Interview with Rudy Kuhn, Evangelist at Celonis on Process Intelligence and beyond: Let AI handle the routine and humans the exceptions

If your organisation is struggling to extract value from AI, then you need to speak to Rudy Kuhn. As an Evangelist at Celonis, he has over 20 years of experience helping companies understand where their business struggles. And whilst there is usually a desire for a quick fix, the real skill is to understand the details – what Rudy calls process, and particularly process intelligence.

I caught up with Rudy at VivaTech 2026. You can watch the video in full below, and this article summarises the key parts of our discussion.

Value = insights x decisions x actions

“Everything we do, every system we use, leaves digital footprints behind,” said Rudy. “So, we collect all this data, and from the data alone we reconstruct, visualise, analyse and improve business processes.”

It’s akin to an X-ray, he explains, but Celonis doesn’t stop at identifying what’s causing the pain. “That was like the beginning. Today, we talk about process intelligence, we talk about context, we talk about AI, we talk about orchestration and automation. So we take this understanding of processes to the next level to help people and AI make better decisions.

“What should we do? How can we improve business processes and put this into action with automation and orchestration? So there’s almost like a formula: Value = Insights x Decisions x Actions, and that’s what we do.”

What is Process Intelligence?

This brings us to one of the core planks behind Celonis’ success, an area where it has been named a Leader in the 2026 Gartner Magic Quadrant: Process Intelligence. So what does this term mean?

“Process intelligence is a technology developed 15 to 20 years ago in the Netherlands by Professor Dr Wil van der Aalst, a good friend of mine, and by now our Chief Scientist at Celonis,” said Rudy.

He used the example of ordering material. “You get an invoice, you approve the invoice, you pay the invoice, whatever. Everything we do is recorded, every single step is recorded in your ERP system, in your workflow system, in your manufacturing system.

“From the data we visually reconstruct the process, but not just the one process, because there’s no such thing as one process. You have variations, and sometimes organisations believe they have… maybe only ten variations of a process, and sometimes they are very surprised to realise, ”Oh, we have 10,000 different variations’.”

To become an efficient company, Rudy explains, you need to understand where time is being wasted. “You need to understand the challenges, the bottlenecks, the compliance issues, so you can tackle them and become really the efficient company you want to be.”

Turning Process Intelligence into real savings

Great: you have all this data and Celonis has helped you understand it. Now show me the money: how do you turn Process Intelligence into genuine savings.

“I’ve been in the process optimization business for about 25 years, and over the time I learned there are only two ways how to improve processes: you either increase efficiency or maximise efficiency, and you reduce or minimise risk,” said Rudy.

“So, when we talk about value, it’s super-tied to unnecessary steps, you know, mistakes that happen, double payments, for example. Compliance issues. So it’s really about understanding where do we lose time, where do we spend too much effort, where we do make mistakes, where do we have loops? What is causing inefficiencies?”

Rudy gave examples of finding how why half of a company’s trucks are driving around empty. “Can we optimise the logistics process? And if you think about companies like BMW, for example, every single car produced by BMW is touched by Celonis Process Intelligence, because BMW is using this technology to monitor the production processes, the purchasing, the manufacturing, the service, everything, so they are really in control of what they do.”

Why AI is failing to deliver value for organisations

Naturally, we had to talk about the impact of artificial intelligence. And it’s safe to say that Rudy has a view on this.

“Generative AI like ChatGPT [is] fun, but in terms of real, tangible value for organisations, it doesn’t move the needle,” he said. “We can write emails faster, we can create presentations faster. But if you really look at the outcome, you don’t get rid of 50% of your employees, or you do not create more cars just by using generative AI.”

The magic only starts to happen, Rudy believes, when you properly train AI systems on processes that already take place in your organisation.

“One of our customers in Spain, a company called Cosentino, they produce building material for construction sites, and not all of their customers have already paid invoices in time before they ordered new products. So the ERP system automatically created order blocks. We can’t process this order because first we need the money.

“So what happens? [Cosentino] needs to decide, okay, should we take the risk and deliver before having the money? If we don’t, there’s an unhappy customer, there’s an unhappy salesperson, and it’s a lose-lose-lose situation, because we want to make the deal, but we need to know if we [will] get the money.”

This is where Celonis trained “the AI system on all the patterns from the past on every single order, every single customer, and then we started to show the finance people, okay, AI recommends this, this credit block can be deleted”. But they also show reasoning, such as examples of “they pay two or three days late, but they pay”.

“In other cases, AI said no… we first should have the money from this customer.” In the end, says Rudy, people almost always followed the AI’s recommendation because they could see the reasoning and it backed up their own intuition and experience.

There are still exceptions, of course, and this is where the employee’s value comes to the fore. “Humans talking to humans, managing exceptions, is what really drives the value.”

Forrester Research and Process Intelligence

It isn’t just Rudy and Celonis that believe Process Intelligence plays an important role. Forrester Research goes further still.

“Recently I was attending a conference from Forrester, and they made some very bold predictions for 2026,” said Rudy. “They said Process Intelligence will rescue, not improve, but rescue 30% of all the failed AI projects in 2026. Why? Because Process Intelligence provides the missing context of processes.”

He goes on to call AI a “very good generalist” but limited. “You can ask a 10-year-old child, if everybody wants ice cream, should we produce more ice cream? Yes, of course, go ahead. But if you ask AI in the right context, AI will say, okay, we see a spike in demand in the EU for SKU one or two, whatever it might be, and the delivery time or replenishment time in our ERP system says ten days, but in 90% of all cases we see 20 days for the real delivery if we order by train.

“So, if we need to restock and reorder, I suggest to have it shipped by plane,” he said. “That’s an answer that actually has all the context, understands the process. You [don’t want] the generalist, but the specialist, knowing your situation, knowing where you come from, knowing the past, knowing the present, being able to predict the future, because this is where AI really provides value.”

Why do so many AI projects fail?

Sometimes it’s obvious why AI projects fail: when organisations simply tick a box. “Recently, I visited Australia, and we talked to a government agency, and we asked about what is your AI strategy. ‘Oh, we have a great strategy. We have just rolled out Microsoft Copilot to everyone.'”

To which Rudy said, great start but what’s next? “Nothing. Yeah, okay, that’s not a strategy. You know, I’m happy for Microsoft, but Copilot needs to be to be grounded in the real processes, needs to understand what’s going on before you can really use it.”

While that example sits at the extreme, Rudy is unimpressed by any company that “uses generative AI disconnected from the real business”. He cites chatbots as one example, which may deliver some value but certainly don’t justify a huge AI investment.

“These investments can only be justified if you really use it within your existing business, and you are faster,” he explained. “One of our customers in the US, a company called Vinmar, Vishal, the CEO, he… told his entire company, guys, we have to use AI, but not AI to replace you.

“I want every one of you to be more efficient, so let’s assume we grow the company by 100% we double our business, and I am not willing to hire any new stuff. So, look into your daily life, your daily work, and tell me, how can you use AI to be twice as effective and efficient as you are today? I think that’s a good approach.”

Can we all really become more efficient through AI?

I pressed Rudy on this: he really thinks everyone can improve their productivity by such a drastic amount? That he can? “I do a lot of thought leadership, so looking out for new trends, relating these trends to our technology. [For] core work, for example, for repetitive tasks, automating these, yeah, definitely, it makes me more efficient. I can produce more content. I’m better in my reasoning, text, emails, whatever.”

Perhaps not everyone can enjoy such an instant boost, but to open your eyes to possibilities he suggested I read – and everyone read – a book called AI 2041 by Kai Fu Lee and Chen Qiufan.

“In this book you find ten really captivating stories about ten visions for the future: autonomous driving, autonomous enterprise, deepfakes.” All current hot topics, despite the fact the book was written before ChatGPT arrived.

“They call it scientific fiction, because you first read the story, and then you get a scientific explanation why they believe this vision can become a reality, like in ten years, and it’s amazing. If you read it, it really will open your eyes, and it’s very easy to consume.”

I’ve placed my order, Rudy, and it’s due to arrive next week.

Lessons all organisations can learn

To finish, I asked Rudy for one final piece of advice to organisations that are thinking about how they best use AI.

“I have a history in automation, and in automation we typically automate repetitive and rule-based tasks,” said Rudy, “but if you think about decisions, many decisions are also rule-based and repetitive.” Like his earlier example of Cosentino in Spain.

“So if you have activities like this, ask yourself, is this really a decision a human needs to make, or can this be handed over to AI? And if it’s routine, if it’s a predictable outcome [like] deleting the credit block… let AI take the job, and let people talk to people, humans talk to humans, the company talk to their support, to their customers, to sales.

“This is where we need humans in the loop, to resolve exceptions. And that’s a pattern we see quite often, you know, routine tasks handled by AI and exceptions handled by humans.”