Five years ago, the agency chief technology officer was essentially the head of IT: keeping the tools running, managing vendors, and staying out of the way of the people who made the real decisions. That role has flipped. Today’s agency CTO increasingly sits at the table where strategy gets decided, not down the hall from it. Nobody has watched that shift more closely than Olivier Pepin, the CTO of NOVUS Media.
According to Olivier, two forces drove the change at once. Privacy regulation, the end of third-party cookies, and walled gardens hoarding their best signal meant agencies could no longer rent their way to good targeting. Building a first-party data capability stopped being optional. At the same time, AI collapsed the distance between raw data and a business decision. “When a model can turn raw performance data into a budget recommendation, the bottleneck stops being ‘do we have the insight’ and becomes ‘is our data trustworthy enough to act on,’” Olivier says. “That’s an engineering problem.” In his view, that is exactly why the CTO moved from the basement to the table.
From Manual Craft to Governed Infrastructure
At NOVUS Media, that shift shows up in a business that has been doing geography-first intelligence since 1987, answering ZIP code by ZIP code which markets deserve investment and what the people who live there actually respond to. Olivier is quick to point out that the intelligence itself isn’t new. What changed is the machinery underneath it. “The moment our local intelligence stopped being something a smart analyst produced market by market, and became infrastructure, a governed data layer where every campaign feeds back in and the picture compounds, technology stopped supporting the business and started running it,” he says. Most agencies, he adds, treat data as fuel that gets burned every quarter. NOVUS treats it as infrastructure that gets better the longer a client stays.
Ask Olivier to explain his job in plain language and he skips the jargon. “I build and manage the systems that answer three questions before anyone spends the next dollar: which markets are actually worth the money, who really lives in them, and what will those people respond to,” he says. Then comes the harder part: proving whether the spend turned into something real, like subscribers, patients, or sales, measured against a client’s own numbers rather than impressions.
Build What Compounds, Buy What’s Commodity
When it comes to deciding what to build in house versus buy from a vendor, Olivier runs everything through one test. “Build where it compounds, buy or rent where it’s a commodity,” he says. A data layer, audience models, and measurement all get more valuable the more they’re used, so those stay proprietary. Cloud computing and other widely available tools don’t set anyone apart, so NOVUS uses them without hesitation. “Rebuilding a solved commodity is just ego with a budget,” Olivier says.
That discipline extends to AI itself. His rule is to start where AI removes friction from the team’s day: automating manual QA, anomaly detection, and data orchestration. Prove the value of boring, high frequency work first, he says, before chasing moonshots.
A Commerce Background Changes What You Notice
Olivier spent twenty five years in technical architecture for brands like Nestlé and AstraZeneca, and he credits that background with shaping how he thinks about agency technology today. “When you build and run the platform where the customer checks out, you can’t hide behind activity metrics,” he says. “There’s a number at the end of the funnel and it’s either going up or it isn’t.” That experience, he says, forces you to care about the entire chain rather than just your own slice of it, and teaches you to build systems that hold up under conditions you don’t fully control.
The Layer Nobody Wants to Fund
Data engineering rarely gets the investment it deserves, in Olivier’s view, because it’s invisible when it works and nobody likes funding invisible. A single brand’s media might run across a dozen platforms, from connected TV to streaming audio to broadcast, each reporting back in its own format with its own definition of a conversion. The job, as he describes it, is turning that pile of fragments into one governed source of truth that every dashboard, report, and AI model downstream can rely on. “A beautiful insight built on bad data is worse than no insight,” Olivier says.
Winning the Pitch Now Requires the Machinery
Clients now bring their own data and technology people into the pitch room, Olivier says, and they want to see the machinery: how an audience gets built, how a campaign’s impact gets proven rather than just claimed. Agencies that can’t answer those questions don’t make the shortlist anymore. Technology is also what makes a client relationship sticky once it’s won. When an agency plugs directly into a client’s systems and first party data, Olivier says, it stops being a vendor that can be swapped out and becomes part of how that business actually runs.
Where the Role Goes Next
Olivier expects the CTO, platform, and data functions to consolidate under one owner over the next few years, closing the fragmentation that slows agencies down today. The bigger shift, he says, is in pricing. The hourly billing model has been losing value for three decades, and AI just put a clock on it. “The agencies that survive the decade will be the ones that reprice around the outcomes they deliver instead of the hours they bill, running on technology they own instead of technology they rent,” Olivier says. The ones still selling hours, in his view, are optimizing for a model the market has been quietly ending for thirty years.
