Building & Shipping PM

The vendor dashboard nobody reads had the signal the whole time.

Building & Shipping Building & ShippingAdvanced

Supply-Side Root Cause Analysis, When Your Marketplace Metrics Drop and the Problem Is Not Your Users

Consumer-side signals move first, but they rarely contain the actual diagnosis. This article shows how to read supply-health signals independently from demand signals, and why most teams run the wrong experiment when conversion drops.

The Wrong Dashboard Is Open

Your conversion rate dropped. Your product manager scheduled a checkout audit. Your designer pulled heat maps. Your growth team queued three A/B tests.

None of this is wrong, exactly. It is just aimed at the wrong half of the system.

Marketplaces are two-sided. The metrics most teams watch, session-to-order rate, checkout abandonment, consumer DAU, live entirely on the demand side. They are real signals. They are also the last place the problem shows up when something breaks on the supply side.

By the time a supply failure registers in consumer conversion, it has already been degrading for weeks. The vendor dashboard nobody reads had the signal the whole time.


What Two-Sided Root Cause Analysis Actually Means

A two-sided Root Cause Analysis is not a checklist. It is a discipline of reading supply signals and demand signals independently before attempting to cross them.

Most teams cross them immediately, they see falling conversion and start looking for a cause anywhere in the funnel. This conflates two separate systems that fail in completely different ways and at completely different timescales.

The demand side fails fast and visibly. A broken payment gateway, a confusing step in checkout, a price increase, these show up in session recordings within hours. Consumer behavior is legible because consumers leave traces you designed the product to capture.

The supply side fails slowly and invisibly. Vendor churn, onboarding degradation, and matching layer latency accumulate over days or weeks before they surface in any consumer-facing metric. You did not design the product to surface these traces, which is the core problem.

Reading them independently first means: before you look at where consumer drop-off increased, you look at whether supply coverage, quality, and responsiveness changed over the same window.


Consumer-Side Signals vs Supply-Side Signals

Dimension Consumer-Side Root Cause Analysis Supply-Side Root Cause Analysis
First metric to check Session-to-conversion rate, checkout abandonment by step Active vendor count, fulfillment rate, median vendor response time
Failure timescale Hours to days Days to weeks
Who typically owns diagnosis Product + Growth Operations + Marketplace Health (if it exists)
Biggest diagnostic mistake Adding more consumer tracking when the data is already sufficient Not having supply instrumentation until after an incident
Signal legibility High, consumer events are logged by design Low, vendor-side events are often absent or siloed
Correct first experiment Funnel step A/B test, UX audit Vendor cohort analysis, onboarding funnel audit
Lagging indicator risk Low, consumer drop-off is usually contemporaneous with the cause High, supply degradation can be invisible for weeks before consumer metrics move

The table above is not a framework. It is an acknowledgment of a structural asymmetry that exists in every marketplace product: you know far more about what your consumers do than what your vendors experience.


Zomato's Non-Metro Collapse, The Case Study

In its expansion into Tier 2 and Tier 3 cities across India, Zomato faced a failure mode that looked, from the outside, like a demand problem.

Consumer daily active users in these markets held steady. Session volume was up. The brand recognition campaign was running. By every demand-side metric, the product was working.

What was not working was the restaurant onboarding flow.

As Zomato scaled beyond the dense urban markets where it had been built, the onboarding system, menu digitization, photo upload, compliance documentation, began breaking under geographic pressure. Restaurant operators in smaller cities had lower digital literacy, different document requirements, and less reliable internet connections than the metro partners the flow had been designed for.

The result: restaurants that had been onboarded were churning off the platform at a rate that did not trigger any consumer-facing alert. Existing restaurants stayed live. New restaurants were failing onboarding silently. Menu coverage in key categories thinned.

When a consumer in Nagpur searched for "biryani" at 8 PM on a Friday, the platform surfaced two restaurants instead of eight. Both had long delivery times. One cancelled. The fulfillment rate in that market dropped 14% over six weeks.

Consumer DAU held. Conversion dropped. The growth team audited the checkout flow.

The actual problem had been accumulating in the vendor onboarding funnel, a dashboard that operations reviewed weekly, not in real time.

The intervention that mattered was not a checkout A/B test. It was a geo-segmented onboarding audit, followed by a redesign of the documentation and photo upload flow for low-bandwidth conditions. The consumer-side metrics recovered only after supply coverage recovered.


How to Build Supply-Health Instrumentation Before the Next Incident

Most teams build supply instrumentation reactively, they add vendor dashboards after an incident surfaces a gap. This is the wrong order of operations.

Supply health instrumentation belongs in the product before you need it, the same way error monitoring belongs in an API before it goes down.

What the instrumentation needs to track:

  • Active vendor count by geography and category, with a week-over-week delta
  • Onboarding funnel completion rate by cohort, not aggregate, by cohort
  • Vendor-side cancellation rate and reason code distribution
  • Matching layer latency: how long between a consumer request and a qualified supply match
  • Menu freshness signal: when did this vendor last update availability or pricing

The design constraint that most teams miss: these metrics need to be co-observable with demand metrics in a single view. If diagnosing a conversion drop requires opening two separate dashboards maintained by two separate teams, the instrumentation is fragmented.

Supply health is not an operations metric. It is a product metric that operations happens to own the inputs for. The moment you separate them organizationally, you introduce a 48-to-72 hour delay into every incident response.

The threshold question: what triggers an alert? Not "supply count fell below X", that is a lagging trigger. The leading trigger is: vendor onboarding funnel completion rate in a given geo dropped more than 15% in a seven-day window. That signal predicts a fulfillment rate drop four to six weeks before the consumer conversion metric moves.

Building to the leading indicator is the difference between catching a supply failure in week two and explaining it to leadership in week eight.


The Judgment Turn

If your marketplace's conversion dropped and your team ran a checkout A/B test, you ran the wrong experiment.

This is not a criticism of A/B testing. It is a criticism of a default diagnostic posture that treats the consumer funnel as the primary system and the vendor layer as a supporting detail.

In a two-sided market, supply is not a supporting detail. It is the product. The consumer funnel is a presentation layer on top of supply coverage, supply quality, and supply responsiveness. When those degrade, the consumer funnel fails, but the failure looks like a UX problem from where most teams are standing.

The uncomfortable position: if you do not have real-time supply-health instrumentation today, you are operating your marketplace on half the data your decisions require. You will misdiagnose the next incident. You will spend two to four weeks optimizing the wrong half of the system. And the vendor problem that caused it will keep compounding while you iterate on button color.

The teams that get this right do not have better analytical skills. They made one structural decision earlier than everyone else: they treated vendor-side telemetry as a first-class product requirement, not an operations nice-to-have.


Key Takeaways

  1. Consumer metrics move first, but they do not contain the diagnosis when the failure is on the supply side. Read both sides independently before crossing them.

  2. Supply failures operate on a longer timescale than demand failures. By the time conversion drops, the supply degradation is already weeks old.

  3. The Zomato non-metro case is the pattern, not the exception, consumer DAU held, fulfillment collapsed, and the root cause was a vendor onboarding flow that was never designed for geographic scale.

  4. Supply-health instrumentation must be built before an incident, co-observable with demand metrics, and triggered by leading indicators, not lagging ones like vendor count.

  5. If your team defaulted to a checkout audit when conversion dropped, the problem is not the audit. The problem is that supply-side signals were not available to redirect the diagnosis.


Related Articles

Warm-up Reps

Did it land?

0 / 1 CORRECT
Three quick checks on the ideas above. Pick an answer and you will see why it is right or wrong. Consider it the warm-up before the real gym.
Q1
Zomato's non-metro order fulfillment drop is an example of which failure mode?
The root cause was the restaurant-side onboarding flow breaking under geographic scale pressure, not any consumer-facing change. Consumer DAU held, which masked the supply degradation until fulfillment rate collapsed.
AW

Anmoll Wadhwa

Senior PM · writing The PM Code

Field notes on product judgment: essays, teardowns, and reps for PMs who would rather think than template. A sharper take most days on LinkedIn.

More like this. Once a week.

Tactical essays on the calls that actually matter. In your inbox before they are on the feed.

LEARN·BUILD·COLLABORATE