Recall vs. Discovery Interface Split
Match interface density to user intent: dense grids for recall, low-density sound-rich feeds for discovery.
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 4
- Confidence
- 92%
Two user modes need opposite interfaces. 'Recall' (I know what I want) demands a dense UI packed with many items so users can pounce on the known target. 'Discovery' (I don't know what I want) demands a sparse, high-real-estate, often sound-on feed tolerant of a low hit rate. Using one interface for the other job breaks the experience.
Origin
Söderström's diagnosis of Spotify's 2023 home-feed redesign backlash, cross-checked against quantitative traffic data.
Core principles
- 01High-intent recall needs a dense, predictable, many-items-per-screen layout
- 02Low-intent discovery needs a low hit-rate-tolerant feed with lots of pixels per item and often audio
- 03A discovery UI used for recall behaves like an unpredictable slot machine
- 04Recall demands ~9-in-10 accuracy; discovery succeeds at ~1-in-10
How to run it
- 1
Classify the job as recall or discovery
Determine whether the user arrives knowing the specific thing they want (recall) or genuinely open with no signal (discovery). These are fundamentally different problems even though both look like 'recommendation.'
Pro tip For a completely new genre you have zero signal by definition — 'if you had no idea, it wasn't new to them' — so plan for a very low hit rate.
Watch out Do not treat breaking someone out of their taste bubble as just another recommendation problem; the required hit rate is an order of magnitude lower.
- 2
Set the accuracy bar per mode
For recall, aim for at least nine correct in ten — a few duds and users think you broke their session. For discovery, one gem in ten tries can be a win, provided the user expects that ratio.
Watch out A background recall stream with five duds out of ten reads as a broken playlist.
- 3
Build a dense UI for recall
When users know what they're looking for, pack many items on screen; they need little real estate per item because they're scanning for a specific known target.
Watch out A sparse single-item feed forces users into a slot-machine hunt for a playlist they already know they want.
- 4
Build a spacious, sound-on feed for discovery
For open-ended exploration, give each candidate lots of pixels and let it play audio, because the user cannot judge an unknown item from a small label; make rejection a cheap swipe so users can churn through many low-hit candidates fast.
Pro tip Swipe-through feeds work because the per-item cost of a miss is near zero and users can burn through candidates quickly.
In the wild
Spotify's home was ~90% recall / 10% discovery. Responding to user requests for more discovery, they flipped it to ~10% recall / 90% discovery. Users flooded Twitter — not really about the discovery, but because they could no longer find their known playlists; traffic visibly shifted from home into search and library.
→ Spotify reverted toward keeping strong recall while making discovery tools available but voluntary — treating recall as a thing they did better than competitors, not to be thrown out.
Common mistakes
Using a discovery feed for recall tasks
A low-density, unpredictable feed is terrible for retrieving a known item — it becomes a slot machine where users may never reach the workout playlist they already know they want.
Assuming users want maximum discovery because they asked for more
Users requested better discovery, but flipping to 90% discovery removed the recall they relied on; wants must be weighed against existing behavior, not maximized literally.
Is it for you?
Best for
Consumer product teams redesigning a home/feed surface that serves both known-intent retrieval and open-ended exploration.
Not ideal for
Single-purpose apps where every visit is the same intent, so one interface mode suffices.
From the transcript
“when you do recall you want the dense UI with many items on screen because you know what it is you're looking for”
“our hit ratio needs to be at least nine out of 10”
“If you find one gem out of 10 tries, you're very happy.”
“you want a lot a lot you want a lot of user interface a lot of pixels and you probably want sound because you don't…”
From the episode
Lessons from scaling Spotify: The science of product, taking risky bets, and how AI is already impacting the future of music
Gustav Söderström (Co-President, CPO, and CTO at Spotify)