Skip to content

04 · Growth Product Management

Growth product management is not marketing with a Jira board, and it is not "the team that does the signup page". It is the discipline of treating the business as a system with measurable inputs, then systematically improving the inputs that matter most per unit of effort.

The distinction that makes it a separate job: a core PM asks what should we build for this customer? A growth PM asks where in the system is a unit of engineering worth the most, and how do we find out cheaply? The two need each other. A growth team optimising signup for a product nobody keeps using is polishing a leaking bucket, and a core team building a beloved product that 3% of visitors ever discover is doing charity.

This module covers the growth model, loops versus funnels, activation, sizing an idea before you build it, and running a portfolio of bets where most of them fail.

Core PM vs growth PM

Core product Growth product
Question Does this solve the problem? Does the system convert better?
Unit of work Feature Experiment
Typical horizon 1–2 quarters 1–3 weeks
Success Retention, satisfaction, job done Conversion rate, LTV, payback
Failure is Expensive and slow to detect Cheap and expected — 60–80% of tests lose or are flat
Main risk Building the wrong thing Local maxima; short-term wins that cost retention

The growth model

Write your business as one equation, then argue about the terms. Everything downstream — roadmap, OKRs, team structure — gets easier.

For a self-serve B2B SaaS like ListUp:

New MRR = Visitors × (Visitor→Trial) × (Trial→Paid) × ARPA
MRR(t+1) = MRR(t) × (1 − Churn) + New MRR + Expansion MRR

ListUp's current model, measured over the last 8 weeks:

Term Value Note
Monthly site visitors 16,000 Organic 61%, direct 24%, paid 15%
Visitor → trial 2.80% 448 trials/month
Trial → activated (2 channels connected + 1 change published, ≤7 days) 41.1% 184 accounts
Trial → paid 22.0% 98.6 new paying accounts/month
ARPA $80.78 Blended across tiers
New MRR/month $7,962 448 × 22.0% × $80.78
Monthly logo churn 2.66% 47.3 accounts lost/month
Net accounts/month +51.2 +2.88%/month, which compounds to 40.5%/year

That last line is the sanity check that makes the whole model trustworthy: a 2.88% monthly net add compounds to 40.5% annual account growth, which matches the 41% ARR growth reported in Module 1. When your bottom-up funnel reproduces your top-down growth rate, you can start making decisions with it. When it doesn't, find out why before you plan anything.

Funnels and loops

A funnel is linear: you pour in at the top and some fraction comes out. A loop feeds its own input, so the output becomes next period's input.

Funnel Loop
Shape Visitors → trial → paid Paid user → produces something → attracts new user
Growth behaviour Linear in spend Compounds, with a lag
Effort to improve Bounded; conversion rates plateau High up front, cheap later
ListUp example Pricing page → trial Agency manages 8 sellers → sellers see ListUp → some sign up direct
Loop type Mechanism Works when
Viral / invitation Users invite collaborators The product is multiplayer
Content Usage produces indexable pages Output is public and searchable
Paid Revenue funds acquisition Payback is shorter than your cash cycle
Sales-assisted Customers become references Deals are considered and social proof matters
Ecosystem Partners bring their customers You have a platform (Module 3)

ListUp is single-player for the seller, so no viral loop exists. The available loops are the agency loop (an agency onboards its clients) and the paid loop. Naming this honestly is more useful than running an invite experiment that has no mechanism behind it. Most products have one real loop; pretending you have four wastes a year.

Activation is where the leverage usually is

Activation is the moment a new user gets the value they came for. It is the highest-leverage stage in almost every SaaS funnel because it sits between two expensive things — acquisition you paid for and retention you need.

Finding it is an analysis, not a workshop:

Step Method ListUp result
1. Candidate actions List everything a new account can do in week 1 14 actions
2. Correlate with month-3 retention For each action, retention of those who did it vs didn't Connecting a second channel: 68% vs 24%
3. Find the threshold Where does the curve bend? 2 channels; the 3rd adds little
4. Add a time bound When does it stop predicting? Within 7 days
5. Validate causally Hold-out or onboarding change Guided connect flow test, below
Definition 2 channels connected and 1 change published within 7 days

Step 5 is the one teams skip, and skipping it is why so many "aha moments" are just descriptions of the users who were going to stick anyway. The correlation tells you where to look. Only an experiment tells you whether pushing people through it changes anything.

Sizing an idea before you build it

Every growth idea should be converted to money before it is scheduled. The arithmetic is short and it kills a lot of debate.

Method: re-run the growth model with one term changed, take the difference in new MRR per month, then compound 12 monthly cohorts forward with churn at 2.66%/month to get the MRR uplift at month 12, and multiply by 12 for ARR.

Experiment Term changed ΔNew MRR/mo ΔMRR at month 12 ΔARR
Guided second-channel connect Trial→paid 22.0% → 24.5% $905 $9,401 $112,815
Pricing-page rework Visitor→trial 2.80% → 3.20% $1,137 $11,819 $141,824
Agency referral loop Visitors 16,000 → 18,000 $995 $10,341 $124,096
Onboarding upsell checklist ARPA $80.78 → $84.98 $414 $4,301 $51,617
At-risk churn playbook Churn 2.66% → 2.40% $4,539 $54,466

Then discount by your honest probability of success and divide by cost:

Experiment ΔARR if it works P(success) Eng-weeks Expected ARR EV per eng-week
Pricing-page rework $141,824 30% 3 $42,547 $14,182
Onboarding upsell checklist $51,617 50% 2 $25,808 $12,904
Guided second-channel connect $112,815 45% 4 $50,767 $12,692
Agency referral loop $124,096 20% 6 $24,819 $4,137
At-risk churn playbook $54,466 35% 5 $19,063 $3,813

Read the table carefully, because the naive reading is wrong. By raw ΔARR the pricing-page rework wins. By EV per engineer-week it still wins, narrowly — but the three leaders are within 12% of each other, which is well inside the error bars on a probability you guessed. When options are that close, pick on a second criterion: the second-channel connect experiment also produces information about activation that the other two don't, so it goes first.

The referral loop ranks last on this table and may still be worth doing, because loops compound and one-off conversion gains don't. Expected value per week is a ranking tool for the quarter, not a philosophy.

Worked example — ListUp's activation experiment

Hypothesis. New accounts that connect a second channel in week 1 retain at 68% vs 24%. Only 41.1% get there. A guided flow that asks for the second channel during setup — instead of leaving it to a dashboard tile — will raise trial→paid from 22.0% to 24.5%.

Design. 50/50 split at trial signup. Primary metric: trial→paid within 14 days of trial start. Secondary: activation rate, time to second channel. Guardrails: day-30 retention of paying accounts, support tickets per new account.

Sample size. Detecting 22.0% → 24.5% (a 2.5-point absolute lift, 11.4% relative) at 80% power and 95% confidence needs roughly 4,500 trials per arm. A 50/50 split gives 224 trials per arm per month, so that is 20 months — the experiment is not runnable as designed.

This is the single most common growth-team collision with reality, and there are only four honest responses:

Option Effect Chosen?
Accept a bigger MDE Test for +5pts instead; ~1,160/arm, ~5.2 months No — a 5pt lift is implausible
Move up the funnel Use activation as the primary metric: 41.1% → 50%, ~490/arm, ~2.2 months Yes
Lower confidence to 90%, power to 80% Cuts sample ~20%; raises false-positive risk Partially
Ship it on judgement, monitor No causal read, fast No

Decision: run on activation as the primary metric with trial→paid as a directional secondary, and pre-register that a positive activation result plus a non-negative trial→paid trend ships. Write that rule down before the data arrives, because the temptation to reinterpret a borderline secondary metric is enormous.

Result after 10 weeks (515 and 508 trials per arm):

Metric Control Variant Read
Activation (2 channels + 1 publish ≤7 days) 41.2% 49.6% +8.4pts, significant
Median time to 2nd channel 4d 6h 1d 2h Strong supporting signal
Trial→paid (14 days) 21.9% 23.8% +1.9pts, not significant — directionally right
Day-30 retention of new payers 91.4% 92.0% Guardrail held
Support tickets per new account 0.31 0.29 Guardrail held

Shipped, per the pre-registered rule. The important discipline is what came next: trial→paid was tracked as a monitored metric for two further quarters. It settled at 23.4%, below the 24.5% assumption, so the modelled $112,815 was revised down to about $86,000. Going back and correcting your own forecast — out loud — is what makes the next forecast believed.

Traps

Trap Symptom Fix
Optimising a leaky bucket Conversion up, retention flat, revenue flat Fix retention before acquisition; churn compounds against you
Local maxima Twelve wins, no change in the growth curve Reserve 20% of capacity for structurally different bets
Dark patterns Signup up, day-30 retention down Make a retention guardrail mandatory on every growth test
Vanity loops An invite feature with no multiplayer use case Name your one real loop and invest there
Underpowered tests everywhere Everything is "trending positive" Compute sample size first; if it's not runnable, change the metric
Growth team owns growth Core team feels unaccountable for retention Growth owns the system; every team owns its own retention

How It Actually Works

Growth PM work runs on a specific mathematical model: the AARRR funnel (Acquisition, Activation, Retention, Referral, Revenue) treats a business as a system of conversion rates chained together, and the compounding nature of that chain (each stage's output is the next stage's input) is exactly why fixing retention almost always beats fixing acquisition for the same effort — a leaky-bucket business pouring more users into a funnel with poor retention loses most of them anyway, so retention improvements compound forward through every future acquisition cohort, while acquisition improvements only affect the current cohort once. Activation is the highest-leverage stage specifically because of a measurable phenomenon: users who reach a defined "aha moment" within their first session show dramatically different long-term retention curves than those who don't (this is the empirical basis behind metrics like Facebook's famous "7 friends in 10 days"), meaning activation rate functions as a leading indicator that predicts retention weeks before it's observable directly. Growth experimentation velocity (running many small tests per week) works because of the same statistical logic behind A/B testing sample sizes: most individual growth ideas fail (a well-documented ~70-90% failure rate across published growth case studies), so the expected value of a growth program comes from the volume and speed of testing, not the brilliance of any single idea — a slow-moving growth team with "better" ideas will still underperform a fast-moving team running many mediocre ones, purely from the statistics of iteration count.

Exercise

Build the growth model for a product you know, then plan a quarter from it.

  1. Write the growth equation with your product's actual terms, and fill in every number from data — not estimates. Include churn and expansion.
  2. Reconcile it. Compound your monthly net add over 12 months and check it matches your reported annual growth. If it doesn't, find the discrepancy before going further; that is the most valuable hour in this exercise.
  3. Identify your activation moment with the five-step method. Report the retention split for your top candidate action, the threshold, and the time bound — and state honestly whether it has ever been validated causally.
  4. Name your one real loop, and one loop people talk about that your product does not actually have.
  5. Size five ideas. For each, change one term, compute ΔNew MRR/month, compound 12 cohorts with your churn rate, and report ΔARR.
  6. Rank by expected value per engineer-week, with your honest probability of success. Then say which ranking you would override, and why.
  7. Sample-size your top experiment. If it is not runnable in under a quarter, pick one of the four responses and justify it in writing.
  8. Write the pre-registered ship rule before you run anything: which metric, which threshold, which guardrails, and what happens on a borderline result.
  9. Schedule the forecast correction. Put a date 6 months out to compare your modelled ΔARR against what actually happened, and commit to sharing the difference.