Build Your Growth Model: The Equation That Connects Every Loop

A growth model is a spreadsheet equation that turns your acquisition, activation, retention, and referral loops into projected users and revenue over time. You build it by defining each loop's inputs and output multipliers, chaining them together, then testing which single input moves the whole model most.

Quick Answer: Growth = (New Users from Acquisition + Referred Users) × Activation Rate × Retention Curve, compounded monthly, minus churn. Map each loop's channel inputs to its output multiplier in a spreadsheet, then run a bottleneck analysis to find the one lever with the highest leverage.

Most growth strategy documents are prose: "we'll invest in content, improve onboarding, and build a referral program." That's directionally fine, but it's not testable. A quantitative growth model forces every claim into a number with a source, a formula, and a projected output — which means you can finally ask "which of these three actually matters most?" instead of guessing.

Why a Spreadsheet Beats a Strategy Deck

A growth model beats a strategy narrative because it exposes hidden assumptions as explicit numbers you can stress-test, instead of leaving them buried in adjectives like "strong" or "significant." Once a claim has a cell reference, you can change one input and watch the whole forecast move.

Growth strategy decks tend to bury the real assumption inside soft language. "Referral will be a meaningful growth driver" sounds like a plan. It isn't one until you can answer: meaningful compared to what, at what referral rate, from what user base?

Brian Balfour, co-founder of Reforge and former VP Growth at HubSpot, has long argued that growth teams fail not from lack of effort but from lack of a systems view — treating growth as a list of tactics instead of a connected loop structure with measurable inputs and outputs. A spreadsheet model is how you operationalize that systems view.

The other benefit is speed of disagreement. When a debate is "will content marketing or referral drive more growth," a model turns it into "content marketing's k-factor and CAC assumptions are on row 12; referral's are on row 24 — which set do you believe less?" That's a much shorter argument.

What the Model Actually Needs to Contain

At minimum, your model needs four connected sections: acquisition inputs, activation conversion, a retention curve, and a referral coefficient, all flowing into a compounding user-count formula. Each section is a small table, not a single cell.

• Acquisition: channels, spend or effort per channel, and resulting new users
• Activation: percentage of new users who reach your activation metric
• Retention: a decay curve showing what percentage of activated users remain active at week 1, 4, 12
• Referral: the rate at which retained users generate new users (your k-factor)

Each of these is a loop with its own math. The model's job is to connect them, not treat them as parallel silos.

The Core Growth Equation, Broken Into Loops

The core equation is: Total Users(t) = Users(t-1) × Retention Rate + New Users from Acquisition(t) + New Users from Referral(t), where Referral(t) is a function of Users(t-1) and your k-factor. This single line is what most "growth model" spreadsheets are actually computing underneath their tabs.

Written as loops instead of algebra:

1. Acquisition loop: marketing spend or content output → impressions → signups
2. Activation loop: signups → activated users (those who hit the aha moment)
3. Retention loop: activated users → returning users over time, governed by a retention curve
4. Referral loop: retained/engaged users → invites sent → new signups (feeding back into acquisition)

The compounding nature of loop 4 is what separates a growth model from a simple acquisition forecast. A linear acquisition-only model grows arithmetically. A model where referral feeds back into acquisition can grow geometrically — or flatline, if the loop coefficient is below 1.

Translating Loops Into Spreadsheet Rows

Each loop becomes a labeled block of rows: one row per input variable, one row for the conversion or decay rate, and one row for the resulting output number, so any teammate can trace an output back to its assumptions.

Build this as an actual monthly time series — columns for months 1 through 24, rows for each metric above — not a single static snapshot. A one-period model tells you where you are; a time-series model tells you where you're headed and how fast.

The Input-Driver Structure: Channels In, Loop Outputs Out

The input-driver structure means every loop has upstream "driver" inputs you actually control (spend, headcount, content cadence) and downstream "output" metrics the loop produces (signups, activated users, referrals), and the model's real value is in making that channel-to-output mapping explicit and adjustable.

This matters because teams often optimize the wrong layer. A PM might obsess over the activation rate (an output) without examining the inputs that determine it — like time-to-value or onboarding friction. The input-driver split forces you to ask, for every metric, "what upstream lever actually moves this?"

Example Input-Driver Map

• Acquisition output (new signups) is driven by: ad spend × CPC-implied traffic × landing page conversion, or content pieces published × average organic traffic per piece × signup conversion
• Activation output (activated users) is driven by: onboarding completion rate × time to value × in-product guidance quality
• Retention output (retained users) is driven by: product habit strength, notification/re-engagement cadence, and how well the product maps to a genuine job to be done
• Referral output (referred signups) is driven by: invites sent per active user × conversion rate per invite, both of which are themselves functions of product design (share prompts, incentives, natural sharing moments)

Once every output has explicit drivers, your model stops being a black box. You can label each driver cell with a source: "measured from last quarter's cohort," "benchmark from Reforge's growth loop data," or "assumption — needs validation." That labeling discipline alone catches a lot of bad forecasting before it ships.

Bottleneck Analysis: Finding the One Lever That Moves Everything

A bottleneck analysis means changing each input variable by a fixed percentage, one at a time, holding all others constant, and recording the resulting change in total users at month 12 — the input producing the largest swing is your bottleneck, and it's usually not the one your team is spending the most time on.

This is the single most useful exercise a spreadsheet model enables that a strategy deck can't. Prose strategy has no mechanism for sensitivity testing. A spreadsheet does it in minutes.

Worked Example

Say your baseline model, at month 12, projects 50,000 total users from these inputs:

In this example, retention is the bottleneck, not acquisition. A 20% lift in week-4 retention outperforms an equivalent lift in ad spend by nearly 10x. That's a finding no amount of qualitative debate would have surfaced with this clarity — and it directly redirects roadmap priority toward retention work, which the growth-retention complete guide covers in more depth.

Run this test on every input, not just the obvious ones. Teams often assume acquisition is the bottleneck because it's the most visible spend line. The model frequently says otherwise — and retention or activation improvements, because they compound through every later period, often carry more leverage than a single acquisition channel ever will.

Why Compounding Loops Change the Math

Retention and referral inputs tend to win bottleneck analyses because they're compounding — a retention improvement doesn't just help month 1, it raises the base every subsequent month builds on, while an acquisition spend increase is a one-time addition that doesn't compound unless it also improves referral inputs downstream.

This is the mathematical reason "fix retention before scaling acquisition" is repeated so often in growth circles — it's not folklore, it's what the equation does. Pouring acquisition spend into a leaky retention curve is mathematically equivalent to filling a bucket with a hole in it faster.

Making the Loop Structure Explicit Before You Model It

Before the spreadsheet, you need clarity on which loops actually exist in your product and how they causally connect — a model built on a fuzzy mental picture of the loops just encodes the fuzziness into a spreadsheet with more decimal places.

A Practical Sequence

1. Map the loops — diagram the causal structure (what triggers what) before assigning numbers
2. Assign input-driver values — pull real numbers from analytics for each driver row
3. Chain the loops in a spreadsheet — build the monthly time series described above
4. Run the bottleneck analysis — vary each input, isolate the biggest lever
5. Validate against the actual customer journey — check that your activation and retention assumptions match where real users actually stall

Key Takeaways

• A growth model is a testable equation, not a strategy narrative — every claim gets a cell, a formula, and a source.
• The four core loops — acquisition, activation, retention, referral — each need explicit input-driver rows feeding a single output metric.
• Build it as a monthly time series, not a static snapshot, so compounding effects (especially referral feeding back into acquisition) show up in the projection.
• Bottleneck analysis — varying one input at a time and measuring the output swing — usually reveals that retention or activation, not acquisition spend, is the highest-leverage lever.
• Compounding loops (retention, referral) tend to beat one-time inputs (a single acquisition spend bump) in long-run impact, which is the mathematical basis for "fix retention before scaling acquisition."
• Diagram the causal loop structure before spreadsheeting it — a fuzzy mental model of the loops just produces a precise-looking spreadsheet built on fuzzy assumptions.