinates. It is powerful, but it is not infallible. Trust, but verify.

And the evidence is now overwhelming that this is not just a nice theory. Slack’s 2025 Workforce Index, based on more than 5,000 desk workers globally, found that people who use AI daily report 64% higher productivity, 58% better focus, and 81% greater job satisfaction than colleagues who do not. Daily AI use more than doubled in six months. The productivity promise is beginning to pay off, but only for people who have made AI part of their daily rhythm, not an occasional experiment.

The catch is that both of those gifts, capacity and capability, are only as powerful as the person wielding them. An executive who does not understand what AI can do will never ask the right questions. A frontline manager who has never been given permission to experiment will never discover the use case that changes their team’s work. The technology is ready. The more important question is whether the people are.

If you listen to the headlines, the answer seems obvious. The problem, we’re told, is the AI talent war. The scramble for data scientists, machine learning engineers, and prompt engineers. Every conference keynote, consulting deck, and LinkedIn post seems to repeat the same message: hire faster, pay more, build a centre of excellence.

They are not wrong. But they are incomplete.

Because focusing only on standalone AI capability misses the biggest opportunity.

After 25 years leading transformations at Nike, Shell, and Sanofi, and after designing and delivering one of the most ambitious AI upskilling programmes in the pharmaceutical industry, I can tell you where the real AI skills gap sits. It is not primarily in the data science team. It sits with senior leaders: the people approving AI budgets they often do not fully understand, delegating “digital” as if it belongs to someone else, and sponsoring transformation programmes without really grasping what they are sponsoring.

Sanofi’s CEO Paul Hudson put it bluntly in a Semafor interview. Asked about leaders who hand off the execution of AI strategy to their Chief Digital Officer, he said: “You don’t delegate the revolution.”

That line matters because it captures the central leadership truth of this moment. The companies winning with AI are not necessarily the ones with the best algorithms. They are the ones where business leaders own the AI agenda. And that distinction changes everything about how you should design an upskilling strategy.

McKinsey’s 2025 research makes the problem plain. Nearly half of C-suite leaders say their organisations are developing AI tools too slowly, and the number one reason they cite is skill gaps. At the same time, 92% say they plan to increase AI spending, yet only 1% describe their organisations as anything close to mature in AI deployment.

That is the gap in one statistic: buying AI is easy; knowing what to do with it is harder.

Meanwhile, employees are already further ahead than many leaders realise. McKinsey found that workers are using gen AI more deeply than their bosses imagine, with 47% saying it could handle more than 30% of their daily tasks within a year. Microsoft and LinkedIn found the same pattern from another angle: 75% of knowledge workers already use AI at work, and 78% of AI users are bringing their own tools without formal organisational sanction.

In other words, the workforce is moving. Leadership is still catching up.

There is another irony here. Slack’s research shows that executives are actually the heaviest personal users of AI. Forty-three percent use it daily, compared with 35% of senior managers, 23% of middle managers, and just 10% of individual contributors. So the C-suite is often using AI to boost its own productivity, while failing to create the conditions for the rest of the organisation to follow. Half of workers say AI use is not explicitly encouraged by their company, and 29% say they have received no formal guidance at all. Workers in companies that actively promote AI use are nearly three times more likely to become power users.

Permission matters. Education matters. Encouragement matters.

And in pharma, where I spent my last corporate chapter, the picture is even starker. A ZS survey of 115 technology executives across pharma and biotech found that only 40% of AI pilots make it to scaled deployment. Technology and data capabilities are under strain, yes, but so are talent, skills, and business engagement. Sixty-eight percent cite weak data quality and governance as the main reason initiatives fail.

But notice where the real bottleneck is. It is not simply that companies lack data scientists. It is that business leaders often cannot articulate what they need AI to do, cannot distinguish a viable use case from a buzzword, and cannot create the organisational conditions for AI to move from pilot to production. The leadership gap creates a vacuum that no amount of technical hiring can fill.

This Was Never About a One-Off Programme

That realisation is what took me back to 2023, and to the most ambitious AI upskilling programme I have ever designed.

But to understand why it worked, a little context matters.

What we built at Sanofi was not a one-off initiative. It sat inside a deliberate annual rhythm that I had been refining for years.

At Nike, I had led enterprise learning as Chief Learning Officer, where I created NikeU, the company’s first centralised Corporate University. Later, I led HR for Nike Digital, reporting to Nike’s first-ever Chief Digital Officer, at a time when the company was trying to integrate and transform fragmented digital capabilities into something coherent and commercially powerful. Between those two roles, I got to see both sides of the same problem: how you build capability at scale, and how you support transformation when technology is changing faster than the organisation around it.

By the time I arrived at Sanofi as Group Head of Organisational Capability & Transformation, with enterprise learning in my remit, I had become convinced of one core principle: each year, choose one enterprise capability that matters most, build it from the top to the bottom of the organisation, and deliver it differently at each level.

The year before our AI focus, still in the aftermath of COVID, we had done exactly that around wellbeing and what we called “Well Working,” in partnership with MindGym. The top 150 experienced it one way. The next layer got a scaled version. Everyone else got access to foundational resources. Same capability. Same language. Different depth.

There is something powerful about that rhythm. When an entire organisation, from ExCom to frontline, is investing in the same capability in the same year, it creates a shared focus that cuts through noise. It stops being a programme in one function’s budget and becomes an enterprise commitment.

In 2023, in close partnership with Sanofi’s Chief Digital Officer, the choice was obvious. The focus would be AI, digital, and data.

Why We Started at the Top

The context mattered. Sanofi had set itself an audacious goal: to become the first biopharma company powered by AI at scale. Paul Hudson had been relentless in communicating that ambition. The company had already launched plai, an AI-powered decision platform rolled out at scale in June 2023, and Digital Accelerators were being established across R&D, commercial, and manufacturing. It would go on to deepen that ambition further with the Formation Bio and OpenAI collaboration announced in 2024.

So the ambition was clear. The strategy was clear. But the leadership population that needed to drive the transformation did not yet have a shared language, a common framework, or enough confidence to lead it well. That was the opportunity.

The top 150 executives were approving AI budgets and sponsoring digital initiatives, but many were still delegating the thinking along with the doing. And in a company of more than 90,000 people, transformation only cascades when senior leaders genuinely understand it and own it. Otherwise it stalls.

I could see this tension from another vantage point too. I was part of what Hudson later described as his “AI Fight Club,” a cross-functional group with one person from each major function brought together to challenge and disrupt how the company used AI. I represented People & Culture. It was a fascinating place to sit: close enough to the technology strategy to see where it was going, and close enough to the people agenda to see what was missing.

So we made a deliberate decision. We did not start with the data scientists. We did not start with an e-learning module for everyone. We started with the people who set direction for the entire company.

Why This Was Especially Hard in Pharma

Pharma makes this challenge particularly instructive because it combines three pressures that many other industries will increasingly face too.

The first is regulatory caution bleeding into everything. In pharma, the compliance mindset is necessary and justified in areas tied directly to patient safety. But over time, that mindset can spread far beyond its proper domain. Controls designed for drug development or manufacturing start being copied onto lower-risk activities where they do not belong. The instinct to seek pre-approval for everything makes experimentation feel dangerous. Moving from that mindset to one where leaders can say, “let’s test this AI application on a low-risk process and learn,” takes cultural work, not just technical deployment.
The second is the difference between genuine complexity and inherited complication. I wrote about this in my Volcano case study on simplification, and the same logic applies here. A clinical trial protocol may genuinely require sophisticated AI governance. An AI tool that helps a commercial team prioritise daily calls probably does not require the same level of oversight. Teaching leaders to distinguish the two is not a technical skill. It is a leadership skill.
The third is that the upside is enormous. Strategy& has estimated that pharma companies that industrialise AI across the enterprise could double operating profit, adding $254 billion in annual operating profit by 2030. They are also explicit that top-down programmes are required to address concerns and drive adoption. The companies that build AI-literate leadership populations will capture a disproportionate share of that value. The ones that treat AI as a technology project will watch it happen from the sidelines.

Designing Drive Digital

Working with HEC Paris, and specifically with Julian Schirmer and René Eber as academic co-directors, we co-created the programme that became Drive Digital.

I have long believed that the best executive programmes come from genuine collaboration between practitioners and academics. You need the conceptual rigour, but you also need a design that works in the room with senior leaders and survives contact with the real world.

One detail mattered a great deal to me: I did not just help design the programme. I went through it as a participant. Everyone reporting to an ExCom member was in scope, so I sat in the cohort sessions, did the pre-work, developed value theories with peers, and experienced every element of the journey I had helped shape. That dual perspective kept the design honest. When something did not land, I felt it directly, not weeks later through a feedback form.

I should also credit Ryan, my Head of Leadership & Executive Development, who was instrumental in the design, delivery, and iteration of the programme. This was very much a team effort.

What mattered most, though, was that Drive Digital was not an HR training programme dressed up in modern language. It was a joint intervention, co-owned by the business, the digital function, and my team. That showed up in how we presented it to the ExCom, and in the fact that ExCom members actively sponsored cohorts. AI upskilling fails when HR tries to solve it alone. It also fails when the digital team tries to change leadership behaviour without changing the leadership development model. This only worked because both sides owned it.

From the beginning, we were clear about the design principles.

This was not AI education for AI’s sake. Business leaders had to stay in the driving seat. Every element was anchored in Sanofi’s strategy. Participants were not students. They were executives accountable for translating digital ambition into business reality.

We also wanted accountability, not just awareness. Executives needed to understand their personal role in shaping the company’s transformation, gain practical frameworks for agile working, data-driven decision-making, and user-centric innovation, and then apply all of that in ways that produced something tangible.

That is why the value theory mechanism became so important. Each cohort had to produce concrete proposals for how AI and data could create business value across Sanofi’s value chain. These were not academic exercises. They were coached, challenged, pitched to juries, and in the strongest cases presented to the ExCom and taken forward.

And finally, we designed for collision. Cohorts were deliberately cross-functional: leaders from R&D, manufacturing, commercial, medical, and corporate functions learning together. The power came from the overlaps. A head of supply chain working with a head of R&D on a shared AI use case will discover dependencies and opportunities that would never surface inside a silo.

The programme itself ran over 12 weeks for each cohort of roughly 35 executives. It began with six weeks of virtual learning, around two hours a week, combining simulated dialogues, applied exercises, and live sessions. That was followed by a three-day in-person bootcamp at HEC Paris, where teams developed and pressure-tested their value theories. Then came four weeks of on-the-job learning, coaching, and refinement, culminating in jury pitches.

Across 2023, we ran four cohorts, with 133 senior executives in total. And critically, the entire Executive Committee went through a condensed version of the programme as well. That changed more than people might think. When the ExCom has done the same pre-work, heard the same frameworks, and wrestled with the same questions as the people reporting to them, the conversation afterwards changes qualitatively. It becomes much harder to fall back on, “I’ll delegate this to someone who understands it.”

What Changed in the Room

The headline metrics were strong. The programme achieved a 4.6 out of 5 satisfaction rating, an NPS above 90, and 95% of participants said it had a sustainable impact on how they viewed Sanofi’s digital transformation.

But the numbers do not capture what actually changed.

Most executives arrived at HEC expecting a technical deep dive. Many assumed they were about to be lectured on machine learning and data architecture. A number told me afterwards that they had been quietly anxious about being exposed as not technical enough.

By the end of the first day, you could see the shift in their body language. The relief was almost visible.

They were not being asked to code. They were being asked to think differently about their business, their decisions, and their teams. Once the anxiety about technical inadequacy evaporated, something much more useful took its place: urgency.

The recurring insight across all four cohorts was remarkably consistent. This was not, at heart, a technology problem. It was a culture and mindset problem. Executives realised they needed a common language. They realised peer learning across functions mattered far more than they had expected. They realised that the business, not the technical function, had to stay in the driving seat. And they realised the window for action was not theoretical or comfortably distant.

One executive captured it perfectly: “I came here thinking it would be quite technical and am surprised at how un-technical it really is. This is a behavioural issue, not a tech issue.”

That single insight, internalised by 133 of the company’s most senior leaders, was worth more than any standalone technology investment.

24 Value Theories: From Learning to Doing

Across the four cohorts, 24 value theories were developed. These were concrete proposals spanning the value chain, from early research and development through manufacturing, commercial activity, market access, and corporate functions.

Several moved beyond the programme and received dedicated funding and resources. They became real business initiatives. The ExCom took multiple ideas seriously enough to invest in them, with Blockbuster Booster and Turing among the notable examples.

The programme also influenced how leaders set goals. By early 2024, more than 1,600 AI-related stretch goals had been recorded across the senior leadership population, with the vast majority embedding AI, plai, or digitalisation into annual objectives. We even used AI to analyse the goals themselves, clustering them around themes such as AI tool adoption, AI in R&D, AI in business processes, and AI-driven operational streamlining.

This goes back to the Venn diagram I opened with.

The strongest value theories did not come from the most technically savvy executives. They came from the people with the deepest domain expertise: leaders who understood R&D pipelines, manufacturing operations, commercial trade-offs, or market access in the grainy detail that only experience gives you, and who now had enough AI literacy to see how the technology could reshape those domains.

That is the sweet spot: subject matter expertise plus AI literacy.

Learning without doing is tourism. The value theory mechanism forced leaders to move from, “I understand AI,” to, “this is what I am going to do with it in my part of the business.” That transition, from comprehension to commitment, is where most corporate learning programmes fail.

The Cascade: From 150 to 1,500 to Everyone

This is where many organisations lose the plot. They invest heavily in a flagship programme for the top team and then assume the insight will naturally cascade down. It will not. Cascade has to be designed.

Our model was a pyramid with three layers, each with a different level of investment, a different learning design, and a different definition of success, but with the same core language and strategic intent running through all of them.

At the top sat the Top 150: Drive Digital in 2023. High-touch, immersive, and intensive. Twelve weeks. Small cohorts. HEC Paris. Value theories. ExCom pitching. This was where conviction was built and where the common leadership language was forged.
The next layer was the broader senior leadership population: Drive Digital@Scale, rolled out from 2024 onwards. Same core content, different economics and delivery model. Larger cohorts, regional in-person events rather than a three-day Paris bootcamp, but still anchored in teamwork, use-case development, and leadership application.
The final layer was the wider organisation. Through Sanofi’s Digital Learning Institute, the core ideas became accessible to everyone through digital learning, tools, and self-serve resources. Not the same depth, but the same message.

I left Sanofi in late-2024, so the continued cascade was delivered by the team without me. That is exactly how it should be. A programme designed well does not depend on its architect. Ryan and the HEC/OAO team carried it forward brilliantly.

The underlying logic was straightforward: intensive for the few who set direction, efficient for the many who execute, accessible for everyone who needs to participate. What mattered was consistency. Every tier had to hear the same message, use the same vocabulary, and understand the same strategic intent.

Many organisations make one of two mistakes with AI upskilling. They either spray generic e-learning across the whole organisation, which creates reach without depth, or they run an exclusive programme for a small elite, which creates depth without reach. Neither works on its own. You need a deliberate cascade.

What Any Organisation Can Take From This

What I learned from this approach can be applied to any company looking to drive top to bottom AI literacy and adoption.

First, demand deliverables, not just learning. Learning changes what people know. Deliverables change what people do. For us, the value theory mechanism was the most important design choice because it forced leaders to translate insight into action.
Second, focus on the overlap. The highest-value people in any AI transformation are not pure technologists and not pure domain experts. They are the people who combine deep subject matter expertise with enough AI literacy to recognise where the technology can genuinely shift outcomes. Build that overlap deliberately. It will not emerge on its own.
Third, design the cascade on purpose. Intensive for the few, efficient for the many, accessible for everyone else. Same message, different depth. And remember that casual AI use changes very little. The Slack Workforce Index showed that people using AI less than once a week see almost no measurable benefit over non-users. The gains come with frequent, ideally daily, use. Upskilling has to create enough depth to change behaviour, not just awareness.
And finally, one of my long-standing rules of transformation applies here as well: roughly a third of any population is relieved you are doing it, a third is on the fence, and a third is actively resistant. Design for the middle third. Win them over with early, visible, tangible progress. That is why the value theories mattered so much. They turned abstract intent into real projects with real resources and real legitimacy.

The Bigger Picture

We are living through a moment in which almost every large company is investing in AI, yet almost none would honestly describe themselves as mature in how they deploy it. McKinsey’s latest State of AI research found that CEO oversight of AI governance is one of the factors most associated with higher EBIT impact from generative AI, and that workflow redesign has a greater effect on value capture than technology deployment alone.

That tells us something important. The gap will not be closed by better technology on its own. It will be closed by better leadership. By executives who understand what AI can and cannot do. By leaders who can separate genuine value from expensive theatre. By people with the conviction, the vocabulary, and the authority to redesign how work happens in their organisations.

Every company I have worked with, from Nike to Shell to Sanofi, has eventually arrived at the same conclusion: transformation is a leadership challenge that uses technology, not a technology challenge that merely requires leadership support.

AI is no different. The real AI skills gap is not in your engineering team. It is in your boardroom.

And closing it does not start with a hire.

It starts with a mirror.

let's talk about ai readiness

about the author

Andrew Kilshaw

Founding Partner, TalentOptima | Founder, phoque.ai | Guest Speaker, IMD

Andrew Kilshaw has spent 25 years leading capability, learning, and transformation work inside Nike, BlackRock, Shell, and Sanofi. At Nike, he created the company's first centralised Corporate University (NikeU) and later led HR for Nike Digital.

At Sanofi, he built and led the Group Organisational Capability & Transformation function, a team of 300+, with enterprise learning, transformation, people analytics, and organisational consulting in scope. He has been a guest speaker, teaching digital transformation at IMD in Switzerland alongside Professors Stéphane Girod and Michael Wade.

Explore examples of his experience

closing the ai skills gap

Why Executive AI Literacy Matters Now