Top-Down / Bottoms-Up Core Action Analysis
Cross two analyses — return-propensity ranking and product purpose — to find the one action worth optimising
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 4
- Confidence
- 93%
The concrete exercise Pinterest ran to identify its core action. The bottoms-up half ranks every action a user can take by reach and by propensity to return the following week. The top-down half asks what the product is fundamentally for and whether a user who never does the action can be said to understand it. The action that wins both halves becomes the north star, and every experiment is then graded against it.
Origin
Run by Sarah Tavel and the Pinterest team in 2011 when it was genuinely unclear whether Pinterest was a photo-graph social network like Instagram/Twitter or something else. It became the underpinning of the Level 1 layer in her Hierarchy of Engagement, and Lenny Rachitsky notes the same result usually surfaces as a company's north star metric.
Core principles
- 01Rank actions by two variables: what percentage of users complete them, and their propensity to bring the user back next week.
- 02Quantitative ranking alone is insufficient — a high-correlation action can still be a proxy rather than the point of the product.
- 03The top-down test: if a user never does this, do they understand what the product is for?
- 04The winning action becomes the experiment pass/fail criterion, not just a reported metric.
- 05Core actions are revisable — re-run the analysis as the product evolves.
How to run it
- 1
Enumerate every action in the product
List everything a user can do — Pinterest listed liking, following, clicking through, time on site, pinning and repinning. Do not pre-filter to the actions you hope will win.
Pro tip Include passive actions (time on site) so you can see them lose on merit.
- 2
Rank actions bottoms-up by reach and return propensity
For each action compute what percentage of users complete it, and, if a user does it in a given week, the probability they return the following week. Pinterest found pinning/repinning and clicking through carried return probability above 90%.
Pro tip Weekly is the right cadence for most consumer social products; pick the cadence that matches natural usage.
Watch out An action only a tiny fraction of users take cannot be the core action no matter how predictive — it will not scale to enough users.
- 3
Apply the top-down purpose test
Ask what the product is fundamentally for, and whether a user who never performs a candidate action really understands it. Clicking through predicted return well, but a user who never saves a pin to a board has not experienced what Pinterest is — so pinning won.
Pro tip Where the two analyses disagree, the top-down test is the tiebreak: the correlated-but-shallow action is usually a step on the path to the real one.
- 4
Convert the winner into an experiment gate
State it as a rule: if a launch does not raise the percentage of users completing the core action or the number of core actions per cohort, the experiment is not a success. Pinterest named it weekly active pinners.
Pro tip Also test whether the action serves both sides of your network if you have one — YouTube's subscribe does, which is a signal you have chosen well.
Watch out Revisit periodically. YouTube's core action moved from watching to subscribing as the platform matured; freezing the metric freezes the strategy.
In the wild
Twitter and Instagram were exploding on asymmetric follow graphs, so Pinterest genuinely did not know whether to optimise the follow graph. The bottoms-up ranking showed pinning and click-through drove >90% weekly return; the top-down question — does a user who never saves to a board understand Pinterest? — eliminated click-through.
→ Weekly active pinners became the north star and the experiment gate, resolving months of contradictory experiment data.
YouTube ran its own analysis and found subscribing, not watching, was the core action: subscribers are what make creators loyal to YouTube rather than another host, and subscriptions are why viewers return rather than going elsewhere.
→ The core action was changed mid-life, and it turned out to serve both supply (creators) and demand (viewers) — the ideal case.
Common mistakes
Running only the quantitative half
Return-propensity correlation alone would have made click-through Pinterest's core action. The top-down purpose test is what separates the action that IS the product from an action on the way to it.
Measuring everything and picking nothing
Pinterest measured follows, likes, click-throughs, pins and time on site simultaneously — experiments moved some up and some down and nobody could say whether they had worked. Clarity on one action is the point of the exercise.
Is it for you?
Best for
Product and data leaders at a consumer product with live usage data and no agreed north star metric
Not ideal for
Pre-launch products with no behavioural data — you cannot rank return propensity on an empty dataset
From the transcript
“we looked at every action that you could do on Pinterest so we had liking following clicking Through Time on site pinning repinning um and…”
“in a week what's your propensity to come back the following week and we basically ranked that”
“and then there's like the top down way of thinking about it which is well what is Pinterest for and if a user does come…”
From the episode
The hierarchy of engagement
Sarah Tavel (Benchmark, Greylock, Pinterest)