Stated vs Revealed Preferences: Holding Conviction Through User Outrage
When users scream and keep refreshing, believe the behaviour — then debug details, not vision.
- Difficulty
- Advanced
- Time to result
- ~weeks to results
- Steps
- 5
- Confidence
- 90%
The News Feed launch produced universal outrage and an immediate doubling of usage. Bosworth's framework for product conviction is built on that cognitive dissonance: read revealed preferences against stated ones, and then run a separate diagnostic to determine whether the backlash means the thing itself is wrong or merely the details are wrong. The corollary is that you choose your customers as much as they choose you — an early cohort can drag you orthogonal to the larger market if you let their stated needs govern.
Origin
Andrew Bosworth's experience shipping the original Facebook News Feed in 2006 (with Chris Cox on engineering and Ruchi Sanghvi as PM) — the first algorithmically ranked feed of any social network, and among the first consumer AI ranking systems deployed on a website. The stated-versus-revealed preference distinction is borrowed from economics (Paul Samuelson's revealed preference theory).
Core principles
- 01You are choosing your customers as much as your customers are choosing you.
- 02Stated preferences and revealed preferences diverge — when they do, revealed behaviour is the stronger signal.
- 03'Don't listen to your customers' is the wrong lesson; the right lesson is to know which of their signals to weight.
- 04There is a hard distinction between the thing being wrong and the details being wrong, and telling them apart is an art.
- 05You need a prior: a clear vision and intuition for what you expected to happen, so you can diagnose the delta when something else happens.
- 06Early cohorts can hold you hostage and pull you away from the larger market.
How to run it
- 1
Hold a vision that isn't the thing you currently ship
Conviction requires a specific picture of what the product should become, distinct from what exists today. The people using today's product will not necessarily want that change — accept that in advance.
Watch out Conviction without a vision is just stubbornness. Bosworth is explicit that they were not always right — the discipline is in the diagnosis, not the certainty.
- 2
Instrument revealed preference before you launch
Make sure you can measure what people do, not just what they say. News Feed's outrage was universal — and so was the doubling of usage, which is what let the team keep their nerve.
Pro tip The tell is the refresh: users saying 'I hate this so much' while refreshing over and over is cognitive dissonance, not rejection.
Watch out Absent behavioural data, you have nothing to weigh the complaints against and you're back to opinion.
- 3
Run the wrong-thing vs wrong-details diagnostic
Ask explicitly: did we screw something small up, or something big? Is the concept wrong, or is the execution around the concept wrong? These call for opposite responses — kill versus fix.
Pro tip Compare what you expected to happen with what actually happened, and diagnose the delta. Without a prior expectation, you have no way to run this test.
Watch out This is where teams flinch: they read detail-level backlash as concept-level rejection and kill the right idea.
- 4
Fix the details fast and publicly
Acknowledge the genuine screw-ups. With News Feed the concept was right, but the team broadcast content people had shared in a technically public place at a volume they never intended — Bosworth's metaphor is the party where the music cuts out mid-sentence.
Pro tip Fixing detail-level mistakes quickly is what earns you the room to hold the line on the concept.
Watch out Refusing to admit any mistake because the concept is right is how conviction becomes arrogance.
- 5
Choose which customers you are for
If your early cohort's needs run orthogonal to the larger market, you must choose. Serving them completely means becoming their hostage.
In the wild
Facebook shipped the first algorithmically ranked feed and users were universally outraged — while immediately doubling their usage of the product, refreshing over and over. The team read the cognitive dissonance between stated and revealed preferences, judged the concept sound, and held. They also accepted a genuine detail-level error: wall posts that were technically public got broadcast at a volume users never intended.
→ News Feed became what people now mean by 'Facebook' — and, in Bosworth's framing, the News Feed/ads Venn diagram created roughly a trillion dollars of value.
Bosworth explains the News Feed detail-level mistake as being at a party talking over loud music when the music suddenly cuts out — you were speaking in a public place, but you never meant to broadcast at that volume. That is precisely what Facebook did to its entire user base with wall posts.
→ A clean articulation of the distinction that lets teams admit a real harm without abandoning a correct product bet.
Common mistakes
Concluding 'don't listen to customers'
Bosworth explicitly rejects this reading of the News Feed story. The lesson is about weighting revealed behaviour over stated complaint — not about ignoring users, whom he says they cared tremendously about.
Shipping without a prior
If you never wrote down what you expected to happen, you cannot diagnose the delta when something else happens — which means you cannot tell a broken concept from a broken detail, and you'll guess.
Becoming hostage to your first cohort
Early users' needs may point orthogonal to the larger market. Serving them faithfully feels like customer obsession and is actually how startups get trapped in a niche.
Is it for you?
Best for
Product leaders shipping a change their existing users are loudly against but that they believe serves the larger market
Not ideal for
Situations where the outrage is accompanied by falling usage, or where no behavioural instrumentation exists to test stated complaints against
From the transcript
“their earliest customers now uh so we've T time and time again we've had a vision for what we thought this should look like and…”
“there's cognitive dissonance here uh between what the stated preferences and what the revealed preferences are in the economic sense”
“people sometimes misunderstand that they think oh the lesson is don't listen to your customers not about and we certainly care tremendous about”
“when is the thing itself wrong versus when were the details wrong that is an art that is a real art and you don't always…”
“happened instead and can you diagnose that Delta there”
From the episode
Making Meta
Andrew ‘Boz’ Bosworth (CTO)