Two companies report 30% growth in what management calls “business scale”. One is a SaaS company and leads with ARR. The other is a marketplace and leads with GMV.

A dashboard can make those figures look like peers. They are not. ARR is normally a recurring or run-rate business measure whose exact construction depends on the issuer. GMV is transaction activity flowing through a platform. Neither label, by itself, tells you recognised revenue or the economics left after serving that activity.

That makes the first job surprisingly mundane: work out what the denominator actually represents.

Start with the object being measured

A SaaS company might organise analysis around an account, subscriber, contract, seat or workload. A marketplace may need several units at once: consumers, merchants, suppliers, orders or transactions. If revenue is attached to one unit, acquisition cost to another and retention to a third, a neat ratio can become analytically useless.

I use “economic unit” as a working discipline here: the smallest decision unit to which revenue, variable cost, retention and acquisition economics can be attached consistently. It is not an accounting-standard term.

Issuer-defined metrics show why this matters. Spotify’s Premium ARPU is constructed from quarterly Premium subscription revenue, average daily Premium Subscribers and a three-month period. The familiar label hides a specific numerator, denominator, population and time basis. Spotify 2025 Annual Report

So two companies can both publish “ARPU” and still be measuring different populations. Before comparing the result, I want the calculation contract.

Transaction volume and company revenue separate very quickly

Marketplaces make this visible because the transaction number is often enormous. Airbnb’s Gross Booking Value includes host earnings, service fees, cleaning fees and taxes, net of cancellations and alterations. That is a measure of activity through the marketplace, not another name for Airbnb’s recognised revenue. Airbnb 2020 Form 10-K

The next question is how much of that activity becomes value captured by the platform. Take rate is one way to express it, but even that ratio inherits the definitions underneath. eBay defines take rate as net revenues divided by GMV. Period, currency, transaction population, cancellations and refunds therefore matter before two percentages can be placed side by side. eBay 2024 Form 10-K

SaaS metrics have their own version of the same trap. ARR and MRR are useful measures of recurring business scale, yet they do not automatically equal GAAP or IFRS revenue. Contract structure, usage, minimum commitments and services can change the bridge between a run-rate measure and recognised revenue.

A 30% increase in ARR and a 30% increase in GMV may both be good news. They still describe different movements in the business.

SaaS and marketplace metrics mapped across economic unit, activity, value capture, variable economics and cohort durability. ARR is not GMV, GMV is not revenue, and NRR above 100% can coexist with churn.

Rows align economic questions, not identical metric types. ARR/MRR summarise recurring business scale; take rate is a marketplace monetisation bridge from GMV/GOV towards platform revenue.

NRR can rise while customers disappear

Suppose a cohort starts with 100 accounts. Ten churn, so logo retention is 90%. The remaining 90 expand enough for cohort revenue to move from 100 to 110. Revenue retention can exceed 100% even though customers were lost.

That single example is enough to keep logo retention, GRR and NRR in separate boxes. Logo retention asks whether units survive. GRR asks how much revenue the starting cohort keeps after churn and contraction, normally before expansion. NRR puts expansion from that same cohort back into the calculation.

Snowflake’s Net Revenue Retention Rate fixes an existing customer cohort and measurement window before comparing revenue from that cohort. Snowflake Fiscal 2025 Annual Report Cloudflare’s dollar-based net retention follows the same broad discipline by comparing annualised revenue from a fixed set of paying customers. Cloudflare 2021 Annual Report

A marketplace complicates the question because there may be no single customer population. DoorDash has consumers, merchants and fulfilment-side participants with different acquisition, activation, retention and service economics. Side-specific CAC and retention therefore matter. A marketplace-wide “customer retention” or CAC/LTV number can hide movements that management needs to see separately. DoorDash 2024 Form 10-K

The age of the cohort can move the headline number

Now take two businesses with identical underlying behaviour but different growth profiles. The faster-growing company has a larger share of young cohorts; the slower-growing one has more mature cohorts. If churn, usage, order frequency or margin changes with tenure, their aggregate retention rates can diverge even when the underlying curve has not improved.

Fader and Hardie show the methodological reason: customers surviving to different tenures are not a homogeneous population, and survivor composition affects the retention pattern observed in aggregate. How to Project Customer Retention

Time-aware analysis appears in other businesses for the same reason. LendingClub discusses originations and credit performance through vintage and cohort views, which helps keep maturity from disappearing inside a single average. LendingClub 2024 Annual Report

For a comparable cohort, I want the entry date, observation window, product and pricing state, population definition and cost basis to remain visible. When retention or economic value changes materially with tenure, CLV or survival methods can model those conditional paths instead of forcing one average across everyone. Berger and Nasr’s work is a foundational reference for that family of customer-lifetime-value models. Customer Lifetime Value: Marketing Models and Applications

Margin comparisons need a cost boundary before they need a benchmark

Once activity, revenue and cohorts are aligned, the analysis can move below revenue. This is where a second denominator problem appears: what costs are included?

For SaaS, infrastructure, support and third-party service costs influence the economics of serving usage. For marketplaces, consumer and merchant incentives, payment costs, fulfilment, support and refunds may sit in different accounting or non-GAAP layers. A percentage can improve because operations improved, but it can also improve because the metric boundary changed. Different treatment of incentives, including whether they are netted against revenue, can also change the apparent monetisation layer.

DoorDash’s Q1 2025 results separate Marketplace GOV, revenue and monetisation/contribution measures. That lets an analyst follow transaction activity into platform revenue and then into the economics remaining after the relevant cost boundary. DoorDash Q1 2025 Financial Results

For a marketplace, a useful working bridge is:

orders × average transaction value → GMV / GOV → monetisation → revenue → variable contribution

It is a diagnostic sequence, not a universal accounting identity. A higher take rate can coincide with heavier incentives, more expensive fulfilment or a different transaction mix, so the ratio alone does not settle whether the business became more attractive.

Any bridge also has to reconcile. If the change in revenue is attributed to volume, average order value, mix, incentives and monetisation, those mutually exclusive components plus an explicit residual should return to the original movement. That discipline is consistent with reconciliation thinking used more broadly in financial analysis. IAS 7 Statement of Cash Flows

Snowflake and DoorDash become comparable only after translation

At this point the native KPI names matter less.

For Snowflake, the analysis can begin with a customer or account cohort, then trace usage or contracted activity into product revenue. Service costs define the next boundary. Retention then needs to separate surviving logos from revenue expansion within the starting cohort.

DoorDash starts somewhere else. Orders and Marketplace GOV describe activity across a multi-sided network. Monetisation converts part of that activity into platform revenue. Incentives, payments and fulfilment or service costs determine how much economics remain. Durability has to be checked on the relevant side of the network, including repeat frequency and cohort maturity.

Those are different businesses, but they can be translated into the same management sequence: identify the unit, observe the activity, trace the value captured, locate the variable cost boundary, then ask whether the result persists for a comparable cohort. The translation makes comparison possible without pretending that ARR and GMV are substitutes.

A hybrid business makes missing definitions more dangerous

Company labels blur over time. SaaS businesses add usage pricing, payments, services or advertising. Marketplaces add subscriptions, logistics, financial services or first-party commerce. The more revenue engines a company adds, the less useful a one-word business-model label becomes.

When two KPIs land on the same slide, I use five checks before treating them as comparable.

Unit. Are both metrics attached to the same kind of economic object?

Activity. Is each figure measuring transactions or usage, a run-rate measure, or recognised/captured revenue?

Value capture. Are numerator, denominator, period, currency, cancellations and refunds aligned?

Variable economics. Do the cost boundaries match, and where do incentives, payments or fulfilment sit?

Cohort durability. Are the populations comparable in maturity, product state and commercial terms?

If one of those answers is missing, the comparison stops there.

References