← Back to articles

How to Model a Two-Sided Marketplace: Lessons From Building the the platform Financial Model



Author: Yanni Papoutsi - Fractional VP of Finance and Strategy for early-stage startups - Author, Raise Ready Published: 2025-03-22 - Last updated: 2025-03-22

Reading time: \~10 min

Why Marketplace Models Are Different

A SaaS company acquires one type of customer and charges them a subscription. A marketplace operates two customer types whose behavior is interconnected. A staffing marketplace like the platform needed workers (supply) and employers (demand). Without enough workers, employers would not post shifts. Without enough employers, workers would not sign up. The financial model had to capture that interdependence or it was structurally wrong.

The most common mistake founders make is modeling a marketplace like a SaaS business: project total customers, multiply by average revenue, done. This ignores the dynamics that define marketplace economics. Fill rate (the percentage of demand that gets matched to supply) is a variable that changes based on the supply-demand ratio. Take rate can compress as volume grows. Unit economics differ by side, by geography, and by segment.

*Key insight: A marketplace model is a system model, not a customer model. You are modeling the interaction between two groups and the transactions that result. If your model does not capture the supply-demand balance, it will produce projections that look reasonable in a spreadsheet and are wrong in reality.*

The Architecture of a Marketplace Model

At the platform, the financial model was built around seven interconnected modules. Each one fed the next, and changing any single input rippled through the entire system.

Module 1: Supply Acquisition and Retention

Workers registered per month, activation rate (percentage who completed their first shift), retention at 30, 60, and 90 days, and reactivation rate for dormant workers. The supply pool at any given time was the cumulative registered workers minus churned ones, plus reactivated ones. The key insight from building this: worker retention was not a simple churn curve. Workers would go dormant for weeks and then reactivate when a shift matched their availability. The model needed to distinguish between true churn (permanently left) and dormancy (temporarily inactive), because the two required completely different strategies and had different financial implications.

Module 2: Demand Acquisition and Retention

Employers onboarded per month, conversion rate from sign-up to first shift posted, retention (percentage still posting shifts after 3, 6, 12 months), and frequency (shifts posted per active employer per month). Demand-side metrics were more stable than supply-side because employer contracts were stickier, but frequency varied significantly by segment. A large hospitality chain posted 50+ shifts per week. A small office posted 2-3 per month.

Module 3: Matching and Fill Rate

This is the module that makes marketplace modeling hard. Fill rate = shifts filled / shifts posted. It depends on the supply-demand balance in each market. A market with 500 available workers and 100 shifts per day has a very different fill rate than one with 500 workers and 1,000 shifts per day.

We modeled fill rate as a function of supply density (available workers per shift). Below a certain density threshold, fill rate dropped sharply. Above it, fill rate plateaued near 90-95%. This S-curve relationship was critical for the model because it meant that adding supply in a market with low density had a disproportionate impact on revenue.

Module 4: GMV Calculation

GMV (Gross Merchandise Value) = shifts filled x average shift value. Average shift value = hours per shift x hourly rate. This sounds simple, but hourly rates varied by role type, geography, and time of day. Night shifts and weekend shifts commanded premiums. London rates were higher than Manchester rates. The model needed to capture the mix, not just the average.

Module 5: Revenue and Take Rate

Revenue = GMV x take rate. The take rate was the percentage of each transaction that the marketplace kept as revenue. But take rate was not a constant. It varied by client segment (enterprise clients negotiated lower rates), by worker type, and by market. Over time, as the marketplace scaled and competition increased, there was pressure on take rate.

Modeling take rate as a single number would have been wrong by 3-5 percentage points depending on the mix. We segmented it by client size (SMB, mid-market, enterprise) and geography, which produced much more accurate revenue projections.

Module 6: Cost of Fulfillment (COGS)

For each filled shift: worker payment, employer liability insurance, payment processing fee, and a per-shift allocation of the operations team cost. Gross margin per shift was the take rate minus these costs. At early stages, gross margin per shift was thin, 15-20% of net revenue. At scale, it improved to 25-30% as insurance rates came down with volume and operations became more efficient.

Module 7: Unit Economics by Cohort

Each employer cohort's LTV was calculated as: average monthly gross profit per employer x average retention duration. Each cohort had its own acquisition cost (tracked by channel and month). The LTV:CAC ratio by cohort was the number that told us whether the marketplace was working, and it was the number the acquirer scrutinized most during exit diligence.

The Supply-Demand Balance: Modeling the Flywheel

The marketplace flywheel is the dynamic that makes or breaks a two-sided platform: more supply attracts more demand, which attracts more supply. In the model, this translates to:

Higher worker density in a market leads to higher fill rates. Higher fill rates lead to better employer experience. Better employer experience leads to higher employer retention and referrals. More employers lead to more shifts posted. More shifts lead to better worker experience (more work available). Better worker experience leads to higher worker retention.

Modeling this loop requires careful thought about which direction to start from. We chose to model supply first (worker acquisition and retention), then demand (employer acquisition and retention), then the matching layer (fill rate as a function of the supply-demand ratio), and finally economics (revenue, COGS, unit economics).

The flywheel creates a non-linear relationship between investment and output. Doubling marketing spend in a market that is already past the liquidity threshold might improve fill rate by 2 points. Doubling spend in a market below the threshold might improve fill rate by 15 points. The model needed to capture this non-linearity, or capital allocation decisions would be wrong.

*Key insight: The financial model for a marketplace is also the strategic model. It tells you which markets to invest in (those closest to liquidity threshold), which to slow-play (those too far from threshold for current budgets), and which to monitor (those already at high fill rates where additional supply has diminishing returns). At the platform, the model directly informed market expansion sequencing.*

How This Model Supported a Successful Exit

When the acquirer completed the acquisition, the financial model was the centerpiece of the valuation discussion. The acquirer's diligence team ran their own analysis on our data, but the structure of our model gave them a framework to work within.

Three specific elements of the marketplace model made a material difference during exit:

Cohort-level employer economics. The model showed that employer cohorts were not just retained, they were spending more over time. This expansion pattern, visible only because we tracked cohort economics, was one of the key factors supporting the revenue multiple.

Fill rate dynamics by market. The model demonstrated that newer markets were still climbing the fill rate curve, meaning significant untapped value in the existing footprint. The acquirer could see where growth would come from without entering new markets.

Take rate sensitivity. When the acquirer asked what would happen if take rate compressed by 2 points, we could show the impact across every segment and geography within minutes. That responsiveness built confidence in the model and, by extension, in the management team.

Common Marketplace Modeling Mistakes

Modeling revenue without modeling both sides. If your model shows revenue growing without an explicit supply and demand layer, it is a SaaS model wearing a marketplace costume. Model each side independently and connect them through the matching function.

Using a constant take rate. Take rates evolve with scale,

competition, and client mix. Model the components that affect take rate rather than hardcoding a single percentage.

Ignoring geographic variation. A marketplace in London and a marketplace in Manchester have different supply costs, demand densities, and competitive dynamics. Aggregate models hide these differences and produce misleading unit economics.

Confusing GMV with revenue. A $10M GMV marketplace with a 15% take rate generates $1.5M in revenue. Investors have seen founders pitch GMV as if it were revenue. It destroys credibility instantly.

Not modeling the liquidity threshold. Every marketplace has a point where supply-demand density is sufficient for reliable matching. Below it, the marketplace does not work. Above it, network effects kick in. If your model does not capture this threshold, it cannot tell you how much you need to invest to get a new market to viability.

Summary

A two-sided marketplace model is a system model that captures the interaction between supply and demand. Build it in modules: supply acquisition and retention, demand acquisition and retention, matching and fill rate, GMV, revenue and take rate, COGS, and unit economics by cohort. Model the supply-demand balance explicitly because it drives fill rate, which drives revenue, which drives everything. Segment by geography and customer size. Track take rate evolution. And remember that the marketplace flywheel is a real dynamic that the model must capture, not just a slide in the pitch deck.