Close the Loop: Intent-Signal Retention Engine
Turn every low-commitment user signal into a trigger you can legitimately fire later.
- Difficulty
- Moderate
- Time to result
- ~months to results
- Steps
- 5
- Confidence
- 88%
Etsy's retention work started from a simple audit: users take lots of low-commitment actions (favoriting an item) that generate intent data the product does nothing with. Holley's team applied the habit loop — trigger, action, reward — by treating each stored intent signal as licence to send a specific, personally-relevant trigger later, aggregating them into an updates feed and delivering via push. The discipline is that the trigger must be a genuine change in the world, not a nag.
Origin
The habit loop (trigger → action → reward) originates with behavioral-science work popularized by Nir Eyal's 'Hooked' and B.J. Fogg's behavior model; Holley describes applying that existing framework to Etsy's favorites and cart data.
Core principles
- 01Every intent signal has a strength: favorite < add-to-cart < purchase. Store them all.
- 02The reward must be new information the user genuinely wants (price drop, low stock), not a reminder that they exist.
- 03Retention experiments run on 30/60/90-day horizons — a different measurement discipline than same-session conversion tests.
- 04Aggregate the loops into one surface (an updates feed) so notifications have somewhere to land.
- 05Push notification is the delivery mechanism, not the strategy. The strategy is having something worth pushing.
How to run it
- 1
Inventory your unused intent signals
List the low-commitment actions users already take that reveal intent — favorites, likes, saves, cart adds, repeat views — and note which ones the product currently does nothing with. That gap is your retention backlog.
Pro tip Rank them by signal strength. Cart-add and purchase are strongest; a favorite is 'strongish' but plentiful.
- 2
Define the reward that closes each loop
For each signal, define what real-world change would make returning worthwhile: the seller put it on sale, only one is left, it is selling out. Reward must be event-driven, not calendar-driven.
Pro tip The best notification is the one the user is pleased to receive — 'the thing that I really liked is now on sale.'
Watch out A trigger with no genuine change behind it is spam. It trains users to ignore the channel and burns the signal permanently.
- 3
Build the aggregation surface
Create a single feed of activity the user has taken, showing how each item has changed and what is new. This gives every closed loop a destination and makes the notifications coherent rather than scattered.
- 4
Wire the delivery channel
Use push notifications to make the change visible where the user already is. The user is on their phone constantly; a well-earned notification is a welcome one.
- 5
Adopt a retention measurement horizon
Shift the experiment discipline from 'did they purchase this visit or this week' to 'do they come back in 30, 60, 90 days.' Accept slower reads and design cohort analyses to see them.
Pro tip Keep asking the incrementality question — is this actually adding value to the business? — even when the read is slow.
Watch out A team trained on fast conversion A/B tests will be uncomfortable here. That discomfort is the point; do not let it push you back to short-horizon proxy metrics.
In the wild
A buyer favorites an item — a real but weak intent signal. Etsy stores it, watches for a change (seller puts it on sale, stock drops to one), surfaces it in the updates feed, and pushes a notification. The user gets a specific reason to return, not a generic re-engagement nudge.
→ A retention loop built entirely from data the product was already collecting and previously throwing away, and a newly credible push channel.
Etsy has a long-standing, strong A/B-testing culture optimized for fast conversion reads. Post-COVID, with a huge influx of new and reactivated buyers, the mandate became frequency and retention — which meant waiting 30, 60, 90 days for a read instead of days.
→ The team moved out of its comfort zone, expanded into cohort-over-time analysis, and kept the incrementality question intact rather than reverting to conversion proxies.
Common mistakes
Notifying without a real change
The loop only closes if the reward is genuine new information. Reminder-style pushes with no underlying event degrade the channel and the trust behind it.
Judging retention features on conversion-test timelines
Retention shows up at 30/60/90 days. Measuring it on a one-week A/B horizon will make every retention feature look like a failure and cause you to kill the ones that work.
Collecting intent signals you never act on
Favorites, saves, and wishlists are cheap to build and easy to leave inert. An unactioned intent signal is a retention loop you have paid to build and left open.
Is it for you?
Best for
Consumer product or marketplace teams sitting on rich but unused intent data (saves, favorites, wishlists) who need to move from conversion to frequency.
Not ideal for
Products with genuinely low natural repurchase frequency, where manufactured return triggers will read as noise regardless of relevance.
From the transcript
“how do we think about the habit loop? Of if you take an action, what's the trigger and then what's the reward?”
“that's where we've we've worked on things like we call it the updates feed. Essentially a feed of activity that you've taken that we're demonstrating…”
“you're you're looking at do they come back in 30 days, in 60 days, in 90 days? And so, that forced us out of our…”
From the episode
Inside Etsy’s product, growth, and marketplace evolution
Tim Holley (VP of Product)