Supply chain transformation: why only 30% succeed, and the basics Knut Alicke says to fix first

About Knut Alicke

Thirty years in supply chain: research, then the planning software company ICON (later sold to E2open), then about 20 years at McKinsey as a partner, alongside a professorship at the University of Cologne. Co-author of Source to Sold, a book of leadership stories from 26 supply chain leaders who made it to the board. Listen to the episode.

Only 30% of transformations succeed

Knut Alicke spent two decades watching companies launch supply chain transformations. The number he keeps coming back to is from McKinsey's broader research on transformations of every kind: "It's a stunning number, it's only 30%, and that number stays constant over the years." The rest take longer, under-deliver, or quietly revert.

The framework McKinsey built around that finding is called performance and health. Performance is the part everyone plans for: reduce inventory, lift service, grow EBITDA. "If you do that, what often happens is that you have a short success phase, where you increase the performance, but then mid-term, long-term, you go back to the old habits." Health is the other half, leadership, people, capabilities, "all the soft topics", and in a technical field like supply chain it is "completely underestimated, and not really focused on".

The 30% that succeed are, in the analysis, the companies that work on both. Or as Knut puts it: the ones that build an organisation that "lives and breathes continuous improvement", rather than treating the project as the finish line.

The myth of the big bang

Knut's lesson from lean manufacturing is that step changes are risky and small steps are not. "It doesn't make sense to aim for a big step, and then stay there. It rather makes sense to do it in small baby steps, and with this take the organisation with you."

His example is a multinational pharmaceutical company, multi-billion revenue, that implemented a new advanced planning system. "After two years of implementation, they had seven users. Seven. Everyone else is still working as they did before, downloading into Excel, doing some stuff, and then uploading it again." The impact was never captured, "just because they forgot to work with the people".

That doesn't mean the ambition has to shrink. The 10x programmes borrowed from digital companies get one thing right, the vision: "We don't want to only improve a little bit. We really want to be the best." What they get wrong is the route. "How to get there doesn't mean that we need to do a big bang and boom, we are there. We need to get there step by step."

Three basics most companies are missing

Ben asked which fundamentals companies most often lack. Knut named three.

Master data. "The most unsexy topic of the world. No one addresses it. It's not rocket science, you just need to do it. But no one is awarded for cleaning up the master data." The consequence for anyone moving to algorithmic planning is blunt: "If your master data is crap, then the result is crap. And then people thought, I knew that I need to correct it anyhow, so why do I need the algorithm?"

Experience. Twenty-five years ago Knut's team wrote planning software for Hewlett-Packard, a company full of people who understood their supply chain inside out. "A lot of companies these days are not at that level that we and Hewlett-Packard were 25 years ago." The reason is tenure. Planners rotate out after two years to manage their careers, "so they have no chance to really understand the business from inside out". Wim asked whether rotating through supply chain roles builds end-to-end capability. Knut would rather people stayed: "If you have a complex planning task and you stay there for two years, then you're more moving around numbers, but you're not really challenging. You're not really trying to improve."

Support from the top. If the board sees supply chain as a cost centre, "you go back to your silo. You work only on forecasting, but you don't have the perspective of integrating sales, supply planning, production planning, procurement." The money is in the end-to-end view: "If you just reduce costs a little bit, that's not interesting. It's interesting to improve availability, to increase revenue, and increase margin."

And if one of the three is missing? Capabilities can be built; there are smart people in every company. Top-level support is different. "If we don't have support from the top, then you should really reconsider whether it makes sense to continue, because your impact will be limited."

Where GenAI helps with the basics, and where it doesn't

Knut doesn't think the basics have to be done by hand. On master data, GenAI "helps the experienced people in the business to be much faster in cleaning the data". His example is lead times: compare the master data lead time with the measured one, then let the model connect the gap to reasons. "This lead time was four weeks, and the reality shows that it's very often six weeks. Why is that? Did we have different customs processes? Did we change the route and not communicate it?"

The caveat is firm: "I would always be very careful in having no human in the loop to do everything automatically, because that can fire back. Hallucination is not yet solved."

Brilliant basics: three scatterplots

Some of the most useful analysis in planning, Knut argues, needs no AI at all. He calls it brilliant basics, and his three favourites are all scatterplots.

Measured lead time against master data lead time, which shows where the parameters are wrong. Days of coverage per SKU against replenishment lead time: "If the average days of coverage is 50 and the replenishment lead time is only 10, then you immediately know that there's something going on that cannot be correct." And service level against customer importance or revenue. Companies tell sales they have platinum and gold customers, then never check. "Just plot the OTIF over the customer. People forget that if a customer is smaller, they keep calling: where's my order, can we prioritise my order? So it gets into this loudest-shout-first principle, which contradicts the service level strategy of the company."

"Sometimes I feel that with all of the AI stuff, we forget this super basic analysis."

Do you need AI in the project title?

Ben asked whether a project now has to be pitched as AI to get approved. Knut has seen the label change many times: "The ask was always, we need to do something that is innovative, new, transformative. And now it's AI, GenAI hype." Painting the vision is fine. The failure is skipping the clean-up. "You enter with the vision, we need to really be digital. And then you realise, oh, you still use fax machines. Let's start to clean up the master data first." Otherwise, "worst case, you automate a crappy process and then you have an automated crappy process".

On the question of whether smaller companies can keep up with multinationals, his answer is that they usually move faster. Big companies "go from committee to committee, and the message is blurring and the decision does not happen or comes too late". In an SME "you often have the decision makers in the team or very close to the team. If you want to move something, it's decided and then it's moving."

What GenAI will change in planning

Knut is "a strong believer that the current development of GenAI will change the supply chain more than everything else", and compares it to the container: a standard box that, over time, changed how global supply chains worked. His own test was the travelling salesman problem. The first models failed, and he filed it under the next blockchain. Then a later model called an optimisation library and came back with the right answer. "It's still a large language model. Why could a large language model do math? Because you have a lot of stuff around you that you can use."

He sees three areas. Root-cause analysis, the five-whys walk through the data that every planner does when a delivery is missed. A conversational layer between the planner and the planning system: "Most planners don't trust their planning system. They don't understand exactly what is going on. That's why they have shortcuts with Excel." If the user who knows the business can talk to the planning system and the ERP through a GenAI layer, "this will enable the planner to focus much more on the value-adding task, on the exceptions". And bespoke programming: 60 to 70% of a project is standard, and the specific rest, which used to be forced into the standard, "can now be programmed very fast".

He doesn't expect the big planning vendors to build that layer themselves. "Here we have a little bit of the innovator's dilemma. If we are able to sell more licences and more programming hours, why would you offer this one that would cut part of your revenue stream?" More likely, "a couple of start-ups will provide this layer on top of planning systems, and at some point in time might be acquired by the big guys".

An experiment: inventory optimisation by conversation

After Nicolas Vandeput launched his inventory competition, Knut set himself a rule: solve it using only a GenAI bot, no additional programming. The bot wrote Python, handled the data and played back results. It also got things mathematically right and practically wrong, until he explained a two-week period where demand burns down inventory before supply arrives. "As an experienced supply chain person, you now have at your hand a super fast and super smart supply chain analyst, just coming from university, not knowing all the nitty-gritty of your business."

The implication for teams is about onboarding. "The young colleagues, the supply chain analysts, we now need to find a way to train them very fast in business. It cannot be that they take five years to get into it."

What the leaders had in common

For Source to Sold, Knut and his co-authors interviewed 26 leaders with a supply chain background who reached the board. Jim Rowan, then at Dyson and later Volvo Cars CEO, told his supply chain people to stop complaining about forecast accuracy: help sales make it as good as possible, then "set up your supply chain to still be able to deliver". Ivanka Janssen, then at Philips, had just completed a Python course as chief supply chain officer, and ran reverse mentoring to learn from younger colleagues. Claudio Strobel invited his most important suppliers into the S&OP meeting; delivery performance went up.

The pattern became the CHAIN framework: collaborative, holistic, adaptable, influential, and narrative. The last one matters most. "Telling the story of supply chain not in the nerdy way, we improved our truck utilisation by three percentage points, which honestly no one would understand and value. But: with this lever, we improved availability, increased sales."

Your next steps

• Check your transformation plan for the health half: who owns capability building, leadership behaviour and the continuous improvement routine after go-live?
• Count the real users of your planning system. If most planners still plan in Excel, the transformation isn't done.
• Run the three scatterplots: measured versus master data lead time, days of coverage versus lead time, OTIF versus customer importance.
• Use GenAI to speed up master data clean-up, with an experienced planner in the loop on every change.
• Keep good planners in role longer, and build a fast business onboarding for new analysts.

Knut's advice for anyone in the field is two words long. "One is be curious: try things out, question things, speak up. The other one is communicate. Understand what the others you're talking to understand. Not supply chain nerd to supply chain nerd."