A Fireside Chat with Roche at the Agentic Analytics Playbook, London
If there's one thing that stuck with me from Yannick Misteli's session at the Agentic Analytics Playbook event in London, it's this: most AI pilots don't stall because of technology or budget. They stall because nobody answered the "day after" questions.
I had the opportunity to sit down with Yannick Mistelli, Head of Engineering at Roche, the global pharma company with 100,000+ employees and heavy regulation across 25+ countries.
In other words, if Roche can operationalize agentic analytics at scale, the "you have to be small and nimble" excuse stops holding water.
Yannick cut straight to the pain points data leaders bring up in nearly every conversation: pilots that never reach production, governance fears that block democratization, pressure from leadership while data still isn't perfect, and business users who will never become data scientists but still need answers.
Here's what landed.
The Top Agentic Analytics Takeaways from Roche
1. Leadership is the Blocker, Not Technology
Yannick was blunt: sponsorship and budget for AI are easier to get than ever. The reason pilots fail is that leadership doesn't define what comes next.
These are the questions leaders must answer before scaling, not during:
Who owns the pilot afterward?
Who is responsible for what?
What governance runs at scale?
What level of risk is acceptable?
If you don't define the operating model that follows the pilot, your pilot becomes a permanent experiment.
2. Get Comfortable with Imperfection (By Being Explicit About Trade-Offs)
One of the most practical moments in the session came when Yannick talked about standards. The temptation is to lower them to move fast; the better path is to be explicit about where you will and won't compromise.
At Roche, AI governance, security, and lineage are non-negotiable. But completeness and latency? Acceptable trade-offs, which is why their dashboards run on nightly batch updates rather than real time.
"You don't need a perfect data foundation before doing AI," Yannick said. His counter: have a solid foundation, then "start small, think big, iterate fast."
Data leaders who feel pressure to have everything ready before launching should take note. Run data readiness and production deployment in parallel rather than waiting for a perfect warehouse.
3. Conway's Law Explains Your Silos
Data silos and "too many technologies" are complaints Yannick hears all the time. His framing cuts deeper: "The technical artifacts are just a mirror of our organizational problems."
That's Conway's Law: if your org is siloed, your tech will be siloed, and no new platform fixes that on its own.
This is why Yannick's throughline across the entire session was people, process, and technology together. Treating AI as just a technology to scale while ignoring the other two? He called that his biggest mistake.
Trust-First Adoption Among Non-Technical Users
Roche's adoption target wasn't data scientists; it was 20,000+ go-to-market professionals across commercial teams who will never learn SQL.
What worked wasn't persuasion. "Trying to convince people to change doesn't work," Yannick said. Building a trust relationship first, where users believe you have their best interest at heart, is what drives adoption.
Does that approach sound soft? Look at the numbers.
Roche now has more than 40,000 users on ThoughtSpot, with 5,000 logging in every month; commercial teams hit roughly 2 million queries in a single month.
Roche users are the second most active Spotter users globally, scaling to 500 active Spotter users within a year.
That adoption curve doesn't happen without trust.
How to Bring Insights into the Workflow
At the core is a key framework: analytics spans descriptive, diagnostic, predictive, and prescriptive questions. But the real shift is insights coming to you inside your workflow, not you going to a BI tool.
Roche embedded analytics into the tools teams already work in: Salesforce, Google, Slack, Teams. When you meet people where they are, they feel like they've been given "new superpowers."
Embedded analytics becomes the difference between a tool people visit and a capability they rely on.
The Biggest Myth, Mistake, and Advice from Roche
To wrap up the session, Yannick shared his take on:
Myth to retire: You need a perfect data foundation before doing AI. Start small, think big, iterate fast.
Biggest mistake: Treating AI as just a technology and adding complexity. Simplify rather than layering on new committees.
Advice for pilot mode: Get the agent into users' hands with the right use case, right risk, and good-enough data.
What to Take Away
If pilots are stalling, governance is causing anxiety, or adoption isn't sticking, Roche's approach offers a clear starting point. Above all, bring analytics into the tools your team already lives in.
Roche didn't fix people, process, and technology one at a time: they worked on all three together.
Ready to operationalize agentic analytics? Start your free trial today.
