RevOps AI Earns Its Keep Answering The Questions Dashboards Never Could

For years, data quality was the unglamorous background task of revenue operations, a thing you cleaned up when there was time. AI has ended that. The moment a company connects its CRM and go-to-market tools to AI and opens the floodgates, every weak spot in the underlying data stops being a tolerable flaw and becomes the reason a use case collapses. The work that used to sit at the bottom of the RevOps priority list is now the input that determines whether any of the AI investment pays off, and the teams treating it that way are getting real leverage rather than a pile of abandoned experiments.

Victor Embrechts is Head of Revenue Operations at the expense-management software company Rydoo, where he owns the data and systems behind the go-to-market motion. He came up through RevOps roles in Belgium before moving to Portugal, and has spent the past year watching his function move from scattered AI tinkering to something far more structured and consequential.

"When the data quality is poor before it's connected to AI, all the use cases fall apart," Embrechts says. That single dynamic has reordered what his team spends its time protecting.

From siloed experiments to structured implementation

In Rydoo, Embrechts describes a rapid AI evolution over two phases. He describes last year as scattered and individual, a phase of siloed experimentation without a real shared foundation. "Some people bought a ChatGPT subscription. Others did something else," he shares.

This year hit an acceleration point when the company bought AI access for everyone and wired it into the CRM and the full GTM stack. What followed was a course correction. Rather than everyone building private one-off automations, the company converged into what he calls an AI guild, cross-functional teams of engineers, salespeople, and others building durable, adoptable workflows. "Instead of 'I built something that saves me 10 minutes a day,' now AI goes broader and deeper into our full stack," he says. The shift is from personal shortcuts to shared infrastructure.

One of the most concrete applications Embrechts sees is lead qualification, and it shows how he divides labor between automation and AI. The hard criteria stay fixed: a lead that meets the thresholds on company size and location is routed to sales automatically, as it always was. What changed is what happens to the leads that fail those rules. "All the ones that don't get sent flow into an AI agent that does more research."

The agent scrapes the prospect's website and does the deeper digging a static rule can't, catching the cases the old model discarded: the department head at a large company who listed only their team's headcount, or the small firm in an industry where nearly everyone is a fit. He's deliberate about the boundary. Standard automations still handle the clean cases that work every time. AI is reserved for the judgment calls at the margins, deciding whether an edge-case lead is worth a salesperson's time.

Two departments, two different AI stories

Because Rydoo's marketing and sales already share revenue targets under one leader, the classic silo between them barely existed to erode. The more interesting split is in who gets more out of AI, and the answer comes down to volume. "Marketing gets thousands of leads every month, while for a salesperson with the correct qualification they maybe get five opportunities a month," he explains. At that level, a rep can qualify by hand, but at thousands of leads, AI is transformative.

That gap is compounded by a deliberate choice to keep the sale itself human. As a premium-priced product, Rydoo wins partly on the trust that real people, not just software, will deliver, so the sales motion stays intentionally manual even as marketing automates at scale. That efficiency shows up in the numbers rather than in headcount. Embrechts says the goal has been to do more with the same people, optimizing workflows so the existing sales team closes more and bigger deals instead of expanding the roster.

The clearest result is win rate. "Our pipeline creation per month stayed quite stable, but we can close more new logo ARR because our win rate increased by 10%," he says, a year-over-year gain he attributes to heavy investment in training, enablement, and tooling that made the same sellers more effective. It's a blended approach in lieu of a replacement of people with software, aimed at raising the output of the team already in place.

Some fields belong to humans on purpose

Even as he automates, Embrechts draws firm lines around data he insists a person must own. His team runs a meeting-recording tool that transcribes every sales call, but he refuses to let AI populate certain fields from those transcripts. Competitor tracking is one example. "I want the sales team to fill in manually which competitors we're going up against," he says, because a name surfacing in a transcript is ambiguous. A competitor might be mentioned as a past vendor, a current evaluation, or an offhand reference, and only the rep knows which is which. So, the system forces the choice at a specific moment: to advance an opportunity to stage two, after discovery, the rep has to name the competitors and stand behind that call. Historical gaps are where he's willing to take more risk, parsing old transcripts and closed-lost notes in bulk to backfill data the company never captured systematically.

Where AI earns its keep, in his view, is in questions that used to be unanswerable. Straight Salesforce analysis doesn't need it. "But, if I have some Salesforce data and I have some back-end product usage data and I have an Excel file from the finance team with costs in it, AI is very nice for structuring and correctly matching which opportunity is related to which customer," he explains.

The result is nuance RevOps could never deliver on demand before. A regional team can now get a real answer to a narrow question, like the historical win rate for a specific industry in the UK, and use it to reconsider where to focus. "Sometimes we have ICP criteria and people don't know why they're ICP criteria," he says. Being able to show the win rate and shorter sales cycle behind a target, and tie it to commission, is what gets sellers to believe the strategy.

Speed doesn't transfer the accountability

For all the leverage AI enables, Embrechts is pointed about the risk, and about where responsibility ultimately sits. His team never lets AI make decisions cold. New use cases run a pilot phase where the AI only suggests, monitored until it clears a bar like 98% accuracy over a month or more, with the human feedback fed back into the model before it's trusted to run on its own. Even then, it's still checked monthly.

Underneath the process is a principle Embrechts tries to drill into everyone on his team. "Every person is responsible for the output they generate, manually or with AI. It's still you as a human that is responsible," he says. Better tools don't dilute ownership, and output that's wrong isn't excused by the fact that AI produced it. In a function racing to automate, his discipline is a reminder that speed changes who does the work, not who answers for it.