Brian Balfour demonstrates the core problem in workshops by asking five people at the same company to independently whiteboard their answer to that question. The result is always the same: everyone gives a different and incomplete answer. He describes this as "a BIG problem," and it is not a communication failure but a structural one. The team lacks a shared, systemic understanding of the mechanics driving their growth.
The AARRR funnel answered the question of what to measure. The Growth Model Framework answers the question of how those measurements connect into a system that either compounds or does not. The distinction matters because a funnel can be fully instrumented, with all metrics improving, and the business still growing linearly, requiring more money, more people, and more tactics every quarter just to maintain the same rate.
A self-reinforcing mechanism where the output of one user cycle becomes the input for acquiring the next. Distinguished from a funnel, which is linear (more in at the top produces more out at the bottom, indefinitely requiring fresh inputs), and from a tactic, which is a one-time action with no feedback effect.
The complete system linking acquisition channels, retention dynamics, growth loops, and monetisation mechanics into a single equation with measurable inputs and calculable outputs. It has two forms: a qualitative loop diagram (boxes and arrows) and a quantitative spreadsheet model (inputs, conversion rates, scenario analysis).
The Growth Model Framework is community-developed, with Brian Balfour as the primary architect. Its intellectual lineage has three phases. The first is AARRR Pirate Metrics, created by Dave McClure and presented at Ignite Seattle on August 8, 2007. AARRR established growth as a five-stage measurement discipline but represented each stage as a funnel step with no feedback loops between them.
The second phase is the Four Fits Framework, developed by Balfour through his work at HubSpot (VP of Growth, January 2014 to 2016) and published as a seven-part essay series. The series established that sustainable growth requires alignment across four fits: Market-Product Fit, Product-Channel Fit, Channel-Model Fit, and Model-Market Fit.
The third and catalytic phase is the "Growth Loops are the New Funnels" essay, published at reforge.com/blog/growth-loops on July 31, 2018. Co-authored by Brian Balfour, Casey Winters, Kevin Kwok, and Andrew Chen, this essay formally named the shift from linear funnels to self-reinforcing loops and introduced the growth loop as the primary unit of analysis.
A growth model is built in sequence. The qualitative form comes first, then the quantitative form is layered on top.
Qualitative (visual loop diagram). A map showing how users flow through the system, which loops are active, and how outputs feed back as inputs for new users. Balfour: "The first thing we do is sit down and map it all out with boxes and arrows." Each loop has three components: what (the action connecting input and output), who (the people involved), and why (the deep motivation that drives the action).
Quantitative (spreadsheet model). A spreadsheet with a long list of input assumptions (conversion rates, channel volumes, retention curves, viral coefficients), acquisition channels modelled over time, retention dynamics as cohort-based decay curves, growth loops layered on top showing compounding, and scenario comparisons as the core output. Per Dan Hockenmaier: "The key output is the difference between the scenarios, not the absolute values they produce."
Growth models are about inputs, not outputs. Growth rate is an output, calculated as a result of other dynamics. If a team puts growth rate as an input, the model has no analytical or predictive power.
The growth equation is the mathematical expression of a company's specific growth model. No universal formula exists, and each business must derive its own. Three foundational formulations from primary sources:
Chamath Palihapitiya and Andy Johns (Facebook/Wealthfront): Growth = A (Top of Funnel) x B (Magic Moment) x C (Core Product Value). Where A is traffic capture and conversion to meaningful usage, B is the emotional "aha moment" of experiencing core value, and C is market size and product-market fit.
Brian Balfour (Reforge, 2025 formulation): (Acquisition + Retention + Monetisation) x Defensibility
Brian Balfour (Model-Market Fit viability test): ARPU x Total Customers in Market x % You Can Capture ≥ $100M
"Invites sent per user x invite-to-signup conversion rate"
Users invite others and each new user repeats the cycle. The output of one user cycle becomes the input for acquiring the next.
Dropbox (storage for referrals), Slack (invite your team), Zoom (meeting participants sign up)
Shortening cycle time has far bigger impact than changing viral coefficient. A K-factor of 1.5 cycling monthly produces roughly the same growth as a K-factor of 1.1 cycling weekly.
Balfour et al. (2018) make three arguments against using AARRR funnels as the primary growth model: funnels create strategic silos, functional silos, and linear growth. Growth loops combine product, channels, and monetisation into one system and generate compound output. AARRR metrics remain useful as measurements of what happens within each loop step, and the framework critique is architectural, not metric-level.
A growth model only compounds when four fits are aligned. Each fit is a prerequisite for the next.
Market-Product Fit: The product solves a real problem for a real market. Balfour frames it as the specific market determining which product properties are required.
Product-Channel Fit: "Products are built to fit with channels. Channels do not mould to products." Viral distribution requires a broad value proposition and fast time-to-value. Enterprise sales requires high ACV and a procurement-friendly buying process.
Channel-Model Fit: The ARPU-CAC relationship must work. At $25/month, HubSpot Sales sat in the "Channel-Model Fit danger zone," where ARPU was too low for content marketing and inside sales but too high for viral and paid acquisition.
Model-Market Fit: The business model must be large enough to reach the target revenue threshold. ARPU x Total Customers in Market x % You Can Capture ≥ $100M.
Confirm four conditions: product-market fit exists (40% Sean Ellis benchmark), baseline data flows, a North Star Metric is defined, and five people can attempt to whiteboard 'how does our product grow?' to expose the gap.
Document every step from first awareness to retained and monetised user using boxes and arrows. Identify all active and potential growth loops. Classify each loop by type. Name what, who, and why for each loop.
Start with Growth = A x B x C. Break each component into business-specific sub-variables. Identify the primary loops. Name the North Star Metric at the top of the equation.
Gather current values for every variable. Sources: product analytics, CRM, billing data, paid platform dashboards. Use best estimates where actuals are unavailable. Per Hockenmaier: do not wait for perfect data.
Model acquisition channels over time. Apply retention dynamics as cohort-based decay curves. Layer on growth loops by applying viral coefficients and cycle times. Calculate growth rate as an output, not an input.
Run the model against the past 3 to 6 months of actual data. Per Hockenmaier: 'The model does not need to be perfect, it needs to be defensible.' Where it diverges, identify which assumption is wrong.
Compare scenarios by toggling one input at a time. Per Hockenmaier: 'If a growth team could focus on lifting activation rates by 10%, or improving paid conversion by 25%, or increasing referrals by 5%, which drives more revenue?' The model answers this quantitatively.
Select one to three highest-leverage inputs the team can directly influence. Translate model insights into specific experiments. The model should produce a ranked argument for the top three, not a list of 40 priorities.
Update monthly with actual data. Per Hockenmaier: assign ownership to someone 'one step removed from operating, like analytics or strategic finance.' Recalibrate assumptions quarterly. Model loops to predict when they will plateau.
Five people on your team give different answers to "how does our product grow?" Your primary growth driver is paid acquisition with no compounding loops reinforcing it. CAC is rising quarter over quarter with no change in product or competition. Teams are running experiments without a shared model of which levers move the output metric. You have improved a conversion metric and it did not produce the expected downstream effect.
Startup (S4, Critical). The essential application. Product-market fit exists and the team is transitioning from finding what works to scaling it. The growth model answers which acquisition loops are self-sustaining, which retention mechanics compound, and which monetisation structure preserves the viral coefficient.
Growth Stage (S5, Critical). Maximum value. The model becomes a planning system. Multiple loops are spinning and the model shows their relative efficiency, interaction effects, and marginal returns.
Scale Stage (S6, Critical). New S-curves must be sequenced before existing loops plateau. Winters: "Companies must invest in new loops before current ones stop driving the growth the company needs."
Early Startup (S3, Useful, conceptual only). A qualitative loop diagram is valuable for designing the product with growth compounding from the start. Do not build a quantitative model without real data.
Casey Winters joined as growth lead in late 2013. Pinterest had roughly 40M MAU and had stalled. The original growth engine, built on Facebook's Open Graph automatic sharing, had collapsed when Facebook removed default auto-sharing. The friend-based viral loop was broken.
How They Applied ItWinters mapped the actual growth model rather than assuming the existing mechanism still worked. The mapping revealed that users acquired through Google search were being shown content from their Facebook friends during onboarding, which had nothing to do with their search intent. Switching new SEO-acquired users from friend-based to topic-based recommendations immediately doubled activation rates and produced a 5x improvement in conversion to signup. Two self-reinforcing loops were rebuilt: acquisition (pins indexed by Google attracting searchers) and engagement (personalised content based on search intent).
Jorge Mazal joined as Head of Product in late 2017. DAU was stagnating despite hundreds of millions of registered users. The growth team was generating experiment ideas but struggled to identify which lever would produce meaningful impact.
How They Applied ItMazal built a Markov chain model decomposing DAU into seven mutually exclusive user states. Running simulations pulling each lever by two percentage points per month, the model revealed that improving Current User Retention Rate (CURR) had roughly 10x the impact on DAU as improving new user acquisition. This was counterintuitive. A dedicated Retention Team was formed with CURR as its North Star Metric. Leaderboards increased learning time by 17%, tripled highly engaged learners, and improved Day 1 and Day 7 retention.
Brian Balfour joined HubSpot in January 2014 to build a new $100M product line using a freemium, touchless model. Starting conditions: roughly seven people, approximately 2,000 weekly active users, and a few thousand dollars in monthly recurring revenue.
How They Applied ItBalfour applied the Four Fits framework to design the growth model before building the growth machine. The $25/month pricing tier landed in the 'Channel-Model Fit danger zone,' too expensive for viral and paid channels and too cheap for content marketing and inside sales. He eliminated the $25 tier, moved features to $50, reduced reliance on virality, and rebuilt around content marketing and inside sales. The customer profile shifted from 'mice' (high volume, low ARPU) to 'moose' (mid-market, higher ARPU).
Teams build a list of channels with budget allocations and call it a growth model. No equation, no linked variables, no scenario analysis capability.
Terms are used interchangeably in popular growth writing. Channel plans are familiar and fast to produce. Mathematical models require data skills most marketing teams lack.
Apply one test: can you plug in a number and see what happens downstream? If no, it is not a growth model. Start with the growth equation. Decompose into business-specific metrics.
The model assumes 15% month-on-month growth as a starting number. It produces projections that look plausible. They are useless.
Investors ask for growth projections and teams back-fill the mechanics. Assuming a rate is faster than decomposing one.
Follow the golden rule: growth rate is an output, calculated from component inputs. Break it down into acquisition channels, retention dynamics, viral coefficient, and cycle time. Build bottom-up.
Teams add share buttons, referral pages, or invite prompts and call them growth loops. Per Reforge: 'They build a product, launch it, see growth plateau, then try to add sharing. That is bolted-on growth.'
Loop thinking is prevalent in growth writing. Teams want the compounding without the prerequisite: core user action must naturally produce an output that feeds new acquisition.
For each proposed loop: does one user's action actually become input for new acquisition? Is the mechanism embedded in the core product experience or appended to it? If fewer than 1 in 100 users completes the loop, the mechanism is not self-reinforcing.
Teams obsess over K-factor while ignoring how long the loop takes. A K-factor of 1.5 that cycles monthly produces roughly the same growth as a K-factor of 1.1 that cycles weekly.
K-factor is the headline metric for viral growth; cycle time is harder to measure and rarely reported in case studies.
Measure both coefficient and cycle time. Track the full loop from invite to activation. Map timing per step. Reducing a 30-day cycle to 7 days at the same K-factor produces dramatically faster compounding.
The model is built once, presented to leadership, praised, and filed. Per Hockenmaier: 'Growth models usually do not get used. Often useful for the builder but rarely adopted broadly.' Assumptions stale within 90 days.
Building the model is significant effort and maintenance competes with running experiments. Without a standing owner and regular update cadence, the model becomes an artefact.
Assign a clear owner, ideally from analytics or strategic finance, one step removed from operations. Integrate into monthly planning. Update with actuals monthly. The model must be consulted before headcount or budget decisions.
Teams copy the growth model of viral consumer products onto enterprise B2B, high-ACV, or compliance-sensitive products. The loop never compounds because the product's properties make virality structurally impossible.
Growth loop thinking is heavily sourced from consumer and PLG case studies. Teams copy the form (viral loop) without checking the fit.
Apply the Four Fits test before designing any loop. Use the ARPU-CAC spectrum: enterprise B2B requires high-touch sales loops; broad B2B SaaS supports collaboration and product loops; consumer supports viral and content loops.
When a company discovers through growth model analysis that its primary loop cycles every 30 days instead of every 7, and that shortening that cycle would produce 4x the growth, that finding does not stay inside marketing. It changes the product roadmap, the pricing tier structure, the financial projections, and the engineering sprint priorities.
Pick and sequence up to six frameworks for a real broken-funnel scenario. Scored on inclusion and order.
Vote on which framework applies first. Community split bars animate in, then the dependency-based verdict reveals.
Flag the broken steps in a six-step strategy. Results show correct flags, false positives, and reasoning.