Building the Growth Model

Strategic intent: Once your marketplace has reached liquidity, the next question is no longer "can we grow?" but "how do we systematically grow, predict the impact of our investments, and avoid burning cash on the wrong levers?". The Growth Model is what answers that question.

#Why this matters for a founder

Imagine you've crossed liquidity. Your platform is working. The board asks: "What happens if we invest €500K into paid acquisition vs €500K into improving onboarding?" — and you don't know.

That's the gap the Growth Model fills. It's not a strategy document. It's a spreadsheet (or equivalent) that captures the dynamics of your platform in numbers — so that, when you're staring at competing investment proposals, you can simulate each one and see which produces better outcomes 12 months out.

The Boundaryless framing borrows from Chris More: a growth model is "a mashup of funnels analysis and cohort analysis" — a way to relate actions you take to results you get, with enough rigor that the results are predictable rather than hopeful.

"A barely working, experimental growth model will help you perform better than not having one." — PDT Growth & Product Guide

That's worth dwelling on. Many founders avoid building a growth model because they fear it has to be perfect. It doesn't. The first version will be wrong in many places. That's fine — what matters is that having any model forces you to make assumptions explicit, and that surfacing assumptions is half the value.

#When to use it

There are four moments when investing in this technique pays off disproportionately:

• Before annual planning — the model produces the forecast and the scenario range. Without it, your plan is essentially fiction with confidence.
• Before fundraising — investors expect a defensible growth model. They don't expect it to be right; they expect it to show you understand the dynamics.
• When growth has plateaued and you can't pinpoint why — the model surfaces which assumption broke. Without one, you'll guess.
• When choosing between competing investment hypotheses — sensitivity analysis quantifies the impact of each lever. "We should invest in retention" becomes "improving cohort week-12 retention by 5pp grows GMV by X% over 12 months".

You should not run this technique pre-liquidity. There's nothing to model — the dynamics aren't stable enough yet.

#The mental model — output vs input metrics

Before getting into the spreadsheet, internalize the framing. The Reforge essay (Brian Balfour, Shaun Clowes, Casey Winters) puts it cleanly:

"Output metrics represent results and input metrics represent actions."

So new customers is an output metric — it's the result. Conversion rate is an input metric — it's the lever you can pull to affect the output.

This distinction is what makes the growth model useful for management. Output metrics tell you what happened. Input metrics tell you what to do about it. A model without input metrics is a dashboard. A model with input metrics is a steering wheel.

#How to build it

The Boundaryless framework breaks the construction into three key steps plus the operationalization in spreadsheet form.

1. #1. Establish your north star metric

   Pick the single metric that, alone, represents progress for your company.

   For marketplaces, the north star is typically a measure of the level of interactions between demand and supply — a transactions-per-period count, weighted by quality. For SaaS-marketplace hybrids it might be active workspaces, or active connections.

   The north star should not be revenue alone. Revenue is too lagging — it tells you what already happened and gives you no early warning. The right north star moves earlier than revenue, and revenue follows it.

2. #2. Establish 4–5 output metrics

   These are the KPIs that represent the value-creation process for the company. They typically cover four areas:

   • New customer acquisition — how many new users (per side, for marketplaces)
   • Customer retention — cohort retention curves on both sides
   • Customer engagement — frequency, depth, breadth of platform use
   • Healthy economics — monetization, LTV, contribution margin

   For two-sided marketplaces, you'll need separate metrics for supply side and demand side — and a small management function that looks at the equilibrium between the two and adjusts targets accordingly. This can be as light as one person spending two hours a week on the balance.

   Every marketplace will struggle in keeping demand and supply in equilibrium. The model should make that struggle visible, and you should have one or more levers that let you ramp either side up quickly when the balance breaks (even if temporarily uneconomic).

3. #3. Establish your input metrics

   These are the leading indicators that, when moved, predictably move the output metrics. Common ones:

   • User satisfaction (NPS, CSAT, qualitative)
   • Customer referrals (referral rate, K factor for viral loops)
   • Conversion rate at the bottleneck step of each loop
   • Activation rate — % of new users completing first transaction
   • Quality signals — review scores, refund rate, complaint volume

   Identifying which input metrics actually move which output metrics is the hardest, most contextual part of this work. Expect to discard several candidates before finding the leading indicators that work for your specific business.

4. #4. Operationalize as a spreadsheet model

   Now translate the framework into something the team can use every week. The spreadsheet has four interlocking pieces:

   a. Loop equations — for each active growth loop (viral, paid, content, sales — see Implementing Growth Loops), write the equation linking next-period new users to this-period activity. For a viral loop:

   NewUsers(t+1) = ActiveUsers(t) × InvitesSent × InviteAcceptance × Activation

   b. Cohort accounting — explicitly track cohorts as they progress through states: New → Activated → Active → Churned. Each cohort retains differently; aggregate views hide the dynamics.

   c. Unit economics — per period and per cohort: GMV (delivered, post-cancellations), Net Revenues, fully loaded CAC by channel, contribution margin per transaction, cohort LTV. Apply Fabrice Grinda's rules of thumb: fully loaded CAC recouped in 6 months, 3× in 18 months, decreasing CAC trend over time as network effects compound.

   d. Sensitivity & scenarios — identify the 3–5 levers with the largest impact on outputs, build Bear / Base / Bull scenarios with internally coherent assumptions, identify the breakeven point and the assumption gating it.

5. #5. Validate by backcasting

   Before forecasting forward, backcast: feed the model historical inputs from the past 6–12 months and check whether it reproduces the actual outcomes within ~10%. If it doesn't, refine before using it for forecasting. A model that can't reproduce the past is unlikely to predict the future.

#The model is for thinking, not forecasting precision

This is the most important mindset shift. Founders who come from operations or finance often expect the model to be right. It won't be.

What the model will do — and what makes it valuable — is surface the right questions: which assumption matters most? where's the bottleneck? what would have to be true for this scenario to play out? When the model reveals that breakeven depends on supply-side multi-tenanting dropping below 30%, you now know what to validate experimentally. That's the real output: the model points to the next experiment.

Once you have a working model, the goal becomes transitioning from one-off experiments to structured, replicable growth engines — the loops you've designed in Implementing Growth Loops become productionized through the discipline of metrics and forecasting that the model imposes.

#Inputs and outputs

#Required inputs

• Active growth loops with conversion rates per step (from Implementing Growth Loops)
• Chosen output and input metrics (from Choosing Platform Metrics)
• Historical data — at least 3–6 periods for cohort calibration

#Recommended inputs

• Comparable platform benchmarks (for sanity-checking conversion and retention assumptions)
• Channel-specific CAC data
• Supply / demand balance metrics for marketplaces

#Outputs

A spreadsheet model containing:

• User cohorts progressing through New → Activated → Active → Churned states
• Loop equations linking transactions to new users
• Per-period and per-cohort unit economics (LTV, fully loaded CAC, contribution margin)
• Sensitivity analysis on 3–5 key levers
• Bear / Base / Bull scenarios with documented assumptions
• Breakeven point with gating assumptions identified

Plus a 1-page brief summarizing the most insightful findings: the dominant lever, the gating assumption, the time-to-breakeven. This is what gets shared with the board.

#Process heuristics

A few practical heuristics from the Boundaryless community of practitioners:

• Document every assumption with a source or rationale. Hard-coded numbers without provenance corrode trust and make refresh impossible.
• Cohort-based, never aggregate-only. Aggregate metrics hide the dynamics that matter — a cohort with poor retention can be invisible behind a strong overall trend.
• Two-sided marketplaces need two separate models plus a balancing function. Don't try to model both sides as one system from day one.
• Refresh quarterly. The model decays as the business evolves. Schedule the refresh as a calendar invite, not a vague intention.
• Linear approximations break for non-linear loops. When you hit a discontinuity (network-effect tipping point, cold-start dynamics), note it explicitly in the model.

#Validation criteria

• North star metric explicitly named and defined
• 4–5 output metrics covering acquisition, retention, engagement, economics
• Input metrics derived from each output, with owning team
• Each active loop has an explicit equation
• User cohorts progress through states explicitly (not aggregated)
• Unit economics computed per period and per cohort
• LTV/CAC ratio computed and tracked
• At least 3 levers sensitized
• Bear / Base / Bull scenarios exist with documented assumption sources
• Backcast produces results within 10% of historical reality
• Breakeven gating assumptions identified and listed for validation

#Common mistakes

• Aggregate-only modeling — hides cohort dynamics, gives misleading averages
• Hidden assumptions — buried in formulas without comments
• Over-optimistic stack-up — every variable set at its best-ever value; the resulting "bull" is fantasy
• No backcast — a model that can't reproduce the past is unlikely to predict the future
• Updating only when convenient — quarterly refresh is non-negotiable; otherwise the model decays into uselessness
• Single-channel CAC — paid, content, viral, sales each have different CACs and LTVs; blend at your peril
• Pursuing model perfection before deployment — a 60% accurate model used weekly outperforms a 95% accurate model built once and forgotten

#Used in pipelines

• Engineering Sustainable Growth — as Phase 3

#Connections

• Requires: Implementing Growth Loops and Choosing Platform Metrics
• Feeds: annual planning, board reporting, fundraising decks
• Updated by: ongoing experimentation — model assumptions are revised as evidence accumulates
• Visualized in: the Growth Model Canvas (draft) — the canvas is the visual companion to the spreadsheet

#Related reading

• Key Metrics for Platform-Marketplaces — the canonical Boundaryless essay (Simone Cicero, Andrea Valeri, Manfredi Sassoli de Bianchi)
• The 4 Key Problems that Hinder Growth in Platforms and Marketplaces — diagnostic for stalled growth
• Seek Growth and Evolve your Platform Strategy — guide release announcement
• The PDT Growth & Product Guide — full chapter "Building a Growth Engine: the Growth Model" in Legacy PDT Growth