All articles Fundamentals

Marketplace weekly retention cohort: a setup guide

Analyst comparing two glowing retention grids in a dark studio office

A marketplace weekly retention cohort groups users by the week they first got real value, then shows what share of each group came back in every week that followed. It is the view that matches how a two-sided market behaves, because it keeps demand and supply in separate rows instead of averaging them into one number that moves with mix.

The hard part is not the chart. It is deciding what week 0 means, which event proves value, and how to keep buyers and sellers apart. Get those three wrong and the grid still renders — it just stops telling you anything.

The setup below follows the order you would implement it.

Define the first-value event before you write the cohort query

Signup is not value. A buyer who created an account and left has not experienced the marketplace; a seller who published a listing nobody ordered has not either. Start the cohort where the market clears for the first time.

Buyer and seller first value are different events

  • Buyer: the first order marked fulfilled or delivered on your side. Not “checkout started”, and not “order placed” if fulfilment routinely fails.
  • Seller: the first order received on one of their listings, or the first payout released — whichever your sellers treat as proof the channel works. Pick one and keep it.

Make the event deterministic

Emit one server-side event per side at the moment the state change is final, carrying the user id, the marketplace-side account id, the order id and the side. Server-side matters: a client pixel can fire twice, fire on a retry, or never fire behind an ad blocker, and each of those quietly reshapes a cohort.

Split buyers from sellers at cohort entry

Buyer holding a first delivered parcel in a dark hallway

One person can hold both roles

A seller who also buys supplies, or a buyer who lists an item next month, is normal. The rule: assign the role that was active in week 0 and let every event carry its own side. Do not move people between cohorts later, or you rewrite history each time someone switches. If a user really acts on both sides in the same week, they can appear in both cohorts — as long as events are tagged and the two grids are never summed.

Give the grouping key a permanent home

Pass the marketplace-side id as the account id when you identify the user, so sessions group per shop or per buyer organisation and open inside your own admin; the call is documented at https://pryntid.com/docs.

Keep acquisition out of the cohort row

Channel, promo and country belong in the segment. Fold paid and organic buyers into one row and a shift in spend looks like a change in retention.

Build the weekly grid

Anchor cohorts to calendar weeks

A cohort is everyone whose first-value event landed in the same Monday-to-Sunday week. Week N is the seven-day window N weeks later, with the same boundaries for everyone. Anniversary windows (“28 days after each user’s start”) are fine for one account and useless for comparing columns.

Decide what counts as a return

For buyers, an order placed that week; for sellers, a listing published or an order fulfilled. Logins are not retention — track them on a separate line so you can watch engagement move without confusing it with trade.

Report counts next to percentages

Week 0 always equals the cohort size, so keep it out of the percentage columns and print the denominator above each one. A cohort of fourteen buyers swings several points when one person returns.

Pair the grid with a first-week funnel

Retention shows who came back; the funnel shows where week 1 broke. Steps: first-value event → second order → third order within 28 days, per side. Prynt Analytics gives you weekly retention, 2–8 step funnels and goals in one view, with bot sessions excluded.

Read the grid with the noise removed

Two people building a grid of translucent tiles on a glass table

Exclude bots, test shops and staff

Automated traffic and AI agents produce sessions that look like returning users. Remove bot and agent sessions before the numbers reach you (https://pryntid.com/bot-detection covers the signals) and keep a suppression list for internal accounts, test shops and QA buyers.

Decide what a refund does to a cohort

A buyer whose week-1 order was refunded did not retain. Drop the membership or count the refund as negative; both are defensible, so pick one and write it down.

Compare like weeks and like cohorts

Read down a column to see whether newer cohorts improve, and across a row to see where each decays. Compare week N only between cohorts of similar size, channel mix and season.

The three questions the grid answers

  1. Which side is leaking? Buyers holding while sellers fade is a supply problem, and no discount fixes it.
  2. Did the product change? Week 1 improving for every cohort at once means you shipped something; only the newest cohort improving suggests a promo.
  3. Where is the leak? A cliff at week 1 points to onboarding; a slow decline from week 4 points to assortment, pricing or trust.

Write those answers for the last four cohorts once a week, with sizes attached, and send each side its own note.

FAQ

Person walking past floating blue orbs, some dissolving into mist

Is weekly retention better than monthly for a marketplace?

Weekly shows the first month, where most marketplace churn is decided; monthly smooths over the week-1 cliff. Track weekly for the first 8–12 weeks of a cohort, then switch to monthly for older ones.

Should buyers and sellers share one retention chart?

No. The value events differ and the cadence differs — sellers act weekly, many buyers order monthly. A blended chart moves with the mix of the two sides rather than with behaviour.

Is a user who logs in but never transacts retained?

Not in this grid. Logging in is engagement; on a marketplace, retention means trading. Keep engagement on its own chart so it can move without distorting the cohort.

Start with your own data. You do not need a warehouse to get the first grid. Prynt Analytics gives you weekly retention, funnels, goals and paths, excludes bot sessions from the count and honours Global Privacy Control; every Prynt product runs on one account and one bill (https://pryntid.com/platform). Start at https://pryntid.com.

Try it free

Prynt is device intelligence with a free plan: 5,000 identifications a month, visitor IDs with confidence and bot detection — no credit card. Upgrade to Pro for VPN, proxy, tampering and behavioral signals.

Keep reading