Decide with confidence: process beats debate

• One-way doors (irreversible, high-consequence): slow down, gather data, build real consensus, decide at the top. Himalaya repositioning as an enterprise vendor or rebuilding its pricing architecture would take years to unwind - clients re-contract, the brand re-anchors, the sales team re-skills.
• Two-way doors (reversible): decide fast, delegate low, pilot small. An activation playbook trialed with 50 clients can be stopped on a Tuesday. The cost of deciding slowly EXCEEDS the cost of deciding wrong.
• The failure mode runs both ways: heavyweight process on two-way doors makes you slow; casual process on one-way doors makes you sorry. Sorting first is the whole trick.

Dave Snowden's Cynefin framework (HBR 2007) sorts problems into four domains, each demanding a different decision style:

The trap for smart teams: treating complex problems as complicated - analyzing harder when no amount of analysis can predict the outcome. "Will an AI copilot win the market?" is complex: no matrix settles it, only a probe does. That is exactly why parking it as a pilot-sized Horizon 2 bet (today's verdict) is the Cynefin-correct move.

Before deciding well, decide whether this deserves deciding NOW. Eisenhower's two-by-two - urgent/not urgent against important/not important - is the doorman for your calendar: important + urgent gets done, important + not urgent gets SCHEDULED (this is where strategy lives, and where it dies unscheduled), urgent + not important gets delegated, neither gets deleted.

The Himalaya read: the churn leak feels urgent AND important - it goes first. The AI copilot is important, not urgent - schedule it deliberately (Horizon 2) instead of letting it gatecrash every meeting. Notice the matrix quietly agrees with the doors: urgency pressure is exactly what tricks rooms into rushing one-way doors.

• Criteria are the strategy in disguise. "Impact on NRR" weighs 30 because Session 1's tree showed the stall is a retention leak. If your criteria could belong to any company, you have not done the diagnosis.
• The verdict, read from the table: the activation fix wins - the reversible two-way door that attacks the bleeding metric, starting NOW. The AI copilot is strong but premature - real NRR upside, weak on time and cost today, so it is parked as a Horizon 2 bet, not killed. The enterprise push scores weakest - a one-way door with the weakest capability fit - and is deferred.
• Nobody lost. Deven's copilot got funded as a bet with a horizon; Ilsa's push got a condition ("revisit when capability fit changes"), not a burial. A matrix verdict everyone watched being built is a verdict everyone can defend upstairs.

When outcomes are uncertain, weight them by probability: expected value = probability × payoff, summed across scenarios. A simple decision tree draws it: each choice branches into outcomes with probabilities and payoffs, and you fold the tree back to compare choices.

• Himalaya's copilot, sketched: say a $2M build with a 30% chance of a $10M ARR line and 70% chance of roughly nothing. EV = 0.3 × $10M - $2M = $1M positive, but concentrated and slow - which is exactly what "Horizon 2 bet" means in numbers.
• The activation fix: maybe $300k for a 70% chance of recovering ~$3M of churn-risk ARR. EV ≈ $1.8M, cheap, fast, reversible. The matrix said it qualitatively; EV says it in dollars.
• The honest caveat: the probabilities are guesses. EV's real gift is forcing you to WRITE DOWN the guess, so it can be argued and updated instead of hiding inside someone's conviction.

Half of all deadlocks are role confusion wearing a strategy costume: three people each believe they are the decider. RAPID (Bain) and DACI (Intuit) both fix this by naming roles before the debate - who Recommends (builds the case, runs the matrix), who must Agree, who gets Input, who Performs, and - the only letter that really matters - who Decides. One name, not a committee.

The Himalaya deadlock decoded: Deven, Ilsa, and Tom were all behaving as D. The unlock was Mara declaring "you three are R and I - I am D, and I will decide from the matrix in this meeting." Decide how to decide BEFORE deciding, and the second part gets shockingly fast.

• 1 · Imagine the failure as fact. "It is July next year. The activation fix failed - completely. This is history now, not speculation."
• 2 · Two minutes of solo writing. Everyone writes their version of why it died - silently, independently. The silence is the mechanism: it stops the loudest voice from anchoring the room.
• 3 · Round robin. One failure reason per person per pass, no debating, no defending, until the list is dry.
• 4 · Mitigate the top 3. Pick the three most plausible killers, assign one owner and one mitigation each. Then commit to the plan - now with its immune system installed.

Why it works: in a normal review, doubting the plan reads as disloyalty to whoever proposed it. In a pre-mortem, the failure is the premise - so the most junior person in the room can name the elephant, because the exercise ASKED them to.

Howard Marks's discipline from The Most Important Thing: first-order thinking asks "what happens?"; second-order thinking asks "and then what?" First-order conclusions are what everyone in the room already sees - the edge is always one consequence deeper.

• Enterprise push, first order: bigger logos, bigger contracts, growth story restored. Sounds great - that is why Ilsa believes it.
• Second order: enterprise clients arrive with enterprise demands - support load triples, security reviews stall the roadmap, and one big client's feature list starts steering product for everyone. And then what? The 1,150 mid-size clients who ARE the business feel neglect, and the churn you were fixing gets a new cause.
• The habit: for any option, write three "and then what?" chains before deciding. It is the pre-mortem's quieter sibling - same muscle, no meeting required.