Eigen Questions
Rank open questions by how many other questions they kill, then answer that one first.
- Difficulty
- Advanced
- Time to result
- ~weeks to results
- Steps
- 6
- Confidence
- 95%
An eigen question is the question that, when answered, also answers the most subsequent questions. Instead of ranking a decision backlog by urgency or importance, you rank it by leverage — which single answer collapses the rest of the list. Shishir Mehrotra used this at YouTube to break a company-splitting deadlock, and it became a named ritual at both YouTube and Coda.
Origin
Coined by Shishir Mehrotra at YouTube in 2008, named after eigenvectors from linear algebra (the most discriminating vectors in a multi-dimensional space). The trigger was the 'Modern Family question' — whether YouTube should link searchers out to abc.com — plus a Google Shopping user-research insight ('I value consistency over comprehensiveness') that suggested the reframe. Mehrotra credits Brian Chesky with an adjacent habit: sorting a to-do list by which item is most likely to create leverage and eliminate the rest of the list.
Core principles
- 01Most teams rank decisions by impact or by the order the questions arrived; both are the wrong axis.
- 02The right axis is leverage: which answer eliminates the most other open questions.
- 03Object-level debates (link out or not) often stalemate as good-vs-evil; the eigen question sits one level above and is debatable on the merits.
- 04Once answered, an eigen question should cascade into many downstream decisions becoming 'easy'.
- 05Eigen-question spotting is a learnable skill — but only if you practise it outside high-stakes situations.
- 06Kids are naturally good at it; the skill is more un-learned than absent.
How to run it
- 1
Write out the full list of open questions
Dump every decision the team is stuck on or arguing about — the link-out question, the app-ownership question, the launch-scope question. Don't filter yet.
Pro tip Recurring meeting-after-meeting arguments are the best raw material; a question that keeps coming back and never resolves is a symptom of a missing eigen question above it.
- 2
Re-rank the list by elimination power, not importance
For each question ask: if I answered this, how many of the others would answer themselves? Sort descending. The top of that list is your eigen question.
Watch out Do not sort by 'most impactful decision'. The biggest decision is usually a downstream consequence, not the lever.
- 3
Reframe upward until the question is debatable on evidence
If the surface fight is a values clash, climb one level to a question about how the market or user will behave. At YouTube the object question 'do we link out to abc.com?' became 'in a decade, is online video more about consistency or comprehensiveness?' — a question two opposed camps could actually reason about together.
Pro tip Look for a distinction with a name (consistency vs comprehensiveness; doc vs app). A named axis is portable across every downstream decision.
Watch out A reframe that no one could argue either side of is not an eigen question, it's a rhetorical device.
- 4
Answer it once, in one forum, with data and framing prepared
Run a dedicated session (YouTube used a full offsite day). One person is assigned to frame the discussion, collect everyone's opinions, and ground it in fact before the debate opens.
- 5
Cascade the answer through the backlog
Walk the remaining questions and apply the answer mechanically. Expect most of them to resolve fast, and expect new decisions you hadn't listed to fall out too.
Pro tip Cement the answer so you stop re-litigating it. Mehrotra named the company 'Coda' — 'a doc backwards' — precisely so the doc-vs-app eigen question could never be reopened.
- 6
Practise on frivolous, low-stakes problems
Build the muscle where nothing is riding on it — a made-up product, a question your kid asks, a company you don't work at. Force a two-question budget: if you could only ask two questions before committing to a plan, which two?
Pro tip The two-question constraint is the whole exercise. Unlimited questions produce a research list; two questions force eigen questions.
Watch out Practising only inside real high-stakes decisions is like only ever playing basketball in real games and only ever playing piano in recitals — you never actually get better.
In the wild
'Modern Family' was a top-five weekly YouTube search query and YouTube didn't have the show; ABC had put every episode on abc.com. The company split — product and engineering wanted to link out ('do right by the user'), while sales, marketing and content partnerships said linking out would guarantee nobody ever put good content on YouTube. Meeting after meeting stalled. The night before a full-day offsite, Mehrotra reframed the agenda: don't debate linking out at all; debate whether the online video market, a decade out, would reward consistency or comprehensiveness.
→ The team concluded the market would value consistency over comprehensiveness. The link-out question then answered itself (no), and so did a stack of others — they stopped embedding third-party Flash players, and they took the YouTube iPhone app back from Apple despite losing default distribution and starting from zero downloads. Mehrotra judges the call correct ~15 years on: no video property is comprehensive.
Mehrotra hands a candidate a hypothetical: scientists have invented a teleportation device and hired you to bring it to market. Candidates generate a pile of questions — how fast, how big, does it disintegrate things, does it need a sender and receiver. He then adds the constraint: the scientists dislike people and will answer exactly two questions, after which they expect a plan. Which two?
→ Strong candidates converge fast. One best answer: (1) is it safe enough for humans? (2) is it more expensive to buy them or to run them? That 2x2 generates the whole strategy — cheap-to-buy/expensive-to-run means deploy everywhere like human fax machines; expensive-to-buy/cheap-to-run means place them strategically, replacing airports; not-safe-for-humans means a freight/goods business entirely. Weak candidates ask about size and cannot say what decision the answer would drive.
Common mistakes
Ranking the decision list by importance
The most consequential-sounding question is usually downstream of the lever. Sorting by impact keeps you fighting the biggest fight instead of the fight that dissolves the others.
Asking questions whose answer changes nothing
Mehrotra's tell for a weak candidate is a question like 'how big is the device?' — followed by no answer to 'what decision would that let you make?' Every question you spend budget on must map to a fork in the plan.
Only practising on real, high-stakes decisions
Eigen-question spotting is a skill and skills need low-pressure reps. Teams that only try it in board-level moments never build the reflex and default back to arguing the object level.
Is it for you?
Best for
Product leaders and founders whose exec team keeps re-arguing the same decision meeting after meeting, where the debate has hardened into two values-based camps.
Not ideal for
Genuinely independent decisions with no shared upstream assumption, or reversible low-cost calls where just shipping and measuring is cheaper than reframing.
From the transcript
“the most subsequent questions and it's a very simple idea”
“rank them by which ones would eliminate the most other questions on the list”
“if you answer just one question do we value consistency over comprehensiveness”
“i think one of the troubles with the concept like eigen questions is we tend to only practice it in real world scenarios that are…”
“we value consistency over comprehensiveness we would much rather be on fewer phones or the more consistent experience than beyond all of them with an…”
From the episode
The rituals of great teams
Shishir Mehrotra of Coda, YouTube, Microsoft