Few people in this series trusted a number on its own. Asked how their organisation decides what to trust in its product data, most described a chain instead, in which people and numbers check each other and someone stays accountable for the answer. Where that chain broke, the costly failure was rarely a wrong number. It was a missing one.
A Founder's Friend published the interviews between July and September 2026. This page sets out what we think they add up to for a team whose numbers depend on code another team ships, and seven questions to put to any number before it reaches a board slide. The people behind each claim are linked so you can check our reading against theirs.
The question behind the series
Each of the first thirteen issues ran under the headline “How does your organization make decisions based on your product data?” The brief Skene commissioned was narrower, and it is the one this page answers:
How does your organisation decide what to trust in its product data, and when does a bad number become a decision that costs you?
What the fourteen interviews add up to
Seven findings, in our words. Each names the interviews it rests on, so you can read the people and decide whether we read them fairly.
Data tests a decision; people still make it
Most interviews put the same order on things. A view forms from domain knowledge, customer calls or watching users, and data is brought in afterwards to confirm or kill it. When an idea keeps returning despite doubt from leadership and no supporting data, the anonymous founder in Issue 1 runs an A/B test to settle it and cross it off.
Arash Farhani trusts testimonials before any chart. Shahin Katebi's team bases its next feature on the domain knowledge of a co-founder with fifteen years in the field.
That is not a dismissal of data. It is a narrower and more honest job description for it, and it raises the stakes: if data is the validator, a broken event does not just leave a gap. It validates the wrong decision with full confidence.
Drawn from: Issue 1, Anonymous founder, Issue 4, Arash Farhani, Issue 8, Shahin Katebi
What happened is measurable; why is a conversation
Maria Petrova put it most cleanly: data tells you what happened and never why. AmirBahador Bahadori calls a dashboard pattern a hypothesis. Niko van Eeghen showed one metric meaning freedom to a patient, budget to a buyer and lost quiet time to a nurse, and only one of those readings appears on a chart.
The practical point is that each number needs a stated question before it needs a dashboard. A metric with no question attached gets read for whatever the reader already believes.
Drawn from: Issue 5, Maria Petrova, Issue 2, AmirBahador Bahadori, Issue 10, Niko van Eeghen
The expensive gap is the number nobody collected
Ask what a bad number cost them and people described absences. The anonymous founder's funnels record when a user leaves but not the repeated failed searches before it, which today can be found only by watching recordings by hand. Marc Olivier Meunier's team used about $500 of $250,000 in cloud credits because nothing measured it. Skene's own CEO, Teemu Kinos, makes the general case in the closing issue: an event added today answers questions in a few weeks, and never answers questions about the past.
A wrong number at least gets argued about. A missing one is noticed only when someone asks a question it would have answered, and by then the history is gone.
Drawn from: Issue 1, Anonymous founder, Issue 6, Marc Olivier Meunier, Issue 14, Teemu Kinos
Ownership comes before tooling
Dino Repo watched daily active users come out five different ways at a company with millions of them, and blamed nobody owning the data rather than any tool. The closing interview, with Skene's own CEO, described the same split from the other side: engineering says everything is tracked, the business says it could be much better, and the two answers come from teams judged on different KPIs.
Buying a tool does not create an owner. Naming one does, and then the tool has someone to answer to.
Drawn from: Issue 7, Dino Repo, Issue 14, Teemu Kinos
Agents changed who reads the data, and what breaks
The most AI-native teams in the series hit problems that analysts never had. John Adib's agents returned wrong numbers because event names were ambiguous to a machine. Isaac Kargar found agents making choices about data that nobody checked. Dino Repo worries about one misunderstanding being copied into twenty places by a system that generates from its own output.
Dino Repo's answer to the twenty-places problem is curation points: each generated action waits for a quick swipe from a person before it spreads. For product data, his answer is a cheap check that runs every time the build loop turns, confirming the product still records what it should, and he puts the setup at ten minutes. John Adib went upstream and made every event descriptive enough for a machine to tell apart, which he says reduced the wrong numbers a lot. Maria Petrova named the pressure behind a check like Dino's: building with AI makes shipping fast without making it self-documenting.
Drawn from: Issue 12, John Adib, Issue 9, Isaac Kargar, Issue 7, Dino Repo, Issue 5, Maria Petrova
Deliberate absence and day-one tracking are both right
Marc Olivier Meunier runs Astra without a dashboard so his team keeps calling customers. Maria Petrova and Dino Repo insist tracking belongs in from day one. Emad Ghorbaninia would rather ask than infer.
These look opposed and are not. Marc is talking about the value of reading the data at ten months, which is low next to a conversation. Maria and Dino are talking about the cost of adding tracking later, which is high.
Put together, our reading is to collect early and read late. Retrofitting is the expensive part, and you cannot interview last year's users about what they clicked. Emad questions whether gathering data should be the reflex at all, and his point holds either way: before opening a chart to guess what a customer wants, check whether a short call would answer it.
Drawn from: Issue 6, Marc Olivier Meunier, Issue 5, Maria Petrova, Issue 11, Emad Ghorbaninia, Issue 7, Dino Repo
Stage and access decide what is even possible
Nima Aksoy trusts market demand before he has users, the activation drop-off once they arrive, and board reporting at maturity. Niko van Eeghen reminded the series that some numbers are out of reach: an open-source framework is never observed in use, and some healthcare data may not be collected at all. Hamid Farmani's answer is to talk to people first, because the right product might be a calculator, not an app.
Drawn from: Issue 3, Nima Aksoy, Issue 10, Niko van Eeghen, Issue 13, Hamid Farmani
Seven questions to ask before you trust a number
Each question comes from one interview. Put them to any metric before it goes into a plan, a board slide or a decision to cut something.
Which layer did this number come from? A database count, a traced error rate and a behavioural funnel deserve different amounts of trust, even on the same dashboard. AmirBahador Bahadori makes the case in Issue 2.
Who owns this metric, by name? When nobody owns it, the same metric comes out several ways depending on who pulls it. Dino Repo describes it in Issue 7.
What does this number not see? Funnels show when a user drops off but not the repeated failed searches that came first, and today only recordings watched by hand reveal those. The anonymous founder describes it in Issue 1.
Could a machine reading this event list pick the right event? Agents querying analytics returned wrong numbers when event names were ambiguous. John Adib saw it happen in Issue 12.
What are the competing explanations, and what would separate them? A metric says what happened. A call or a test says why, depending on how many customers you have. Maria Petrova draws the line in Issue 5.
Who does this metric touch, and does it mean the same thing to each of them? One metric can mean freedom, budget and loss to three different people. Niko van Eeghen shows how in Issue 10.
Which question will we wish we could answer in six months? An event added today answers nothing about the past, so the cheapest tracking is the tracking added before the question arrives. Teemu Kinos makes the point in Issue 14.
The same failures, read framework by framework, are in 14 GTM frameworks and the product events each one assumes. To check the events behind your own numbers, the free audit maps what your product collects today, once, with no card.
All fourteen interviews
- Interview 01: the founder who uses product data to retire a recurring idea
Issue 1, 24 July 2026. Product data confirms or kills a hypothesis that people formed first, and the expensive gap is the number nobody collects. - AmirBahador Bahadori: product data is three layers, not one
Issue 2, 30 July 2026. Product data is three separate layers that answer different questions, and a dashboard pattern is a hypothesis until a person confirms it. - Nima Aksoy: demand is the first dashboard
Issue 3, 6 August 2026. Which data deserves trust depends on the product's stage: market demand first, the activation drop-off next, financial reporting at maturity. - Arash Farhani: the first data he trusts is a testimonial
Issue 4, 9 August 2026. Before a few hundred users, intuition and testimonials lead, and a number earns authority only once direct contact with clients confirms it. - Maria Petrova: data tells you what happened, never why
Issue 5, 12 August 2026. Data establishes what happened and never why, and tracking is infrastructure to build before the first real user, not something to add later. - Marc Olivier Meunier: why Astra runs without a dashboard
Issue 6, 18 August 2026. Trust lives in an evidence chain a person can check, so at ten months the team keeps its own analytics near zero and talks to customers instead. - Dino Repo: one metric, five numbers, and a ten-minute fix
Issue 7, 19 August 2026. Untrustworthy data is an ownership problem before it is a tooling problem, and the cheapest fix is a product-data check that runs every time the build loop turns. - Shahin Katebi: data finds the stuck user, a person asks why
Issue 8, 20 August 2026. When users are not from the tech world, domain expertise decides what to build, and product data has one narrow job: locating where a user got stuck. - Isaac Kargar: the agent decision nobody checked
Issue 9, 21 August 2026. Trust belongs to the harness that decides what to extract, keep and forget, and the dangerous failure is an agent's decision that nobody verified. - Niko van Eeghen: one metric, three meanings at Evondos
Issue 10, 22 August 2026. The number is rarely the hard part: one metric means different things to different people, and in parts of the economy you cannot or may not collect it. - Emad Ghorbaninia: five analytics tools or one five-minute call
Issue 11, 28 August 2026. The data he trusts most is not tracked but asked for, and anything that reaches a client passes a human review first. - John Adib: when agents read your analytics, event names break first
Issue 12, 29 August 2026. Once an agent rather than an analyst queries the analytics, ambiguous event names stop being hygiene and start producing wrong numbers in the business review. - Hamid Farmani: talk to people before you track anything
Issue 13, 30 August 2026. Big organisations earn trustworthy data through interdependence and documentation; solo, he builds that discipline for his agent, and trusts a real person's need first. - The sponsor answers The Question: Teemu Kinos, cross-examined
Issue 14, 15 September 2026. Nobody wants to track things; they have questions, and the questions arrive while someone is using the product or talking to a user, never at development time.
About the series
A Founder's Friend published The Question on Substack between 24 July and 15 September 2026, edited by Nicolas Dolenc. Thirteen interviews ran first, and a closing issue put the same question to Skene's own CEO. Some issues are public and some are for subscribers; each page here links its original and quotes only what the public can read.
Skene sponsored this research and commissioned it on a single question. A Founder's Friend ran the interviews, wrote them up and published them, and the series states that Skene paid for the research and did not direct the questions, the analysis or the conclusions. Each of the thirteen interviews was reviewed and signed off by the person in it, named or anonymous by their choice. Appearing in the series is not an endorsement of Skene, and nobody here is presented as a Skene customer.