Data always tells you what's going on.
It never tells you why.

A conversation with Maria Petrova — ValueLab

Interview by Nicolas Dolenc · 2026-08-04

Portrait of Maria Petrova, co-founder of ValueLab — Maria Petrova, co-founder, ValueLab

How does your organisation decide what to trust in its product data — and when does a bad number become a decision that costs you?

In one line

Trust in product data isn't a binary for Maria Petrova. Data is trustworthy for establishing what happened and untrustworthy the moment anyone asks it why — at which point human judgment takes over, calibrated to how many customers you actually have.

Trust the “what.” Never let it answer “why.”

“Data always tells you what's going on. It never tells you why people behave this way or another.” A metric showing that a feature isn't used is reliable as a fact and empty as an explanation. It produces competing hypotheses — a visibility problem or a motivation problem — and cannot choose between them.

The cost isn't a wrong number. It's an unresolved “why” that gets acted on anyway.

Her example is deliberately mundane: nobody clicks. Make the button red, or accept that the journey leading to it is wrong. Same number, opposite responses, and only one of them is right. Acting before resolving that is where the expense begins.

How you resolve it depends on customer count, not on principle.

Handful of enterprise accounts: “the easiest thing is to ask them, because they will tell you.” Thousands of consumers: one call gets you one unusually opinionated person, so you run experiments instead — and if a few tweaks can't move behaviour, that's a signal to pivot rather than to keep tuning.

The most expensive bad number is the map, not the metric.

Her costliest observed failure isn't a broken dashboard — it's a company mistaking its own service blueprint for its customer's journey and optimising confidently against a document about itself. “That's the perspective of whoever develops the solution, not of its users.”

The second most expensive is the number you never collected.

Her closing, unprompted point. “One trackable action is worth a 1000 words.” Self-reported intent and actual behaviour diverge; AI-assisted building ships fast but doesn't self-document; instrumentation gets harder to retrofit the longer you wait. The failure mode is complete and familiar: you have this application, nobody's adopting it, and you don't know why.

What stays with people.

Not taste — she finds the word underspecified, even while defining it well. Empathy: connecting online and offline context about a person to work out what would help. Models can accelerate surveys, transcripts and pattern-finding; observing a user and reading a situation still belongs to a human.

Data always tells you what's going on.It never tells you why.