The growth model canvas: mapping the engine before you optimise it

Most growth teams are running experiments on a machine they haven't drawn. They optimise the referral rate when the constraint is onboarding completion. They test ad copy when the bottleneck is a 40% D7 churn rate. The experiments produce results. None of them compound. Here is a structured way to map your growth engine first.

The experiment treadmill

In a typical Series A company, the growth team runs eight to twelve experiments per quarter. At the end of the year, they have shipped forty or more changes. The growth rate is roughly where it was twelve months ago.

This is not an execution problem. The experiments are well-designed. The p-values are fine. The problem is that the team is optimising individual cells in a spreadsheet without understanding what the spreadsheet is calculating. They are improving local variables in a system whose structure they have never written down.

Consider what this looks like in practice. A B2B SaaS company has a referral programme that generates about 12% of new signups. The growth team's thesis: if we improve the referral incentive and add an in-product nudge, we can move referral contribution to 20%. They design the experiment. They run it. They get a statistically significant lift of 2.4 percentage points on referral-driven signups.

Six months later, overall ARR growth is unchanged. The referral improvement was real. But the constraint in the system was not referral rate — it was the fact that only 31% of users who signed up were completing onboarding and reaching the activation event that makes someone likely to refer in the first place. Improving the referral programme for 31% of users moves a small number. Improving onboarding completion to 55% would have expanded the entire base eligible to refer, and every other downstream metric along with it.

This is what a growth model canvas is designed to prevent.

What a growth model canvas actually is

A growth model canvas is a one-page document. It has three columns: inputs, multipliers, and outputs. It maps how new users enter the system, what determines whether they stay and expand, and how retained users generate future inputs. It is not a funnel — funnels are linear and don't capture loops. It is not a dashboard — dashboards show you current values, not the structure of the system.

The inputs column captures everything that gets a new user into the system: paid acquisition, organic search, word of mouth, partnership referrals, outbound. Each input has a volume and a cost, and importantly, a quality score — the percentage of inputs from that channel that reach the activation event. Paid traffic that converts at 0.4% to activated users is a structurally different input than organic search that converts at 2.1%, even if the raw volume favours paid.

The multipliers column is where most teams have the least clarity. Multipliers are the rates that determine whether each unit of input produces more than one unit of output over time. The primary multipliers are: activation rate (what fraction of signups reach the moment of first value), retention rate at D7, D30, and D90, expansion rate (what fraction of retained users upgrade or expand seats), and referral coefficient (how many new signups does one retained user generate in a twelve-month window).

The outputs column ties the model to business outcomes: retained users, ARR, and the loop output — the number of new inputs generated by retained users, whether that's referrals, SEO content from user activity, integrations that pull new workspaces in, or API usage that spreads the product through a developer's stack.

How to build yours in an afternoon

Assemble three people: the product lead, the growth lead, and the data analyst. Block three hours. The first sixty minutes are for drawing the structure — not filling in numbers, just agreeing on the shape. What are your actual input channels? What is the activation event? Where do expansion triggers exist in the product? Does your product have a natural referral surface, and if so, where does it fire?

The second sixty minutes are for filling in numbers from your actual data. Every cell should have a number or an explicit "unknown." Unknowns are as valuable as knowns — they tell you where your instrumentation gaps are. If you don't know your referral coefficient, you cannot model the growth impact of improving it, which means you cannot prioritise it correctly against other work.

The third sixty minutes are for the sensitivity analysis. This is the part most teams skip and it is the most important part. Take each multiplier and ask: if this number improved by 10 percentage points, how much does the output change? This calculation identifies your constraint — the multiplier where the same absolute improvement produces the largest output change. That is where you run experiments next.

For most early-stage SaaS products, the sensitivity analysis produces the same answer: activation rate. A 10-point improvement in activation rate has a compound effect on every downstream multiplier. It expands the pool eligible to retain, expand, and refer. It is almost always the highest-leverage lever, and it is almost always underinvested in relative to acquisition.

Reading the loops, not just the rates

The most important insight the canvas surfaces is whether you have a growth loop at all. A growth loop exists when retained users generate inputs — new users entering the system — at a rate that produces compounding over time. The referral coefficient needs to be above approximately 0.15 for word-of-mouth to have a meaningful compound effect at typical SaaS churn rates. If it is 0.03, word of mouth is not a growth loop — it is a rounding error.

Many companies operate without a meaningful growth loop. Their growth is entirely a function of how much they spend on acquisition. This is not fatal, but it means that improving the referral nudge will not produce compound growth — it will produce a one-time step change followed by a return to the baseline growth rate. The canvas makes this structural fact visible, so you can decide whether to build a loop or accept the linear model and invest accordingly in paid acquisition efficiency.

Slack's growth loop is the canonical example: a user invites colleagues, those colleagues activate on the product (because their workspace already has content and context), some fraction of them bring Slack into new channels and new projects, which generates new invitations. The referral coefficient is embedded in the product architecture — you cannot use Slack without inviting someone. Most B2B products do not have this structure, and attempting to bolt a referral programme onto a product with no natural sharing surface produces the 0.03 coefficient described above.

The constraint identification test

Once you have built the canvas and run the sensitivity analysis, you can apply what I call the constraint identification test. Take your current throughput — say, 500 new activated users per month. Now model what happens if you improve each multiplier by 20% independently. The multiplier whose 20% improvement produces the largest increase in 12-month output is your constraint.

An illustrative case: a B2B SaaS canvas showing activation rate at 34%, D30 retention at 58%, expansion rate at 11% and a referral coefficient of 0.08. A 20% improvement in activation rate (to 40.8%) flows through every downstream multiplier, so it lifts 12-month retained users far more than a 20% improvement in expansion rate, which only touches the small base that has already retained. The constraint is obvious — yet teams in this position often run experiments almost exclusively on expansion and referral, because those are the "growth" levers. Onboarding gets treated as a product problem, not a growth problem.

That framing is the enemy of compound growth. Onboarding is a growth problem. It is often the growth problem.

What to do with the canvas after you build it

The canvas is a quarterly document. Your growth model shifts as your product evolves, as your audience shifts, and as your market matures. A canvas built at $500K ARR will not reflect the system at $5M ARR — the input mix changes, new multipliers emerge (enterprise expansion becomes more significant), and the referral dynamics shift as you move upmarket.

Review the canvas at the start of each quarter before setting the experiment roadmap. The review should answer three questions: has the constraint changed? Have any assumptions in the model been invalidated by data? And are there new loops available — new surfaces where retained users could generate inputs — that didn't exist last quarter?

The review does not need to be a long meeting. If the canvas is well-built and the data instrumentation is in place, a ninety-minute review is sufficient. The output is a ranked list of multipliers by leverage, which becomes the experiment prioritisation framework for the quarter.

• 01
  Draw the structure before you fill in the numbers. Agreeing on the shape of the engine — what the inputs, multipliers, and loops actually are — is more valuable than precise metrics for a canvas you haven't built yet. Structural disagreements between product and growth are the most common cause of misaligned experiment roadmaps.
• 02
  Mark unknowns explicitly. A cell that says "unknown" is more honest and more useful than a cell that has a number someone invented. Unknowns become your instrumentation roadmap. If you can't measure a multiplier, you can't optimise it with confidence.
• 03
  Run the sensitivity analysis before the experiment roadmap. The sensitivity analysis tells you which multiplier improvement produces the largest system-wide output change. This is the only defensible basis for prioritising one experiment over another.
• 04
  Distinguish loops from linear inputs. A referral coefficient of 0.03 is not a loop — it is noise. If your canvas shows no loop above approximately 0.10, you do not have a word-of-mouth growth engine. Build one or accept the linear model and plan your acquisition spend accordingly.
• 05
  Treat onboarding as a growth problem. Activation rate is almost always the highest-leverage multiplier in early-stage SaaS. Its improvement compounds across retention, expansion, and referral. Any growth team that does not own the onboarding funnel is optimising a subset of their actual leverage.
• 06
  Review quarterly, not annually. Your growth model shifts. The constraint at $1M ARR is rarely the constraint at $5M ARR. A canvas built once and never updated becomes a false map — it shows you where the roads were, not where they are.

Frequently asked questions.