Decide

Run a real decision process, not a debate: find the question that matters, set the rigor budget, collect independent input, stress-test with a pre-mortem, name the decider, and log the bet so you can learn from it. Built from 396 insights from 177 Lenny's Podcast guests. Lead team: Annie Duke, Shreyas Doshi, Shishir Mehrotra.

When to use

• The team is split on a roadmap, strategy or launch call and meetings loop without a decision.
• You are about to approve a feature, rewrite, vendor, hire or deal and need to know how much rigor it deserves.
• A project is about to launch and you want a pre-mortem that changes behavior, not a ritual.
• You suspect you should quit, pivot or kill something but sunk cost is in the way.
• You face a personal or career fork (offer, move, leaving) with no clean data.
• Nobody knows who decides, or you own a call but lack authority.
• You have a decision doc, PRD recommendation or memo and want it graded before it goes up.

Step 1 — Diagnose (ask before answering)

If the user attached a doc, deck, thread or PRD, read it first and infer answers. Ask only what is missing (max 4):

1. What exactly is being decided, by when, and who is the decider? Why: an unnamed decider routes you to the decision-rights play before anything else.
2. What does it cost to be wrong, and can you undo it? (money, users affected, months, reputation; and your level: below CEO, apparent two-way doors are often one-way.) Why: sets the rigor budget (Play B) and decides whether to skip straight to deciding.
3. Is this a group decision, a stuck project, or a personal call? Why: group means Play C or D, stuck project means Play E, personal means Play F.
4. What evidence exists: analogs and data, or a new category where gut leads? Is this a one-off or a repeating call type? Why: routes the gut-versus-data debate and tells you whether to start a decision log.

Step 2 — Pick the play

Most real decisions chain 3-4 plays: B first, then A or H, then C, then D, then G.

Step 3 — Run it

Play A — Eigenquestion (Shishir Mehrotra)

1. List every open question (10-20). Do not sort by importance or order of arrival.
2. For each, ask how many other questions would be answered if this one were. Rank by that count.
3. Test the top candidate with the follow-up what decision will answering this let me make?
4. Lift it to the market-level theory underneath the fight. At YouTube, the linking-out debate became the question of whether online video values consistency or comprehensiveness; one answer settled linking out, embedding other players and taking back the iPhone app.
5. Debate that single question as a theory about the future, then derive the downstream decisions.
6. Pair with Richard Rumelt's diagnosis test: a stated desire (we want to grow faster) is not a diagnosis; keep asking because... until a cause is named, then deliberately look for a second reading before committing. Done when: one question is written down, the team has picked an answer, and at least three downstream decisions fall out of it. Hard under pressure; practice on low-stakes cases first (Mehrotra).

Play B — Triage and rigor budget (Brandon Chu, Dharmesh Shah, Shreyas Doshi)

1. Rate importance: is it reversible, and does it materially affect many users? (Chu). Few decisions are critical; decide the rest by gut or delegation.
2. Set the budget by cost of being wrong (Shah): Low = decide now, gut or delegate; Medium = one-page memo plus one outside check; High = full memo, pre-mortem, and a pause of two days to two weeks.
3. Apply Doshi's PM-level correction: before approving a two-way door feature, answer customer motivation, differentiation and distribution, and pause two minutes, two days or two weeks. Thinking is cheap; the commitments a shipped feature creates are not (the QBR feature trap).
4. Check what is actually irreversible: Eeke de Milliano says pricing can be grandfathered, but titles cannot be undone; Nickey Skarstad flags standards every seller must follow as one-way.
5. Information threshold: do not decide on under 30% of the data and do not wait for 70% (Shaun Clowes); if the upside is orders of magnitude larger, skip precision (Alex Komoroske).
6. Low-stakes filter: will anyone remember this in six months? (Will Larson). If no, pick something reasonable and move. Done when: the memo header says Low/Medium/High plus the time budget.

Play C — Independent input, discuss, then decide (Annie Duke)

1. Name the decider (one name), the criteria, who is informed, and when you will revisit (Claire Hughes Johnson, Hari Srinivasan). If unsure who decides, it is probably you.
2. Decide beforehand what opinions you need: a ranking, an estimate, a brainstorm.
3. Collect independently and asynchronously, no reply-all: a forced ranking with a three to five sentence rationale. For timelines, ask for a point estimate, a lower bound and an upper bound with rationale (nominal group forecast).
4. Compile and circulate before the meeting. Add a sentiment or pulse table (Lane Shackleton) to surface quiet dissent; group it by decision role and pre-fill what you need from each approver (Mehrotra).
5. Use the meeting only to discuss disagreements. The facilitator reflects back without offering an opinion. Edit the slide live to record the choice and numbered next steps (Naomi Gleit).
6. Decide outside the room: the decider or a private vote. Never ask are we all aligned?; say nevertheless, this is the decision (Duke).
7. Before leaving, ask each person to state what was decided (Alisa Cohn); differing answers mean you are not done.
8. Dissenters write their disagreement, then commit by understanding the reasoning (Bill Carr) or champion it (Sanchan Saxena). Done when: every attendee gives the same one-sentence decision.

Play D — Pre-mortem with kill criteria (Shreyas Doshi + Annie Duke)

1. Invite every function; for large launches run separate product-engineering and go-to-market sessions.
2. Prompt: imagine this has failed miserably six months from now. Five to ten minutes of silent writing: tigers (will actually kill us), paper tigers (others worry, you do not), elephants (the unspoken issue).
3. Go around the room. Each person votes for the scariest tiger someone else raised.
4. Duke's addition, the part that changes behavior: for each top tiger, write the early signal you would see, the threshold that makes it a kill criterion, and a pre-committed action (kill, pursue or probe) with an owner and review date.
5. Leader publishes the action plan and stays accountable. Reuse the vocabulary in later meetings.
6. For multi-quarter bets add quarter-long go/no-go milestones (Jiaona Zhang). Bing spent 100 person-years on social-feed integration after hundreds of flat experiments (Ronny Kohavi): decide the abort point before you start. Done when: at least three tigers each have signal, threshold, action, owner, date.

Play E — Quit, pivot or persist (Duke, Belsky, Caldwell, Ries)

Run in order and stop at the first clear answer:

1. Knowing what I know today, would I start this now? Count only future costs (Duke, Uri Levine). Set a 90-day clock or it never happens.
2. Do I have more or less conviction in the problem and the solution than at the start? More: you are in the messy middle; less: quit or pivot. Never decide on a bad day (Scott Belsky).
3. Count untried, credible growth ideas: six or more, try them first; out of ideas, pivot (Dalton Caldwell).
4. Belief or hope? Hope is confidence without basis (Mark Pincus). Asking whether it is an A means it is a B+.
5. Did the core assumption break? Codeium pivoted while profitable when the architecture assumption died (Varun Mohan). Do not run old and new side by side.
6. Still unsure: time-box a fixed period (weekend to six weeks) of all-in action on one thing; everyone states what they wish they were working on (Eric Ries).
7. Cash crisis: compute runway, pick a target, decide today (Uri Levine); wind down early and return capital beats burning every cent (Tom Conrad).

Play F — Personal decision stack (Ada Chen Rekhi, Paul Millerd, Graham Weaver)

1. Write values from a word list, group, stack-rank into three to five sentences: your inner scorecard.
2. Run the movie forward: extend your current path and score it against those values; score the new option the same way.
3. Curiosity Loop: send a specific, unbiased question to 4+ people, mixing a subject expert and someone who knows you; look for surprises and disagreement, then thank them and close the loop. Advice is input, not instruction.
4. Fear setting: fears, mitigations, and the cost of inaction (Millerd). Ask what your five-years-later self wants (Weaver); expect worse first.
5. If fear is driving, check with someone not in fear (Matt Mochary). Financial fear: do the monthly number; capability fear: a green light to prove it (Molly Graham).

Play G — Decision log and hygiene (Kevin Yien, Annie Duke)

1. At decision time log: decision, rejected options, rationale, forecast with point plus low and high bounds, conviction (low/medium/high), and intermediate signals.
2. For long-horizon bets, pick signals that are necessary or correlated (Series A funding, retention, net new ARR) and forecast them now; Duke: there is no such thing as a long feedback loop.
3. Tag #decision in a running log and set a review reminder (six months out, or weeks for small bets). Start tiny: one bet per week.
4. At review, grade the process and the outcome separately. Do not result (Duke).
5. For repeat evaluations with several judges, use a 1-7 rubric of components plus explicit forecasts, record it, and compare to outcomes before adjusting weights.

Play H — Options, recommendation, stress test (Naomi Gleit, Rahul Vohra, Ebi Atawodi)

1. Generate at least three real options, one you dislike. There are unlimited wrong options and about 10,000 right ones (Farhan Thawar via Toby Lütke); the biggest risk is never seeing the right option (Nan Yu).
2. For each, find the single decisive reason. If an option lists several weak reasons, ask whether any one alone would justify it (Vohra).
3. Build a traffic-light table: options as rows, criteria or functions as columns, each cell red, yellow or green with rationale; rule out the reddest. If nothing wins on every lens, there is no perfect choice; pick the lens that matters most (Jake Knapp).
4. Stress test: what do we do if the result is green? if red? (Tom Verrilli). If you could build only one of five, which? (Ebi Atawodi). Is it better for the customer too, not only for us? (Jason Cohen). How does this make users love the product more? (Doshi, from Jack Dorsey).
5. The PM still recommends one and champions it (Anuj Rathi); do not escalate an A-versus-B pros and cons list. Done when: three options, one recommendation, one decisive reason, one decider.

Where the experts disagree

1. Decide fast on two-way doors (Gibson Biddle, Nickey Skarstad, Mayur Kamat, Claire Vo: fast beats right) vs think more (Shreyas Doshi on two-way door misuse; Austin Hay: delay as long as you can; Dharmesh Shah: effort proportional to consequences). Use fast when it is cheap, reversible and you own it end to end; slow when you are below CEO level and the call creates commitments. Default: Play B budget, then pause two days on anything that commits other people.
2. Single decider, independent input (Annie Duke, Kevin Aluwi, Sanchan Saxena, Dharmesh Shah: debate, decide, unite) vs decide together (Tobi Lütke, Qasar Younis: many inputs, anyone can change the leader's mind, but not a democracy). Duke's model wins when politics or seniority distort the room; Lütke's wins in small expert teams with high trust. Default: collect independently, discuss in the room, one named decider.
3. Gut is data (Jason Fried, Tony Fadell, Maya Prohovnik, Sanchan Saxena) vs bring evidence (Itamar Gilad, Ronny Kohavi, Mayur Kamat). Gut leads in a new category with no analogs and an informed taste-maker; evidence leads when analogs exist or you must persuade others. Default: log your gut as a forecast with a conviction level (Brian Tolkin); explain it objectively (Prohovnik) and make intuition explicit so you learn when it is wrong (Duke).
4. Emotion is signal (Joe Hudson, Jonny Miller) vs strip it out (Qasar Younis; Matt Mochary and Jonathan Lowenhar on fear). Name the emotion as information; do not let it vote. Default: check with someone not in fear, and ask whether different people would reach the same decision (Younis).

Deliverable

# Decision memo: <one-line decision>
**Decider:** <one name> | **Decide by:** <date> | **Informed:** <names> | **Revisit:** <date>
**Rigor budget:** Low / Medium / High (cost of being wrong: <$, users, months>; reversible? <how>)

## Eigenquestion
<the question whose answer settles the most others> -> Our answer: <...>

## Options (3+)
| Option | Single decisive reason | Red/yellow/green by criterion | Biggest risk |
Recommendation: <option> because <one sentence>. Opportunity cost: <what we give up>.

## Facts vs hypotheses
Facts: <...> | Hypotheses and how we test them: <...> | Conviction: low / medium / high

## Independent input (collected before the meeting)
<ranking and rationale per person; where they disagree>

## Pre-mortem
| Tiger / elephant | Early signal | Kill criterion | Pre-committed action (kill / pursue / probe) | Owner |

## Forecasts
<outcome>: point <x>, range <low-high>; intermediate signals: <...>; review date: <...>

## Dissent and commitment
<who disagreed, their strongest argument, how they commit or champion>

## Next 3 actions
1. <owner, date>
2. <owner, date>
3. <owner, date>

Grade existing work

Score each 1-5 (1 = absent, 3 = partial, 5 = strong). Output: score per criterion, total out of 35, and the top 3 fixes with the guest behind each.

Red flags

• Resulting: judging a decision by its outcome, praising or blaming luck (Annie Duke).
• Pre-mortem with no pre-commitment: it rarely changes behavior alone (Duke).
• Loudest-voice estimation and are-we-aligned closings (Duke).
• Two-way door misuse: approving features fast because they look reversible (Doshi).
• Pros and cons memos that dodge a pick (Ebi Atawodi); a tidy flat list hides the trade-off (Gleit).
• Hedging a big bet: Shopify's split platform bet cost about 18 months and 100 engineers (Farhan Thawar); unclear direction beats being wrong (Tomer Cohen).
• A statement of desire passed off as a diagnosis (Rumelt).
• Confidence read as quality, certainty language, hope as evidence (Bret Taylor, Marc Benioff, Mark Pincus).
• Single-issue voter bias: your fix matches your skillset, at least 30% likely chosen for comfort (Bret Taylor).
• False precision: paying to pin down numbers where the order of magnitude decides (Komoroske).
• Quitting late: by the time you think of it you are probably past it (Duke).
• Never seeing the right option (Nan Yu).

Receipts

• "the only thing that's ever supposed to happen in a meeting is the discussion part." — Annie Duke, Lenny's Podcast (00:27:26)
• "a pre-mortem is great, but only if you attach a pre-commitment with it." — Annie Duke, Lenny's Podcast (01:05:25)
• "The eigenquestion, the simplest definition of eigenquestion, it's the question that when answered also answers the most subsequent questions." — Shishir Mehrotra, Lenny's Podcast (00:52:51)
• "most doors that look like two-way doors are actually one-way doors. They are two-way doors at Bezos' level, but as a PM leader, for you, they are a one-way door, and that's making you busy." — Shreyas Doshi, Lenny's Podcast (23:19)
• "the calories you spend on a decision should be proportional to the consequences of that decision, pure and simple." — Dharmesh Shah, Lenny's Podcast (01:07:21)
• "for important decisions, you should be able to identify one, one reason that on its own supports the decision." — Rahul Vohra, Lenny's Podcast (01:19:22)

Go deeper

• references/frameworks.md — every framework in this skill, how to run it
• references/quotes.md — verified quotes with timestamps
• Related skills:
  • strategy — when the decision is a where-to-play choice that needs a diagnosis and moat test.
  • prioritize — when you are ranking many items rather than making one call.
  • experiment — when the decision can be tested cheaply before you commit.
  • exec-comms — to write the decision up for leadership once it is made.