Verve-PM
← All Resources
Strategy names the goal. It does not tell you how to reach it. Closing that gap is the job: a PM and their team place a small number of strategic bets on where the value actually is, and fund the initiatives that chase each bet. Everything below the bet exists to find out, quickly and cheaply, whether the bet was right.
Take FreshCart, the weekly grocery delivery app used as the running example on this page. The strategic goal is to turn one-time buyers into weekly customers, and today 68% of new customers order once and never return. The team's bet is that retention is won at the second order. That bet funds three initiatives, starting with One-Tap Repeat Order, and that initiative rests on one assumption risky enough to be worth testing before the year is spent on it.
Five words, five altitudes, one chain from the goal down to the evidence and back. Most teams use them interchangeably. Here is what each one actually is.
A team, or a single PM, holds several live bets at once, and every step down the chain fans out again. The relationship between each two levels is one to many, never one to one. That is the whole point of the structure: one wager buys you several ways to be right about it, and one piece of work buys you several ways to find out you were wrong.
Read the other direction and it is strict: every initiative belongs to exactly one bet, every hypothesis to one initiative, every experiment to one hypothesis. Anything that cannot name its single parent is orphan work.
The goal at the top, the evidence at the bottom, and one arrow that runs back up. Only the highlighted bet is opened here. The other two fan out exactly the same way, at the same time.
Scroll the diagram sideways to follow the whole chain.
The whole model on one screen. The "how many" column is where the one to many relationship shows up in practice.
| Level | Answers | How many | Horizon | Owner | Written proof it is real | Lives in |
|---|---|---|---|---|---|---|
| Bet | Where do we think value is? | 3 to 5 live at once | 2 to 4 quarters | Product leadership | It spawns several initiatives and names what it costs you | S-Strategy/Strategic-Bets.md |
| Initiative | What are we building? | 2 to 4 per bet | 4 to 12 weeks | The team | It has an end state and a named parent bet | I-Initiatives/ |
| Hypothesis | What could be wrong? | 1 primary per initiative | 2 to 6 weeks | The PM | There is a number that would kill it | H-Hypotheses/ |
| Experiment | How do we find out? | 1 or 2 per hypothesis | 1 to 4 weeks | PM and data | Criteria and guardrails were written before the ship | P-Proof/ |
| Learning | What changed our mind? | 1 or more per experiment | Permanent | Everyone | Something downstream moved because of it | Learning.md |
Scroll the table sideways for the remaining columns.
One product, one situation, followed from strategy all the way down to a number and back up again. Every card below ends with this example at that altitude.
Each card gives the definition, the test that separates it from the level below it, and the FreshCart instance.
A directional wager on where value lives, made while you are still uncertain, and funded for quarters rather than sprints. A bet is a claim about the market and the customer, not about a feature. It says what you are pointing the company at, and by implication what you are giving up.
Does it spawn more than one initiative?
If a single team can finish it, it is an initiative wearing a bet's clothes. A real bet is bigger than any one thing you would build for it, and it survives the failure of its first initiative.
Can you name what it costs you?
A bet you would make anyway is not a bet. It should be visibly expensive in the things it does not fund.
"Retention is won at the second order, not the first. We are betting the year on making the weekly re-order effortless."
That wager explicitly deprioritizes the two things the team wanted to do instead: expanding the catalog and launching in a third city.
A committed body of work that serves a bet. It has a scope, a team, a start and an end. An initiative is a decision to spend, so it is written in the language of what will exist when it is done, not in the language of what you hope will happen.
Can you tell when it is finished?
An initiative ends. If there is no state of the world in which you say "that shipped", you have written a theme, not an initiative.
Does it point up at a named bet?
An initiative that serves no bet is orphan work. It may still be worth doing, but nobody can tell you what it is worth.
The bet spawns three:
INI-014 One-Tap Repeat Order, last week's basket on the home screen.
INI-015 Smart Basket, predicts what has run out.
INI-016 Slot Subscription, a standing delivery window.
The riskiest belief inside an initiative, written so that reality can prove it wrong. Every initiative rests on a stack of assumptions. The hypothesis is the one that, if false, makes the whole initiative pointless. It names a population, a metric, a threshold and a window.
Can you write the number that kills it?
If no result would make you stop, you wrote an opinion. "Users will love it" cannot be wrong, so it is worthless as a hypothesis.
Is it the riskiest one?
Teams reliably test the assumption they are most confident about, because that feels safer. Rank the assumptions and take the top one.
HYP-021 "If new customers see last week's basket as a one-tap re-order on the home screen, the 14-day second-order rate rises from 31% to at least 38%."
Population: new customers, first order placed. Metric: second order within 14 days. Baseline: 31%. Bar: 38%. Window: 14 days.
The cheapest procedure that can produce the number that kills the hypothesis. An experiment is a method, not an outcome: who is in it, what they see, how long it runs, what you measure, and what you refuse to break while measuring. All of it written down before anything ships.
Were the success criteria written before the ship?
Criteria written afterwards always get met. A launch with no pre-written bar is a release, not an experiment.
Is it the cheapest way to be wrong?
Interviews, fake doors and prototypes are experiments too. Build the product only when nothing cheaper can answer the question.
EXP-007 A/B test, 50/50 on new customers, 3 weeks of intake, each cohort read at its own day 14. Minimum 4,000 users per arm.
Primary metric: 14-day second-order rate. Guardrail: average basket value must not fall more than 3%, because a one-tap repeat could quietly shrink the cart.
What the result changed about what you believe. Not the number, the consequence of the number. A learning is written as a claim about customers that outlives the initiative that produced it, and it is the only artifact in the chain that flows upward.
Did something downstream change?
If no bet was re-priced, no initiative reordered and no roadmap touched, you wrote a status report. Numbers you did not act on are not learnings.
Does it survive the feature?
"The button worked" dies with the button. "People treat the last basket as a starting point" is still true in three years.
Result: second-order rate moved 31% to 36.4%. Under the 38% bar, above the kill line, guardrail held at minus 1.2%.
The surprise sat in the instrumentation: 62% of one-tap users edited the basket before checkout, and they added a median of 2.3 items.
Almost every misuse is the same mistake: writing one level in the language of the level above or below it.
Four of the five links flow downward, from strategy into work. Learning is the only one that flows up, and it is the reason the system compounds instead of just recording.
The learning did not kill the bet, it sharpened it. The second order is still where retention is won, but the mechanism was wrong: the win is a good starting cart, not a repeated one. That re-prices the three initiatives without a new planning cycle.
One-tap re-order shipped anyway, at 36.4% it paid for itself. But it stopped being the year's main move, and that decision took one meeting because the chain was written down.
AI-SHIPR Workshop by Yaniv Yaakubovich