In this article
A marketplace can have plenty of listings and still be empty when a buyer needs something. The plumber is too far away, the appointment is already taken, or the available provider cannot do that particular job. Counting registered sellers will not tell the product team which problem to fix.
Start with the attempted job. Where did the buyer need help, when did they need it and what would have counted as an acceptable match? Build a demand ledger that includes the failed attempts. The strongest opportunities may be in the requests that never become transactions.
Give one customer need a stable identity
Define a demand episode as one attempt to satisfy a specific job within a bounded period. A buyer may search several times, change a filter and contact two suppliers before succeeding. Counting every search as a fresh need can make a difficult experience look like growing demand.
For an appointment marketplace, an episode might begin when a buyer specifies a service, area and acceptable time range. It ends with a confirmed booking, an explicit withdrawal or the agreed observation deadline. Keep the underlying searches so the team can inspect the path, but use the episode to calculate fulfillment.
Set rules for materially changed requirements. A buyer switching from a routine inspection next week to an emergency repair tonight has changed the job. Preserve that change instead of merging incompatible requests or resetting the record whenever results look poor.
Classify what prevented the match
The useful failure categories describe different operational responses:
- No relevant inventory exists for that area and service.
- Inventory exists, but it lacks a required attribute or capability.
- A suitable supplier appears, but cannot accept the requested time.
- A suitable, available option exceeds the buyer’s stated budget.
- The episode remains unresolved or lacks enough information to classify.
Do not infer “too expensive” merely because a buyer leaves. It requires a stated constraint or another defensible signal. Likewise, no booking after a search is not proof of insufficient supply. The buyer may be comparing options or may have completed the job elsewhere.
Allow multiple diagnostic flags, but choose a documented primary outcome for totals that must add up. Preserve unknowns. A ledger with honest gaps is more useful than a precise-looking report built from guesses.
See how the aggregate hides the weekend problem
Consider an invented local services marketplace with 200 demand episodes, split equally across four segments. All episodes in this teaching example have a known outcome within seven days; real reporting should retain unresolved episodes separately.
| Segment | Episodes | Matched | No inventory | Unsuitable | Unavailable | Over stated budget |
|---|---|---|---|---|---|---|
| North, weekdays | 50 | 35 | 3 | 4 | 5 | 3 |
| North, weekends | 50 | 10 | 20 | 8 | 8 | 4 |
| South, weekdays | 50 | 40 | 2 | 2 | 4 | 2 |
| South, weekends | 50 | 25 | 5 | 6 | 8 | 6 |
| Total | 200 | 110 | 30 | 20 | 25 | 15 |
The overall match rate is 110 / 200, or 55%. The segment rates are 70%, 20%, 80% and 50%. A general campaign to recruit more sellers would overlook the sharpest problem: North’s weekend demand, where 20 of 50 episodes have no inventory at all.
That pattern supports a focused supply investigation, not an automatic recruitment decision. Confirm whether the missing supply could operate at an acceptable price and whether demand is likely to recur. Recruiting suppliers into a segment with little sustainable work can simply move the frustration to the other side.
Separate listed supply from usable supply
A supplier profile is an asset in the catalog. Usable supply is a capability available for the buyer’s requirements at the relevant time. That distinction has to survive search, selection and confirmation.
Airbnb’s calendar guidance gives a concrete example: blocked dates affect search visibility, while advance notice, trip length and preparation time constrain availability. A published listing therefore cannot be treated as interchangeable with a bookable option for a particular request.
In your marketplace, inspect whether availability is stale, whether a supplier’s service area is accurate and whether the system knows the necessary job attributes. If suitable suppliers are repeatedly shown but cannot accept the work, more profiles may make the search results longer without improving fulfillment.
For the illustrative ledger, the 25 unavailable outcomes deserve a different investigation from the 30 no-inventory outcomes. The first may call for calendar accuracy, capacity handling or faster confirmation. The second may require adding a missing type of supply or narrowing the promise made to buyers.
Put time beside the match rate
a16z’s marketplace metrics framework treats match rate, market depth and time to match as separate dimensions. That is useful because eventual success can still arrive too late for the customer’s job.
Measure time from the episode’s qualified start to the agreed success event. A confirmed appointment and a completed service answer different questions; choose the one relevant to the decision and keep the other available. Define how reschedules and canceled bookings affect the record.
Report the share matched within a meaningful deadline across all eligible episodes. You can also report the median time among matched episodes, but label that population. Otherwise, a fast median can look healthy while the marketplace leaves many buyers unmatched entirely. Recently opened episodes also need time to mature before a final failure rate is assigned.
Choose an intervention that matches the gap
For North weekends, interview a small set of buyers about the missing jobs and ask suitable suppliers what would make those slots viable. For unsuitable inventory, improve the attributes that determine fit before changing ranking. For unavailable suppliers, test the freshness of the capacity signal. For budget failures, inspect the actual cost of serving the job before assuming a discount solves it.
Avoid widening the service area or time window without the buyer’s agreement. A distant option can improve a technical match count while failing the original need. Keep the original constraint in the episode so substitutions remain visible.
Review the same segments after an intervention, including whether sellers actually complete the work and receive enough demand to remain active. A descriptive before-and-after change can guide the next investigation; it does not isolate causality when seasonality or demand mix also changes.
The ledger should end in a specific operating choice: recruit a particular capability into a particular window, repair an availability problem, or stop promising coverage the marketplace cannot provide. That decision is easier to connect to a product growth model than a rising count of seller profiles.
Sources
Something we should clarify?
Suggest a correction


