06 · Building Product Culture¶
Culture is the least tractable-sounding word in management and one of the most concrete things a product leader actually controls. The trouble is that almost every attempt to build it starts in the wrong place — with values on a wall, an offsite, a set of adjectives — and adjectives do not survive contact with a quarter-end.
A more useful definition, and the one this module uses: culture is the set of behaviours that reliably happen when you are not in the room. It is produced by three things, in descending order of power: what gets rewarded, what gets tolerated, and what gets said. Values statements are the third category, which is why they change nothing on their own. Promote a PM who ships without evidence and you have communicated your real standard more loudly than any document.
At VP level this matters because it is your only remaining lever at scale. With 76 people in R&D you cannot review the decisions; you can only shape the conditions under which they get made.
Norms, and the artefact that encodes each¶
A norm that lives only in people's heads decays with every hire. A norm encoded in an artefact — a template, a checklist, a recurring forum, a default in a tool — survives.
| Norm | The artefact that encodes it | What happens without it |
|---|---|---|
| Decisions are written down | A one-page decision record with owner, options, and rationale | The same argument is re-litigated every quarter |
| Claims come with evidence | The "how do you know?" column in every review template | Confident people win over correct people |
| Experiments are pre-registered | A registry with hypothesis and success criteria filed before launch | Post-hoc rationalisation of every result |
| Negative results are published | A standing agenda item in the monthly product review | Teams learn to hide failure, then to avoid risk |
| Customers are met directly | A quota — every PM does 3 customer conversations a month, logged | Product opinions drift into internal folklore |
| Strategy is repeatable by anyone | The one-page strategy, restated in every all-hands | Eight teams optimise eight different things |
| Disagreement is aired, then committed to | An explicit disagree-and-commit protocol with a named decider | Passive resistance in the execution phase |
| Bad news travels up fast | Leaders visibly thank the person who brought it | You find out from the churn report |
| Launches have pre-set success criteria | Criteria fixed in the launch doc before code is written | Everything is a success in retrospect |
| Postmortems are blameless | A template with no "who" field and a published output | The cause is recorded as "human error" and recurs |
The test for any norm you claim to hold: name the artefact. If you cannot, you have an aspiration, not a norm.
The four cultures product orgs drift into¶
| Culture | Optimises for | Feels like | Fails at | Common in |
|---|---|---|---|---|
| Craft | Quality of the thing built | Design reviews, high standards, slow | Speed; commercial urgency | Design-led companies |
| Velocity | Throughput | Ship weekly, dashboards everywhere | Coherence; nobody owns the whole | Growth-stage PLG |
| Consensus | Alignment | Everyone consulted, few things decided | Speed; strong opinions leave | Post-IPO, large orgs |
| Command | Speed of decision | The founder or VP decides | Scaling; the org stops thinking | Founder-led companies |
None is wrong; each is a trade. The failure is drifting into one without choosing it, then hiring people suited to a different one. ListUp is velocity-leaning and needs more craft, which is a sentence worth saying out loud at an all-hands, because it tells 76 people which way to lean when the trade-off appears on a Tuesday afternoon.
Diagnosing the culture you actually have¶
Adjectives are unmeasurable. Behaviours are countable. Pick behaviours that would be expensive to fake.
ListUp's R&D culture diagnostic, run over one quarter across 76 people:
| Behaviour | Measurement | Baseline | Target | Reads as |
|---|---|---|---|---|
| Experiments pre-registered | Registry entries ÷ experiments run | 34% | 90% | Results are decided after the fact |
| Negative results published | Count per quarter | 2 | 10+ | Failure is hidden |
| Launches with pre-set success criteria | Share of launches | 48% | 90% | Everything succeeds retrospectively |
| PM customer conversations | Per PM per month | 1.4 | 3.0 | Opinions are internal |
| Decisions with a written record | Share of significant decisions | 29% | 80% | Re-litigation is routine |
| Strategy stated consistently | Variance across 12 random skip-level answers | 6 distinct answers | Under 3 | The strategy is not landing |
| Postmortems published | Share of Sev-1/Sev-2 incidents | 62% | 100% | Learning stops at the team boundary |
| Bad news lead time | Days between a team knowing and leadership knowing | 11 | Under 3 | Escalation feels unsafe |
| Regretted PM attrition | Trailing 12 months | 12.5% | Under 10% | See exit interviews |
| Time in recurring status meetings | Person-hours per week, R&D | 88 | Under 30 | Coordination has replaced work |
The last two rows are the ones executives skip and the ones that cost most. Take the meeting line seriously.
The meeting audit, costed¶
Four recurring status meetings, average 22 attendees, one hour, 50 weeks a year:
| Line | Value |
|---|---|
| Person-hours per year | 4 × 22 × 1 × 50 = 4,400 |
| Fully loaded cost at $97.50/hour | $429,000 |
| Written replacement: 4 authors × 25 min + 22 readers × 8 min, weekly | 670 hours |
| Cost of the written version | $65,325 |
| Annual saving | 3,730 hours, $363,675 |
$363,675 is roughly 1.9 engineers, recovered permanently, from a change that costs nothing to make. And the hours are the smaller benefit: a written update forces the author to have a position, and it can be read by someone who was not invited, which is how context spreads in an org too big to fit in a room.
The rule that makes it stick: a recurring meeting must have a decision to make. If its purpose is for people to hear what happened, it is a document. Audit every recurring meeting annually against that one test and kill the ones that fail, rather than shortening them.
Building the norms deliberately¶
Culture change is not announced, it is instrumented. Four levers, in descending order of power:
| Lever | Mechanism | Example at ListUp | Time to effect |
|---|---|---|---|
| What gets rewarded | Promotion and calibration criteria | Add "publishes negative results" to the Staff PM bar | 1–2 cycles |
| What gets tolerated | What you let pass in review | Stop accepting launches without success criteria — in public, once | Immediate |
| What you model | Your own visible behaviour | VP publishes a written postmortem on a decision they got wrong | Immediate |
| What you say | Strategy memos, all-hands | Repeat the one-line strategy in every forum for a year | 2–4 quarters |
The second row is the fastest and the most uncomfortable. Culture is defined by the worst behaviour a leader walks past, and the first time you decline to walk past something is the moment the norm becomes real. Doing it once, visibly, in a monthly product review, is worth more than a year of memos.
Worked example — ListUp's evidence norm¶
Diagnosis: 34% of experiments pre-registered, 2 negative results published in a quarter, and a monthly product review where the team with the best story consistently won the argument. The consequence was measurable in Level 4, Module 1's org-health data: only 48% of launches had pre-set success criteria, so nothing was ever falsified and the roadmap accumulated features nobody could show worked.
What was tried first, and failed: a values memo on "evidence over opinion", sent once, referenced by nobody after week two. Memos change what people say, not what they do.
What worked, over two quarters:
| Change | Type | Detail |
|---|---|---|
| Launch template gained a mandatory criteria block | Artefact | The doc cannot be marked ready for review with it blank |
| Product review opens with last month's criteria vs actual | Ritual | Every team, every month, including the wins |
| "Learning of the month" award for a disproven hypothesis | Reward | Named in the all-hands; the first recipient was a Director |
| Pre-registration became a one-click template in the experiment tool | Friction | The default path became the compliant path |
| VP presented a personal decision that was wrong, with the data | Modelling | Went first, before asking anyone else to |
| Staff PM promotion bar amended | Reward | "Has changed their own mind publicly with evidence" |
Result after two quarters:
| Measure | Before | After |
|---|---|---|
| Experiments pre-registered | 34% | 91% |
| Negative results published per quarter | 2 | 11 |
| Launches with pre-set success criteria | 48% | 86% |
| Launches meeting their criteria | Not measurable | 44% |
That last row is the point, and it is the one that alarms executives who have not thought it through. Before the change, every launch succeeded. After it, fewer than half did — which was always true and is now simply visible. A leadership team that treats 44% as a failure will destroy the norm in a quarter; one that treats it as the first honest number the company has ever had will compound on it for years.
The things that quietly kill product culture¶
| Killer | Mechanism | Counter |
|---|---|---|
| Promoting the loudest | Confidence rewarded over calibration | Promotion packets require evidence of changed minds and measured outcomes |
| Punishing the messenger once | One public reaction ends escalation for a year | Thank the bearer, visibly, especially when the news is expensive |
| Hero culture | Praise for weekend rescues | Reward the boring prevention; ask why the rescue was needed |
| Roadmap as contract | Dates become promises to sales | Commit to outcomes and near-term dates; publish confidence bands |
| Silent standards | The bar exists only in the VP's head | Write the matrix; calibrate in a room, together |
| Hiring for culture "fit" | Reproduces the current org, including its blind spots | Hire for values alignment and skill difference |
| Under-managing a strong performer | One tolerated exception teaches everyone the rules are optional | The exception costs more than the performance is worth; act early |
| Re-orging to avoid a conversation | Structure used as a substitute for feedback | Have the conversation; keep the structure |
The last one is common at exactly this level and nearly always visible to everyone except the person doing it.
Culture at scale: the onboarding multiplier¶
At 37.9% growth ListUp will add roughly 30 R&D people next year — about 28% of the org will have been there under a year. Norms do not transmit by osmosis at that rate.
| Mechanism | What new joiners get in week 1 | Why |
|---|---|---|
| The strategy one-pager | The same page everyone else has | Removes the folklore version |
| Three decision records | Real ones, including a reversed decision | Teaches the format and the tolerance for being wrong |
| Two published postmortems | Real incidents, real detail | Proves blamelessness is not a slogan |
| A customer conversation | Booked before day 10 | Sets the direct-contact norm before habits form |
| A named buddy outside their team | Not their manager | Cross-group context, and someone safe to ask |
An onboarding pack made of real artefacts teaches culture faster than any values deck, because it is evidence rather than assertion.
How It Actually Works: reinforcement schedules and the meeting-cost arithmetic¶
Culture change is an operant-conditioning problem, not a persuasion problem. Behaviour that is rewarded or visibly tolerated recurs; behaviour that produces no consequence extinguishes. This is why the levers are ranked "what gets rewarded > what gets tolerated > what you model > what you say": a reward changes the payoff of the behaviour every time it happens (a standing incentive), tolerance changes it only at the moment someone tests the boundary (an intermittent, high-salience signal), modelling and speech change only what people believe the payoff is, without touching the payoff itself. A values memo moves zero incentives, which is mechanically why it decays to nothing by week two — nobody's actual outcomes changed based on whether they read it. The reason "decline to walk past something once" outperforms a year of memos is that a single enforced consequence is a stronger, more concrete data point about the true reward function than a thousand words describing an aspirational one.
Why 44% "launches meeting criteria" is the correct number to have, not a crisis. Before the norm change, success criteria didn't exist for 52% of launches, so the implied success rate was 100% by definition — a measurement artifact, not a real number. Setting criteria in advance and measuring against them converts an unfalsifiable claim into a falsifiable one, and any genuinely falsifiable prediction process should produce a realistic hit rate well under 100% — a portfolio of experiments run in good faith, most of which explore genuinely uncertain territory, should fail more often than it succeeds, the same statistical logic behind why a low-power, well-calibrated A/B testing program shows plenty of null results (Level 2's significance-testing material). A leadership team that punishes the newly visible 56% failure rate is optimizing for the appearance of success, which reintroduces the exact incentive that produced 34% pre-registration in the first place — teams stop testing things that might fail rather than stop failing.
The meeting-cost math is a straightforward throughput/latency trade
computed as person-hours, and the reason it lands with executives is that
it converts a fuzzy "too many meetings" complaint into the same currency as
a headcount request. attendees × duration × frequency × fully_loaded_
hourly_rate gives the true cost of a standing meeting regardless of how
useful any individual instance feels — at 22 attendees this is
22 × 1 × 50 = 1,100 person-hours/year per meeting, or $107,250 at
$97.50/hour, before comparing it to any replacement. The written-update
comparison works because it separates the authoring cost (fixed, borne
once by whoever has the update) from the consumption cost (borne once
per reader, but at a fraction of the time — reading a position takes
minutes, generating and delivering one live takes an hour regardless of
audience size). Converting a synchronous one-to-many broadcast into an
asynchronous document is the same efficiency gain as batch processing over
polling: the cost scales with the number of authors rather than the
number of attendees × authors, which is why the saving (3,730 of 4,400
hours, 85%) tracks almost exactly with the ratio of meeting attendees to
document authors in the original setup.
Exercise¶
- Write down five norms you believe your org holds, and name the artefact that encodes each. Delete every one you cannot attach an artefact to — those are aspirations.
- Name which of the four cultures you are, which you need to be, and the single behaviour that would have to change first.
- Run the culture diagnostic: pick eight countable behaviours, measure the baseline honestly, and publish the results to your leadership team.
- Audit your recurring meetings. Count person-hours per year, cost them at a fully loaded hourly rate, and apply the one test: is there a decision to make? Convert the failures to documents.
- Identify the last thing you walked past — the launch with no criteria, the missed commitment, the tolerated behaviour — and decide whether you will walk past it again.
- Change one artefact this month, not one memo: a template field, a review agenda item, a default in a tool.
- Model the norm yourself first. Present a decision you got wrong, with the data, before you ask anyone else to.
- Amend one promotion criterion so that the behaviour you want is rewarded rather than merely praised, and take it through calibration.
- Build the week-1 onboarding pack from real artefacts — three decision records, two postmortems, the strategy page — and compute what share of your org will be under a year tenured twelve months from now.