In July 1994, at the Estádio Universitário in Lisbon, the junior hurdlers of the world lined up for the 400 m final. Ukraine's Hennadiy Horbenko won in 50.56. Second, 0.29 seconds behind him, was a Hungarian teenager named Miklós Róth.

Three decades later, Róth runs an AI marketing agency, a research lab and an AI consultancy. It would be easy to treat the medal as a nice line in the "About" section. But the event he chose says a lot about how he works now, so it deserves to be taken seriously.

Why is the 400 m hurdles such a demanding event?

Because it punishes every kind of shortcut. It is too long to sprint all the way and too technical for pure endurance to save you. The geometry never changes: the first of ten hurdles stands 45 meters from the start, the rest follow every 35 meters, and the last 40 meters is a flat run to the line.

Inside that geometry, an athlete has to solve three problems at once. The stride pattern has to be settled early, because the number of steps between hurdles is decided, not improvised. That rhythm has to hold while fatigue builds, and it typically turns brutal around the seventh hurdle. And form has to survive exactly when it starts falling apart, because one sloppy clearance costs more than a slightly slower stride.

After Lisbon, Róth took that discipline to the University of Nebraska, where he studied finance and was part of the team that won the distance medley relay at the 1996 NCAA indoor championships. Relays add a lesson of their own: your result depends on hand-offs to people you trust.

How does a hurdles race map onto an AI transformation?

Almost phase by phase.

The start: decide early. In the race, the first 45 meters fix the stride pattern for everything that follows. In an AI rollout, the first decision does the same. Pick the wrong first use case, something flashy but disconnected from revenue, and the whole program loses its stride. That is why Róth's work at Roth AI Consulting opens with a 20-minute diagnostic designed to produce exactly one clear priority. Not ten. One.

Hurdles one to six: build a cadence. Good hurdlers don't win with heroics over a single barrier. They win by clearing each one the same way. Organizations adopting AI need the same unglamorous consistency: a weekly rhythm of shipping, reviewing and adjusting, rather than one big launch followed by silence.

Hurdles seven to ten: hold your form. This is where races are lost. Fatigue arrives, technique collapses, and athletes start clipping barriers they cleared easily a minute earlier. In companies the equivalent moment comes a few months into an AI program, when enthusiasm fades and shortcuts creep in: unchecked outputs, shadow tools, nobody owning the risks. That is precisely when governance matters most, including the rules a company needs under the EU AI Act.

The run-in: finish through the line. Sprinters are coached to run through the finish, not to it. AI programs often stop at the demo, with a pilot that impressed everyone and changed nothing. Finishing means measuring business outcomes, not activity.

Does elite sport really translate to business?

Partly, and it's worth being honest about where the analogy breaks. A race has fixed rules, a known distance and a clock nobody can argue with. Business has none of these. Competitors move the hurdles. The finish line drifts. Nobody hands you a result sheet.

That is why Róth didn't stop at sporting metaphors. His S·I·C·T framework, developed at the Roth Complexity Lab, is in a way the grown-up version of the hurdles lesson: a system stays stable while its structure and cohesion keep up with the information and the pressure for change flowing through it. A hurdler's stride pattern is structure. The trust between relay teammates is cohesion. The fatigue and the barriers rushing at you are transformation. Here is the framework in full.

What sport does give you is a rare kind of feedback. In athletics, results are public, timed to the hundredth of a second, and preserved in record books long after the athletes move on. Róth's 50.85 from Lisbon still sits in the official results. In business, and especially in AI, claims are far harder to check. That appetite for verifiable results shows up in how he runs his lab, where every claim carries a label for how strongly it is supported.

What does this mean for leaders planning an AI program?

Three questions, borrowed straight from the track:

1. What is our first hurdle? Name the single use case that sets the rhythm. If you can't name it, you're not ready to run.
2. What is our stride pattern? Agree on the weekly cadence of building, testing and reviewing before the program starts, not halfway through.
3. Who checks our form at hurdle seven? Make someone responsible for quality and governance when the program is tired, not only when it's new.

If those three answers exist, the odds of finishing improve dramatically. If they don't, no amount of speed in the first hundred meters will help.