The Corporate AI Journey: Why maturity is a leadership question
Most companies are somewhere on the AI journey, and most of them are stuck in its first two stages. We have mapped that journey as four waves of value and five levels of maturity. The map shows where an organisation stands. What it cannot show on its own is why it stays there. That, we would argue, is always a question of leadership.
The chatbot that worked
In February 2024, Klarna announced that its AI assistant had handled 2.3 million conversations in its first month: two-thirds of all its customer-service chats, the work of roughly 700 full-time agents. Resolution times fell from 11 minutes to two, repeat enquiries dropped by a quarter, and customer satisfaction was, in the company’s own words, on par with human agents. Klarna expected about USD 40m in additional profit that year.1 It quickly became the most-quoted AI case of the year.
Fifteen months later, CEO Sebastian Siemiatkowski told Bloomberg: “As cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality.” Klarna started investing in human support again and promised customers there would always be a human if they wanted one.2
What strikes us is what did not fail. The technology did what it was supposed to do, and AI’s share of the work kept growing afterwards. What failed was a judgement: which yardstick would define success. No model made that decision. People at the top made it, and the same person corrected it in public (which deserves credit; not many CEOs say “we got this wrong” to Bloomberg).
Klarna’s case is unusually well documented, but the pattern is common. In BCG’s 2024 survey of 1,000 executives, 74% of companies had yet to show tangible value from AI.3 In McKinsey’s survey published this August, only 37% of respondents attribute any EBIT impact at all to their use of AI, and the share of high performers has stayed at about 6%.4 The investments have been made. Most of the return has not arrived.
Four waves, five levels
The Corporate AI Journey & Maturity Framework, which Alexander developed, describes the journey along two dimensions. The first is value: four waves, each a different way in which AI creates a return.
- GenAI-supported organisation. Off-the-shelf tools embedded in daily work, to boost individual productivity and reduce cognitive friction for every knowledge worker.
- Enhanced processes, products & services. Core processes re-engineered, and AI built into what customers receive.
- Product & service innovation. Net-new, AI-native revenue: products that could not exist without AI.
- Business model innovation. A different logic of how the firm creates and captures value.
The second dimension is maturity. In each wave, an organisation moves through five levels: Exploration, Experimentation, Operational Pocket (one area works this way as standard, and the result is proven), Full Scale, and Competitive Advantage.
Two features matter more than the grid itself. First, the waves overlap. They run in parallel, so an organisation can be at Full Scale in wave 1 while it is still exploring wave 3. Second, the waves are not equally valuable. Wave 1 is where everybody buys the same tools from the same vendors, and efficiency gains are symmetric: if your competitor has the same licences, the net effect on your position is roughly zero. Advantage sits further along, and further up.
Where do most companies stand? No large survey measures our waves directly, so we have to read the closest proxies, and they point the same way. In Deloitte’s 2026 survey of 3,235 leaders, 37% use AI at a surface level with little change to processes, and only 20% already see revenue growth from it, against 74% who hope to.5 BCG’s 2025 study classes 5% of companies as “future-built” and 60% as laggards with minimal gains.6 Our reading: most organisations sit in waves 1 and 2, on the lower levels, and many believe they are further along than they are.
Three companies, three places where it gets stuck
The framework places an organisation on the map. The interesting part is what happens at the edges between the fields, and three well-documented cases show it at different points of the journey.
JPMorgan, wave 1 at full scale. In July 2024 the bank launched its internal LLM Suite; within eight months it was in the hands of about 200,000 employees.7 Few organisations have taken wave 1 further, faster. Its executives have also been unusually frank about what should follow. The CIO reported that the coding assistant made software engineers 10 to 20% more efficient.8 At the 2025 investor day, Marianne Lake, who runs the consumer business, said AI would allow a 10% reduction in headcount in operations and account services, and added: “I would take the over on this projection.”9 Then, according to reporting earlier this year, the bank began sorting its engineers into light, heavy and non-users of AI tools, with usage feeding into performance reviews.10 Critics pointed out the obvious: this measures how much people use the tools and says nothing about what the use produces. Whatever one thinks of it, the episode shows where wave 1 ends. A mandate can put a tool on 200,000 screens. Deciding where the saved time goes, and what counts as success, is a different kind of decision.
Ping An Good Doctor, wave 3 with no business model behind it. Ping An built one of the largest AI-supported health platforms in the world, with 400 million registered users by mid-2021.11 A year earlier, in May 2020, the board had already removed the founder-CEO and most of his team for “not having met the expectations of the board”: growth had not turned into profit.12 The company reported its first full-year profit only for 2024, after re-focusing on corporate clients and Ping An’s own insurance customers.13 A product that hundreds of millions of people use is a wave 3 success. Turning it into a different way of earning money is wave 4, and that move was decided in the boardroom.
Tesla, a technology lead that has not become a business model. Elon Musk has talked about licensing Full Self-Driving to other carmakers since 2021. In November 2025 he conceded: “I’ve tried to warn them and even offered to license Tesla FSD, but they don’t want it!” According to industry reporting, prospective partners wanted Tesla to carry crash liability, which it would not.14 Whatever one thinks of the technology, the blockage sits in trust, liability and terms between the leaders of two companies.
We would add a fourth observation, from Siemens. Its CTO, Peter Koerte, said in 2025 of generative AI across the group: “We have it pretty much everywhere. The question is, how much have we saved?”15 It is the most honest one-line summary of wave 1 we have come across.
The leadership dimension: seven moves
A maturity map is static: it tells you where you are. What it does not show is how you get from one field to the next, and who has to do it. That is the dimension we have added to the framework, and for LUMINUM it is the essential one.
There are seven moves: the actions leaders can take. Four go up a level within a wave, and three go from one wave to the next. Each is made by a small number of leaders, and each asks something of them that their role has not asked before. The function head whose pilot worked now has to take apart how their own team works. The sponsor of a success has to hand it over to peers who did not design it and may not want it. A CEO who funds a new way of earning is also running the business it may replace.
| Move | From → to | Who carries it | What is new for them |
|---|---|---|---|
| 1 · Deciding to try | Exploration → Experimentation | The executive sponsor and the first function head willing to host a pilot | Sponsoring uncertainty in public: small exposure in money, large in reputation |
| 2 · Changing how the work is done | Experimentation → Operational Pocket | The head of the function where the pilot runs | Redesigning roles, handovers and measures in their own team, often the job they were promoted for |
| 3 · Adopting what someone else built | Operational Pocket → Full Scale | Peer function heads, the original sponsor, the COO and CHRO | For peers, giving up autonomy; for the sponsor, handing over ownership and credit |
| 4 · Turning capability into choices | Full Scale → Competitive Advantage | The CEO and the executive team | Using the capability for choices competitors can see: pricing, where to compete, whom to hire |
| 5 · From personal help to redesigned work | Wave 1 → wave 2 | Process owners, rather than IT or individual users | Owning a process end to end, across other leaders’ territory, including where people are redeployed |
| 6 · From saving to building | Wave 2 → wave 3 | Business-unit and product leaders, and the CFO | Funding bets with uncertain returns, and stopping them on evidence, in an organisation that has learned to judge AI by cost saved |
| 7 · Questioning how the firm earns | Wave 3 → wave 4 | The CEO and the board | Funding what may replace parts of the business they run |
These moves stall in ways that can be recognised, and the patterns have little to do with the technology. Move 2 stalls when the pilot succeeds and the team quietly goes back to working as before, because nobody switched the old way off. Move 3 stalls on “our situation is different”, which is sometimes true and often a way of keeping control. Move 6 stalls when every AI proposal is judged by the savings it will produce, a test that new products can never pass. Klarna adds a further pattern: the work was redesigned, and then judged by the wrong yardstick.
The survey evidence is consistent with this, although it shows association rather than cause. McKinsey finds that nearly three-quarters of AI high performers have fundamentally redesigned workflows, against roughly one in four of the rest, and that high performers are twice as likely to say their senior leaders demonstrate commitment to AI.4 BCG’s long-standing rule of thumb is that leaders put about 10% of their effort into algorithms, 20% into data and technology, and 70% into people, processes and culture.16
One situation sits outside the seven though, and it is probably the most common: leading through a change that you do not own. Many managers find their area reshaped by AI while someone else steers the programme. They still lead the people it affects, and they are expected to look certain about something they do not control. So the ground moves under them while their role stays the same. As nobody calls that a transition, nobody offers support.
Yet, support is very much needed. This, we think, is the most substantial transformation all of us will see in our lifetimes.
Where to start
If the return on AI depends on a handful of leaders making specific moves, the useful questions for an executive team are fairly concrete:
- Where do we really stand in each wave? Evidenced, in a report someone answers for, rather than claimed in a strategy deck.
- Which move comes next, and who has to make it? By name.
- What does that move ask of that person that their role has not asked before? And who supports them while they do it?
- What yardstick will we judge it by? Klarna chose cost and got quality problems. The yardstick is a leadership decision and deserves to be taken as one.
- Who on the executive team owns the question of what AI should change about our strategy? If nobody does, the capability stays an operational asset, however well it is deployed.
None of these questions needs a new technology budget, but most of them are uncomfortable, because they put a name next to a result.
We have built two short diagnostics around the framework:
The organisation diagnostic takes about six minutes, is meant for the executive sponsoring AI, and shows where the organisation stands, which move comes next and who has to make it.
The leader diagnostic takes about five minutes and shows any leader carrying part of an AI change what their move asks of them personally. Both are on our website, and we are happy to talk through the results.
Where does your organisation stand?
Take our testWhere do you stand as a leader?
Take our test
