The AI Catalyst: Scaling Adoption Without Waiting for the Infrastructure with Patrik Hedljung, Scania

John Emmerson: Hi, and welcome to another aibl Leadership Series interview. Today I am delighted to be joined by Patrik Hedljung, who is AI Catalyst at Scania. Thanks for joining us today, Patrik.

Patrik Hedljung: Thanks for having me.

John Emmerson: We got introduced quite a long time ago now, did we not? I cannot remember who it was that introduced us. But I think you were described to me as Sweden’s AI gorilla. And I do not mean that in an ape sense. How do you feel about that description?

Patrik Hedljung: Apart from the ape aspect of it, I think it is actually pretty spot on. That is how I have approached this subject, and that is what I still do to quite a large extent. Leaving all political aspects of that aside, it captures something important about my approach and what I have been doing.

John Emmerson: I immediately loved it, which is why I reached out to you. And I am delighted that you will be joining us at aiblLIVE in October to speak at the event. But today we are going to have a conversation about your approach. Let us start with your title. AI Catalyst is not a very standard title. I do not think I have ever seen anyone else called that. So what does it mean? What do you do on a normal day?

Patrik Hedljung: I usually explain it in terms of being responsible for AI adoption at Scania. What I typically do is anything and everything I can possibly come up with that helps move the needle on getting people to use the technology once it is in place.

It is a role within our IT organisation, but my background is not IT. I moved into it. I think that also colours my approach and what I do. It has got very little to do with the technology. Of course I need to stay updated and knowledgeable about what those preconditions look like at Scania. But the different initiatives I drive are very little about the technology.

It is more about what creates momentum. It can be anything from trying to squeeze a few standard sentences into all our job adverts, in order to send a signal to applicants that Scania is an AI friendly company, and that we expect people to have a positive or curious attitude. Of course, in a large organisation, just finding the one person who is responsible for that template could be a full-time job. So that is one end of the scale.

At the other end, sometimes it is pushing for a new tool, or arranging a hackathon with a new feature or tool that we may or may not expect to acquire at some point. Just to get people excited about a certain aspect of AI. Right now we are thinking about physical AI, so there are a few startups doing vibe coding hardware that I am looking into. So how can I push that limit? It is anything and everything, and thinking about what creates momentum.

John Emmerson: It sounds like a very cool role, I have to say. And I love that your background is HR. You came out of human resources, did you not?

Patrik Hedljung: Yeah, L&D.

John Emmerson: Regular listeners to this series will know that at aibl we have gone from not really considering the HR and people function as part of the AI adoption journey, to going full swing the other way. We now see it as potentially the most important function in AI adoption.

I do not think you are alone in being a person with high levels of accountability for AI adoption who has come out of HR. I am seeing that more and more. We had a workforce event that was all about this a couple of months ago. So given your background, what do you think you bring to the table that a more tech-focused person might miss? And the flip of that, what do you think they catch that you might not?

Patrik Hedljung: I have sometimes jokingly said that I used to be the one person in HR thinking about AI. Since I moved to IT, I now see it the other way around. I am one of few in a world where everyone thinks about AI, but very few people think about change management, leadership, adoption, and all that goes into that. That is certainly a constant struggle.

So what I bring to the table in IT is definitely everything that is non-technical. The importance of leadership, communication, and competence. The broader brush strokes when it comes to what we think the consequences of AI will be down the line, and how we prepare ourselves as an organisation. Not technically, but workforce-wise, for that change.

To be honest, nobody knows anything about the future. But we have some ideas. And if anything happens along the lines of what we think, the changes are going to be huge. They are going to be about many other things than the technology itself. We tend to overestimate the technological change and underestimate the culture and human ways of working aspects. Even though in this case I think we are also underestimating the technological changes, we are certainly missing a lot of what that will mean for us as an organisation. Certainly for a big old legacy company of 60,000 plus people, over 100 years old.

John Emmerson: You have previously said that the bottleneck is culture, not tech. You have just said you do not know what the future holds, but I am going to ask you to try and predict now. If the current bottleneck is culture, and that is certainly something we hear a lot and something we will be addressing throughout the event in October, what do you think the next bottleneck is? Once people have got over that one, what comes next?

Patrik Hedljung: I think culture gets us only so far. A good culture will enable experimentation. In Scania’s case we have a good culture that allows people to experiment a lot. Our AI journey has been very much a bottom-up journey, and we have a very decentralised decision-making culture.

We have both the resources and a tradition, I would say, of investing in technologies we cannot get the ROI from on a short-term basis. I have no idea how much money we poured into autonomous driving, and we still have not sold many autonomous trucks. But we still believe in that technology.

In our case that serves us well when it comes to AI. Now people and managers are starting to ask questions about ROI. But up until now we have been very blessed in being able to invest in licences and technology that allowed people to experiment a lot. So experimentation is not a problem at all. We can come back to whether experimentation and proofs of concept are a problem or not.

But I think we get to a point where, okay, now we have a thousand flowers blooming. How do we coordinate that? So I think the next bottleneck will be operating model of sorts, and coordination.

At the same time, there are so many voices saying you have to start with the structure, the governance, the data management, all those things. I am not an expert in those, so I am biased, definitely. But I strongly believe in the democratisation aspect, and I strongly believe in letting a thousand flowers bloom. We also get to a point, though, especially if we want to leverage all the data we have accessible within our organisation, where we have to prioritise and streamline. We have to make some decisions about access, tools, and connections. Some people are going to cry out a little bit about that, because they all have their own solution right now.

John Emmerson: It is going to come to senior leaders having to choose. If there are high levels of innovation, idea generation, and experimentation, then at some point someone somewhere is going to have to say, okay, we want to do that, that, and that. And we are not going to do those, and these are the reasons why.

That in itself becomes a challenge, does it not? It becomes a challenge for the people having to make those decisions, to make sure they are using the right metrics to pick the right ideas. And it becomes a leadership challenge in terms of how you manage the people who thought of these amazing ideas you are not taking forward. That is just business, right? That happens now, maybe not with the same level of idea generation, but certainly in that context. Do you think that is right?

Patrik Hedljung: Yes and no. It definitely will come to a point where leaders need to prioritise. The question is on what level we want them to prioritise. And I am not sure use cases is the answer.

Yes, it is great if we can have a handful of high value end-to-end use cases where we get some quick ROI and make people happy about that. That is excellent, and they can be lighthouse cases. But I also see a risk that organisations who take that approach first then become complacent. Okay, we are doing AI, we have these five use cases. And people using those solutions do not even necessarily know there is AI involved when they carry out their work in these processes.

I am talking to several consultancy companies, and we are not going to name anyone, that have this approach. From an outside-in perspective, and from a consultant’s perspective, it is much easier. You have your benchmarks, you know the best practices in the industry on an aggregated level. It is very easy to present those cases to a company and say, okay, you can automate your XYZ process, whatever it is, and here is how we can help you with that. It is not that this is not true. There probably are good use cases in that.

But I believe if we really want to leverage the power of AI, we need to activate everyone within the company. We need to raise the general understanding and competence around AI, so that each subject matter expert, each nitty-gritty person who has a specific workflow, can start thinking about how we can use AI in this work. That is when we are going to get the real effect from AI, from a general purpose technology.

If it was a specific tool we were talking about, it would be different. But this is a technology where we have just begun scratching the surface of what we can use it for. So I believe strongly that we need to put this in the hands of the many, and enable them to become entrepreneurs, creative people. What we need to decide at a higher level is what the methods are. What does good governance look like? How do you scale? How do you take your good idea and deploy it in a compliant way? How do you scale that to other functions? We are not there yet. Certainly not.

John Emmerson: A point there is that we are not scratching the surface of this tech, and linked to that is the rate of change and advancement, and how astonishingly quickly things change. Is there something you believed about AI adoption 12 or 18 months ago that you have since abandoned, or that has changed in a big way?

Patrik Hedljung: I guess maybe I slightly overestimated how far the bottom-up approach would take us. I ran into quite a few walls along the way.

How should I put this? Let me rephrase that. I think you need to understand the organisation you are operating in. I really think there is no one solution that fits everyone. So I am a bit allergic to all attempts at these are the steps, this is the sequence, this is the recipe. I do not think there is one recipe that fits all organisations. You really have to start at your own organisation. What prerequisites are there in terms of executive buy-in, resources, technology, culture, other priorities, and strategy? Then you have to take it from there.

What I had available to me when I started was decentralised decision making, and the resources in terms of tech were there. So I am taking that as far as I can, and that is going quite well in many aspects. But when you start scaling it beyond individual or team level, then you really need some more hardwired things in place. Of course data access, different systems, and legacy systems all become very important. But I am never going to say that you need to start there. Because if you are going to wait for those things to be in place, then you still have not started with AI.

John Emmerson: That is interesting, because that is very different to an approach quite a few of our other interviewees have talked about. There have been a couple of people who said you need to start with your infrastructure and get that set up, and then you can start experimenting. I think what you are saying is that maybe you can do the two in parallel.

Patrik Hedljung: Yeah, and coming back to what the situation is you have right now. What does infrastructure look like in your organisation now? What does the learning culture look like right now? What is the leadership culture like right now?

Setting the infrastructure, and coming to an agreement about that, in the case of Scania is a very gruesome process. If we had waited to get started with AI until that was in place, then we still would not have started. And then we would have been three and a half, soon four years behind even starting.

Because it is a very decentralised organisation, different parts of Scania created their own infrastructure. I am not saying that is only a problem. It served us well historically in many aspects. But now there is an increased need for coherence and alignment, and for using all that data across the company. We need to take some hard decisions. But I think it was one hundred percent the right thing to start with what we could start with, and not wait for that to be dealt with first.

John Emmerson: And I guess, to your point earlier, organisations are all different shapes and sizes, with different setups. Infrastructure looks different, the tech stack looks different, data looks different. So there is no one size fits all approach to this, right? You have to assess based on the environment you are in. We are a startup, we can do what we like, it is easy. How many people at Scania, 60,000? Is that what you said earlier?

Patrik Hedljung: Give or take. To be honest, I do not know what the official number is right now, but roughly.

John Emmerson: But it is a big old place. Very different.

Patrik Hedljung: Yeah. Global.

John Emmerson: And also over 100 years old. Legacy systems that probably should have been updated many years ago and simply have not been. So we have all got this challenge. And I guess the advice to everyone is that you have to figure all this stuff out as you go along, and work out what is best for you, not necessarily for someone else.

Patrik Hedljung: There are several days in the week when I can be envious of a startup starting from a blank sheet of paper. It is like, okay, if we could rethink, if we could start all over, would we do things the way we do? No, of course we would not. One hundred percent not. Especially with AI.

But we have to play the cards we are dealt. We need to rethink things, it is going to take time, we need to involve people in the process. A startup does not have that legacy.

John Emmerson: No, we certainly do not, and that can be fun. Do you at Scania run post-mortems on AI projects? And if you do, what does that look like? What does it consist of?

Patrik Hedljung: I think the short answer is yes, I am sure we do. And together with that, I have no idea.

The reason I say that is that Scania is very heavy on lean as a methodology. Lean and agile, and in general heavy on methods, structure, and processes. Of course, learning, whatever you might call it, is an important part of any method framework. Learning from the last period or the last project, or whatever you have done. So absolutely, we do that.

Do we have a common method for post-mortems on AI projects specifically? No. And I bet that within IT it is done one way, in production logistics it is done a different way, and in a third business area it may not be done at all. So we do not have a common AI post-mortem method. But we have a long tradition of working systematically and methodologically with reassessing the work we have done. So it is part of the culture, for sure.

John Emmerson: That is just standard business advice, frankly, AI or not. Which is quite an interesting general piece of advice that we hear quite a lot. It does not matter that it is AI. Quite often, use common sense. What would you do with any other kind of new tech, or new disruption, or new behavioural change piece?

It is accelerated, and that is partly because of its ability to create opportunity and drive efficiencies. But it is also accelerated because of the coverage it gets in the media. And that leads me on to AI literacy and AI fluency programmes. They can often produce both high levels of enthusiasm within a workforce, and quite often high levels of fear as well. Because people are sometimes worried about what it is going to mean for them and their jobs. Is that something you have seen at Scania? Does that ring true?

Patrik Hedljung: I would assume there are people worried about their jobs. I have a hard time believing there is any organisation where people do not, to some degree at this point, have a fear about that. I think it is valid for a lot of people and a lot of jobs.

Then it is going to be down to each organisation to have a strategy and communication about that. How do we as a company talk about that question? What is our why, what is the reason we give for using AI? Is it to cut costs? People will very quickly equate costs with FTEs and understand, okay, we need to get rid of people. So I would argue for telling a different story. But of course that needs to be aligned in the executive management and in communications.

Our AI literacy programme is not an individual programme, but a team-based programme. You do a, I do not even want to call it a training. It is a facilitated discussion that each team does over the course of several weeks, and the leader or manager gets help to facilitate that discussion.

The idea is that there are a number of questions we need to discuss in our team, in our context, given our jobs and roles. Some of those are more technical and use case oriented. Some are more compliance related, what we can and cannot do. Some of them have more to do with the impact of AI on the work that we do.

So we want to give managers and leaders a chance to have those discussions openly within their teams, without necessarily having all the answers. Because we need to talk about this. We need to have dialogues about it. When we leave each to his own, then people are going to form all different kinds of fantasies and worries. Some are valid and some may not be. So we need to have those dialogues at scale. And not in a town hall.

John Emmerson: Well, people generally do not want to put their hand up at a town hall, do they? We are remarkably nearly out of time. So I always finish on the same question. The vast majority of our audience here at aibl are scale-up and mid-market leaders. If you had one piece of advice for that group of people, if they are starting their AI adoption journey, what would it be? What would be the one thing you would recommend they do first?

Patrik Hedljung: I just spent at least 15 minutes saying there is no one answer that fits everyone, right? So I set myself up for this one.

I would say do not wait. They wait for the technology, they wait for data, they wait for governance, they wait for a mandate, they wait for I do not know what. Do not wait for any of that. Start. Momentum is the important thing here. Nobody has the blueprint yet, and certainly not for you.

The only way to find out what your way in this is, is to experiment. Trial and error, see what works, see what does not, and see what you need in terms of data, infrastructure, and training. But do not wait for any of that. Just start and iterate quickly. Have short loops. Create channels for people to express themselves, and to capture what they are experiencing and thinking and worrying about.

John Emmerson: Well, there you go, folks. Essentially, get on with it, have some fun with it, experiment, see what happens, and go from there. Patrik, it has been an absolute pleasure talking to you today. Thank you very much. And I cannot wait to see what you bring to aiblLIVE in October.

Patrik Hedljung: There is so much time until then, so many things can happen, right?

John Emmerson: Absolutely, God only knows what is going to happen. Thanks very much, Patrik, it has been a pleasure. Thank you.

Patrik Hedljung: All right. Thanks for having me.

Should you fix your data and infrastructure before starting with AI?

Hedljung says no, and he is blunt about the cost of waiting. Agreeing infrastructure at Scania was a slow, difficult process. Had they held off on AI until it was settled, they would still not have started, and would now be close to four years behind. His argument is not that foundations do not matter. Data access, legacy systems, and coherence all became important once the work scaled. His point is about sequence. Start with what you already have, then build the hardwired pieces as you hit their limits, rather than treating readiness as a precondition.

What does an AI Catalyst actually do?

The role is responsibility for AI adoption, meaning getting people to use the technology once it is in place. Hedljung describes it as anything and everything that creates momentum, and very little of it is technical. At one end of the scale, that means adding a few standard sentences to every job advert, so applicants know the company expects curiosity about AI. At the other, it means pushing for a new tool or running a hackathon around a feature the business may not even buy, purely to get people interested. He sits in the IT organisation, but his background is HR and L&D.

If culture is the bottleneck for AI adoption, what comes next?

Coordination, and the operating model underneath it. Hedljung is clear that culture only gets you so far. A good culture produces experimentation, and at Scania experimentation is not a problem at all. The difficulty arrives once there are hundreds of local solutions in use. To make use of the data across the organisation, somebody has to prioritise, streamline, and make decisions about access, tools, and connections. He expects some people to object, because they all have their own solution already. That is the next constraint, and it is organisational rather than cultural or technical.

What is the risk of focusing on a handful of high-value AI use cases?

Complacency. Hedljung accepts that a few end-to-end cases with quick returns are worth having, and can act as lighthouse examples. The risk is that an organisation then decides it is doing AI. The people using those solutions may not even know AI is involved in their work, so no capability is built anywhere else. His view is that AI is a general purpose technology, so the value comes from activating everyone. Each subject matter expert who knows a specific workflow needs enough understanding to ask how AI applies to it.

Why does an HR background help with AI adoption?

Because almost nothing about adoption is technical. Hedljung jokes that he used to be the one person in HR thinking about AI, and now sits among people who all think about AI but rarely about change management, leadership, or adoption. What he brings is the non-technical side: communication, competence building, and a view on how to prepare the workforce for what is coming. He also warns against the standard error. Organisations tend to overestimate the pace of technology change and underestimate the culture and ways of working, and in this case may be underestimating both.

How should a large organisation run an AI literacy programme?

Scania runs theirs as a team activity rather than individual training. Hedljung is reluctant even to call it training. It is a facilitated discussion each team works through over several weeks, with the manager given help to lead it. The questions cover use cases, compliance, and the likely impact of AI on the team’s own work. Managers are not expected to have all the answers. The reasoning is that people left alone will form their own fears, some valid and some not, so the dialogue has to happen at scale. He specifically rules out doing it in a town hall.

What is the one thing a mid-market leader should do first on AI adoption?

Do not wait. Hedljung lists what organisations wait for: the technology, the data, the governance, a mandate. His advice is to start without any of it, because momentum is what matters and nobody has the blueprint, certainly not one built for your organisation. The only way to find your own approach is trial and error, seeing what works and what does not, then learning what you actually need in terms of data, infrastructure, and training. He adds two practical points. Iterate quickly with short loops, and create channels for people to say what they are experiencing and worrying about.