There is a moment Guillaume Bacuvier, CEO, ARIS, keeps returning to, and it is not a technical one. It is the moment a senior executive has to decide whether to switch something on.
“If you’re a decision maker in a company, you would be willing to press the button in confidence and say, okay, I’m going to deploy AI agents and give them large degrees of autonomy and decision making and running a particular process,” said Bacuvier, CEO, ARIS, in a conversation with GEC Media at Riyadh during LEAP 2026. Almost nobody, in his account, is willing to do that yet. “Because those guardrails, those controls don’t exist.”
Bacuvier took over the German process intelligence company in January 2026, arriving from Kantar Worldpanel and, before that, dunnhumby and more than a decade at Google. ARIS itself is only recently a standalone business, carved out of Software AG and backed by Silver Lake, and Bacuvier has been assembling a new executive bench around a single strategic bet: that the constraint on enterprise AI is no longer model quality but organisational self-knowledge.
The productivity numbers are missing because the control layer is missing
Bacuvier’s starting point is the research everyone in the industry has now read and nobody enjoys quoting. Ask how many companies have deployed AI at scale, in production, in a way that has changed how the business runs, and the answer keeps coming back small.
“Every research report that comes out gets percentages that vary, 15, 20, 30%, depending on the definition,” he said. His explanation for the gap is not about capability. It is about the absence of anything resembling operational oversight.
“We’re still not at a point where most companies can, at any point, get a report that says here’s how many agents are running in my processes, here’s what they’re doing, here’s what they’re not doing, here’s the percentage of agent-led processes that have complied with the rules, which ones have gone wrong.”
Until that exists, he argues, the money keeps moving and the output does not. “People are spending money but not yielding results,” he said, connecting the missing governance layer directly to the wider argument about why AI investment has yet to show up in productivity statistics.
“A lot of companies rush and try to deploy AI when they can’t even describe how they work today. So when you’re laying your AI on top of the mess, you get a bigger mess.”
Governance is being retrofitted, and retrofitting is the expensive way
Put to him that governance has become fashionable only after the fact, and that vendors are now attaching it to products designed without it, Bacuvier did not push back.
“There’s a bit of a fear of missing out effect. If I’m too slow, my competitor is going to do it faster, I’m going to be in trouble,” he said. “So people have rushed and tried to deploy AI half the way, and to your point, often trying to solve some problems after the fact. So I deployed agents in a process, and then, now I need to have governance. That typically is very difficult to do.”
What he expects next is a correction rather than an acceleration. Companies that have spent tens of millions, in some cases hundreds of millions, with little to show will step back and redesign processes with governance built in from the start. He was careful about how much of this he claimed as vindication for his own company, though the commercial logic is not subtle: designing the process and designing the governance framework that runs through it is what ARIS sells.
The company’s oldest product became relevant for a reason nobody planned
The most interesting part of Bacuvier’s case is that it rests on something ARIS was doing long before the current cycle. The platform’s core function is a detailed description of how an organisation actually operates, covering processes, IT systems, decision rights and the connections between them, which the company refers to as a digital twin of the organisation.
“Before AI, only a minority of companies saw the value and had the willingness to invest in building that discipline,” he said. “What’s changing today is you actually cannot deploy AI if you have not built that kind of digital twin, because the AI models need that detailed context in order to work from them.”
His framing of the addressable market shift was blunt. A recognised product in an established category can now, he said, address a market potentially 10 to 100 times larger, “a little bit by chance, because it was doing it before AI.”
Asked the obvious challenge, why a well-funded startup could not simply build an agent that does the same thing, he went back to prompting. The quality of any model’s answer depends on the quality of the context it receives. Scale that principle to an organisation with hundreds of thousands of employees, operating across multiple jurisdictions and product lines, and supplying that context becomes the hard problem rather than an incidental one.
“We are not in the business of building agents,” he said. “We are in the business of giving agents what you could call a super prompt.”
Saudi’s advantage is administrative, not infrastructural
The Gulf reads, in Bacuvier’s account, as an unusually favourable environment, and the reason he gave has little to do with data centres.
“Even prior to AI, there was a strong culture in this part of the world to be disciplined in managing your processes and documenting your company,” he said, pointing to a customer roster across both private and public sectors built over years. Organisations that already documented themselves for compliance and governance reasons are, by his reckoning, closer to the starting line for agent deployment than peers elsewhere.
The infrastructure investment, the data centre build-out and the partnerships with US and European technology companies compound that, but they are not the differentiator he cited first.
His plan for the region is narrow by design. ARIS is in active discussion with a shortlist of 5 to 10 organisations in the Middle East, and Bacuvier defined success in terms that are easy to check.
“If we meet again in a year from now, I can publicly reference a few organisations in the area. I can say, talk to these people, and they’ll tell you how you can deploy AI successfully at scale, in partnership with us and others.”
What he tells the people who have to sign off
Bacuvier is careful about giving advice on AI, which he acknowledged is a dangerous business given how much is still unknown. What he offered instead was a sequencing argument.
Build a filter for the noise. Separate what is genuinely known from what remains unproven. Then establish whether you can describe your own organisation to an AI system before you attempt to deploy one inside it.
“Anytime there’s been a big technology cycle, a lot of money is wasted at first, because people rush and they get influenced by all the hype, all the vendors, all the noise,” he said. “And there’s always this impression that people are doing it better than you, which is actually not true.”
On employment, his position was measured and specific about the current limits of the technology. Even in the most advanced organisations, he said, AI functions as a very effective way to speed up information processing and assist interpretation rather than as something that makes decisions. Judgment and accountability, for now, stay with people.
“Employees should be far more productive, should be focusing on high judgment and high agency work, and can automate a lot of the repetitive run work of information processing, that is usually the lot of most employees.”
It is a modest claim by the standards of the week he made it in, delivered in a city where the ambition on display is anything but modest. Bacuvier’s argument is that the modesty is the point, and that the organisations willing to do the unglamorous documentation work first will be the ones still standing when the shortlist of credible reference customers gets written.
