Marketplace template (two-sided)

The two-sided Marketplace template — separate buyer and seller health models governed by liquidity / match rate, GMV, and responsiveness.

A marketplace has two customers with opposite churn models: a buyer churns when they can't find a match; a seller churns when their listings don't sell. So the template ships one bundle per side, and you tag each account with a role of buyer or seller. The governing metric on both sides is liquidity — the match / fill rate — because most marketplaces fail when they can't maintain it.

Split accounts by role

A single tenant holds both sides. Set role: "buyer" or role: "seller" on each account via the account upsert, and provision the matching template (Marketplace — buyer or seller). Value is tracked as GMV (valueBasis: "gmv").

How a value becomes a score

Each component maps a raw value to a 0–100 score at its good / warn thresholds. Rate components (match, fill, response) use the generic ratio op; GMV uses sum_payload. Overall health is the weight-proportional average, banded Healthy ≥ 80 · Watch ≥ 60 · Risky ≥ 40 · Cold < 40. Buyer and seller are independent 100-weight bundles.

Buyer / demand side

The buyer's health is liquidity + repeat + spend. Match rate is the make-or-break signal — a buyer who searches without finding leaves — followed by repeat rate, recency, and GMV.

Evidence: liquidity is "the most critical aspect of a marketplace; most marketplaces fail because they never reach or maintain it" (a16z); the demand-side flywheel is driven by repeat purchase + referral, and GMV retention is the under-watched health metric (a16z – GMV retention).

3 signalscombined weight 62≈62% of score

Match rate (search → transaction)transaction ÷ search_performed

Transactions ÷ searches, as a %.

Higher is bettergood ≥ 20%warn ≤ 2%weight 24

How we collect it: Send `search_performed` on each search and `transaction` on each completed purchase; the `ratio` op divides them.
How to read it: Raw value is search-to-transaction conversion. ≥ 20% scores 100; ≤ 2% scores 30.
How to analyze it: Liquidity is the most critical aspect of a marketplace — most marketplaces fail because they never reach or maintain it. A buyer who searches but can't find a match leaves.
What to do: Fix the supply gap behind low match rates — inventory depth, relevance, or geography for the searches that fail.
Tune it: Anchor `good`/`warn` to your category's realistic conversion; a high-consideration marketplace converts far lower than an impulse one. Weight, thresholds, and the metric spec are all editable in Settings → Components.

Repeat transactions (90d)transaction

Transaction count in the last 90 days.

Higher is bettergood ≥ 4warn ≤ 1weight 20

How we collect it: Same `transaction` event, counted over 90 days.
How to read it: Raw value is the transaction count. 4+ scores 100; a single transaction scores 30.
How to analyze it: Repeat purchasing is the demand-side flywheel — a one-time buyer hasn't formed the habit that sustains GMV.
What to do: Drive the second and third transaction with reminders, saved searches, and category cross-sell.
Tune it: Set the window to your marketplace's natural purchase cadence. Weight, thresholds, and the metric spec are all editable in Settings → Components.

Days since last transactiontransaction

Recency of buyer activity.

Lower is bettergood ≤ 30 dayswarn ≥ 90 daysweight 18

How we collect it: Derived from the latest `transaction` event.
How to read it: Raw value is days since the last transaction. ≤ 30d scores 100; ≥ 90d scores 30.
How to analyze it: A lengthening gap between transactions is the earliest buyer-lapse signal.
What to do: Re-engage as recency crosses your repeat interval with fresh, relevant supply.
Tune it: Match good/warn to how often buyers naturally return in your category. Weight, thresholds, and the metric spec are all editable in Settings → Components.

3 signalscombined weight 38≈38% of score

GMV — buyer spend (90d)transaction

Total transaction value in the last 90 days.

Higher is bettergood ≥ 500warn ≤ 50weight 16

How we collect it: Put the transaction amount on the payload as `value`; the `sum_payload` op totals it.
How to read it: Raw value is summed GMV, banded by your thresholds.
How to analyze it: GMV retention is the marketplace health metric most teams ignore — a buyer whose spend is sliding is churning in value before they churn in logins.
What to do: Watch high-GMV buyers for spend drops; they're the highest-value win-back segment.
Tune it: Set the band from your average order value × expected orders per quarter. Weight, thresholds, and the metric spec are all editable in Settings → Components.

Referrals (180d)referral

Referral signals in the last 180 days.

Higher is bettergood ≥ 1warn 0weight 12

How we collect it: Send a `referral` event when a buyer refers someone.
How to read it: Any referral scores 100; none scores the floor.
How to analyze it: The demand-side flywheel is driven by repeat purchase and referral — referrers are advocates who churn far less.
What to do: Prompt happy repeat buyers to refer at their peak-satisfaction moment.
Tune it: Keep the weight modest since referrals are sparse. Weight, thresholds, and the metric spec are all editable in Settings → Components.

Account silenceany event

Days since any event.

Lower is bettergood ≤ 21 dayswarn ≥ 60 daysweight 10

How we collect it: Computed from every event on the account.
How to read it: Raw value is days since the last activity of any kind. Recent activity scores high.
How to analyze it: Even before transactions stop, a buyer who stops searching and browsing is drifting off the platform.
What to do: Re-engage before silence hardens; pair with recency for the full picture.
Tune it: Tighten for high-frequency marketplaces. Weight, thresholds, and the metric spec are all editable in Settings → Components.

Seller / supply side

The seller's health is utilization + supply depth + earnings. Fill rate (do listings sell?) and responsiveness govern liquidity from the supply side; active listings and earnings show whether the seller is invested and paid.

Evidence: supply-side liquidity (fill / utilization rate) is the governing marketplace metric (a16z glossary); sellers stay where they earn — declining earnings is the strongest supply-side churn tell.

2 signalscombined weight 36≈36% of score

Fill rate (listings → sales)transaction ÷ listing_created

Transactions ÷ listings, as a %.

Higher is bettergood ≥ 30%warn ≤ 3%weight 22

How we collect it: Send `listing_created` on each listing and `transaction` on each sale; the `ratio` op divides them.
How to read it: Raw value is listing-to-sale conversion. ≥ 30% scores 100; ≤ 3% scores 30.
How to analyze it: Supply-side liquidity: listings that never sell are wasted effort, and a seller who doesn't sell churns.
What to do: Help low-fill sellers with pricing, presentation, and demand matching before they give up.
Tune it: Set the band to your category's healthy sell-through rate. Weight, thresholds, and the metric spec are all editable in Settings → Components.

Response / acceptance ratetransaction ÷ transaction_request

Transactions ÷ transaction requests, as a %.

Higher is bettergood ≥ 70%warn ≤ 20%weight 14

How we collect it: Send `transaction_request` on each inbound request and `transaction` on acceptance.
How to read it: Raw value is the acceptance rate. ≥ 70% scores 100; ≤ 20% scores 30.
How to analyze it: Slow or declining sellers strangle liquidity for the whole marketplace — responsiveness is a supply-quality signal.
What to do: Nudge unresponsive sellers; surface the revenue they're leaving on the table.
Tune it: Set the band to your marketplace's service expectations. Weight, thresholds, and the metric spec are all editable in Settings → Components.

4 signalscombined weight 64≈64% of score

Active listings (90d)listing_created

Listings created in the last 90 days.

Higher is bettergood ≥ 5warn ≤ 1weight 18

How we collect it: Send `listing_created` when a seller publishes a listing.
How to read it: Raw value is the listing count. 5+ scores 100; ≤ 1 scores 30.
How to analyze it: Supply depth is the seller's investment in the platform — a seller who stops listing has one foot out the door.
What to do: Prompt inactive sellers to relist; make listing fast and rewarding.
Tune it: Set the band to a healthy listing cadence for your category. Weight, thresholds, and the metric spec are all editable in Settings → Components.

Earnings — GMV captured (90d)transaction

Total transaction value in the last 90 days.

Higher is bettergood ≥ 1000warn ≤ 100weight 18

How we collect it: Put the sale amount on the `transaction` payload as `value`; the `sum_payload` op totals it.
How to read it: Raw value is summed earnings, banded by your thresholds.
How to analyze it: Declining earnings is the strongest seller-churn tell — sellers stay where they make money.
What to do: Intervene on earnings drops with demand, promotion, or pricing help before the seller leaves.
Tune it: Set the band from your sellers' typical quarterly take. Weight, thresholds, and the metric spec are all editable in Settings → Components.

Days since last saletransaction

Recency of sales.

Lower is bettergood ≤ 30 dayswarn ≥ 90 daysweight 16

How we collect it: Derived from the latest `transaction` event.
How to read it: Raw value is days since the last sale. ≤ 30d scores 100; ≥ 90d scores 30.
How to analyze it: A seller who stops selling stops logging in — recency of sales predicts supply-side churn.
What to do: Reach out as the gap grows; a dormant seller is often one good sale away from re-engaging.
Tune it: Match good/warn to your category's normal sales cadence. Weight, thresholds, and the metric spec are all editable in Settings → Components.

Onboarding / first listingonboarding_completed

Whether the seller created their first listing / finished onboarding in the first week.

Pass / failweight 12

How we collect it: Send `onboarding_completed` when the seller publishes their first listing / finishes setup.
How to read it: Completed within 7 days of signup → 100; otherwise a low fixed score. A binary activation gate.
How to analyze it: A seller who never posts a first listing never activates — the first week is decisive for supply.
What to do: Make first-listing frictionless and recover stalled seller onboarding fast.
Tune it: This is a fixed-score activation signal — adjust its weight, not a threshold. Weight, thresholds, and the metric spec are all editable in Settings → Components.

Sending the data

Tag accounts with a role via the upsert; send liquidity events via the SDK or the HTTP API; map your names in Settings → Event map.

Signal	Kind	How to send
`role`	account field	`PUT /api/accounts/{id}` → `role: "buyer" \| "seller"`
`transaction`	event	on each completed transaction — include `{ "value": 49.90 }` for GMV
`search_performed`	event	on each buyer search
`listing_created`	event	on each seller listing
`transaction_request`	event	on each inbound request to a seller

Common raw names (transaction.completed, search.performed, listing.created, booking.requested, …) are pre-mapped, so most marketplaces resolve out of the box.

Enrich account data — set role, GMV, and kind: "person".
Build a custom dashboard — retune weights and thresholds per side.

Buyer / demand side

Seller / supply side

Sending the data

Next