Research question
Which combination of observable measures can distinguish four micro-market states: insufficient demand, insufficient supply, quality or price mismatch, and inefficient allocation between sides?
The programme also asks when aggregation hides simultaneous scarcity and saturation, which measures lead or lag, and how conclusions change under alternative time windows and eligibility rules.
Practical significance
The protocol is intended for product, growth, revenue, operations, and analytics leaders who need to distinguish an acquisition problem from a quality, price, ranking, or execution problem. It does not propose a universal KPI, a company ranking, or a causal estimate of any intervention.
Working definitions
- Demand unit: an observable intention to obtain a product, service, contact, or fulfilment that meets a pre-published qualification rule.
- Supply unit: an available and eligible object or participant able to respond to qualified demand within a stated window.
- Match: a relevant connection between sides; it is not automatically an impression, contact, or completed transaction.
- Successful outcome: a pre-defined terminal event appropriate to the platform archetype.
- Micro-market: an analytical segment defined by category, geography, time, price, and other substitutability constraints. Its boundary is a testable research choice, not a natural fact.
Unit of analysis
The primary unit is micro-market × time window × cohort. Goods marketplaces, classifieds, services, on-demand, and lead-generation platforms must be described separately before aggregation. Inclusion, exclusion, deduplication, retry, and censoring rules must be public.
Measurement families
| Family | Working question | Candidate measures | Main interpretation risk |
|---|---|---|---|
| Availability | Can demand find eligible supply? | fill rate, zero-result rate, unmet demand | treating any exposure as a relevant outcome |
| Speed | How long until response and outcome? | time to match, first response, first transaction | ignoring censoring and seasonality |
| Unused supply | How much available supply meets no demand? | unmatched supply, stale supply, exposure coverage | confusing absent demand with low quality |
| Activation | When does registration become useful participation? | seller activation, time to qualified listing | choosing the event after observing outcomes |
| Reliability | Does a match reach its intended outcome? | cancellations, failed transactions, fulfilment failure | merging distinct failure mechanisms |
| Repeatability | Do sides return after an outcome? | repeat rate, cohort retention | treating repeat behaviour as proof of satisfaction |
| Density | Are enough eligible sides present in the micro-market? | category, geographic, temporal density | comparing density under different boundaries |
| Concentration | Does the outcome depend on a small group? | concentration ratios, top-decile share, Gini | inferring market power without separate analysis |
| Price | Is the gap explained by incompatible prices? | price coverage, competitiveness, dispersion | reading price without quality and terms |
| Match quality | Is the outcome relevant to both sides? | qualified response, acceptance, completion | optimising short-term conversion against long-term liquidity |
These are candidate measures, not a claim that every measure applies to every platform.
Data requirements
Priority sources are openly licensed datasets, public reporting with reproducible definitions, permissioned anonymised company data, and purpose-built survey or observational data. Every field needs provenance, licence, timezone, unit, allowed values, missing-data rules, and extraction date.
No user or employee personal data belongs in the public dataset. Internal company measures remain unpublished without written permission from the data owner.
Collection protocol
- Select one platform archetype and define its terminal success event.
- Register micro-market construction rules before inspecting final comparisons.
- Create an event dictionary and a mapping from source events to research variables.
- Test completeness, duplication, temporal order, missingness, and schema drift.
- Calculate at micro-market and cohort level while retaining numerator and denominator.
- Keep illustrative data separate and label all synthetic values as examples.
- Complete external methods and industry review before publishing findings.
Analysis and sensitivity
Publish distributions, cohorts, and missingness maps first. Then examine relationships between measures and diagnostic stability. Causal language requires a separate identification strategy with explicit assumptions.
Required sensitivity tests cover alternative micro-market boundaries, daily to monthly windows, eligibility definitions, exclusion of dominant segments, seasonal periods, repeated requests, censored observations, and mean-versus-quantile aggregation.
Alternative explanations
A measure can move because of instrumentation, marketing mix, traffic quality, price, seasonality, moderation, logistics, payment flow, or a changed event definition rather than liquidity. The report must assess these explanations before giving an operating interpretation.
Non-claims
Without a separate design, the study cannot establish causality, identify the “best” marketplace, prove future revenue, set a universal health threshold, or transfer values across non-comparable platforms.
Planned outputs
Methodology, glossary, open calculation template, clearly labelled illustrative data, web report, PDF, executive summary, slides, press release, and source-labelled chart pack. Listing an output does not mean it already exists.
External review and release gate
A methods reviewer checks units, denominators, censoring, and sensitivity. An industry reviewer checks operational applicability and omitted mechanisms. Legal review is required for non-public data. Reviewer names are disclosed only with consent.
The annual report remains at protocol version 0.x until every field has a data dictionary and publication permission, calculations reproduce from the public template, findings are separated from interpretation, and unresolved review comments are visible.