Why growth teams count in stages
Most teams know how many people visited their site last month and how much money came in. The interesting part sits in between. Of the people who arrive, how many sign up? Of those who sign up, how many reach the first moment of real value? How many are still around a month later, how many bring a friend, and how many pay? If you only watch the two ends, a drop in revenue tells you that something is wrong but not where. A stage-by-stage view tells you where to look first.
The most widely used way to organise those stages is the AARRR framework, nicknamed the pirate metrics because the initials sound like a pirate's growl. The five letters stand for Acquisition, Activation, Retention, Referral and Revenue. This free lesson turns the framework into numbers you can calculate in a spreadsheet: one conversion rate per stage, for every acquisition channel, plus the overall rate from first visit to payment.
The example is Kestrelwick, a fictional household budgeting app with a free plan and a paid plan called Kestrelwick Plus. All channels, people and numbers in this lesson are invented for learning, and any benchmark mentioned is illustrative. You will use the same eight channels and the same March cohort as in the lab above.
The five stages and what each one means
Every company defines the stages a little differently, and that is fine, as long as the definitions are written down and used the same way every month. For the lab, Kestrelwick uses these:
- Acquisition: people arrive (visitors) and create an account (sign-ups). The channel they came through is recorded, which is what makes a comparison between channels possible.
- Activation: a new user gets the first real benefit. For Kestrelwick that is finishing a first budget. Activation is the bridge between curiosity and a habit, and it is usually the cheapest stage to improve.
- Retention: the user comes back. Here it means still being active in week 4 after sign-up. A product with weak retention fills a leaking bucket however much traffic it buys.
- Referral: the user brings someone else, for example by inviting a friend who then joins. Referral is both a growth source and a sign that users value the product.
- Revenue: the user pays. It comes last in the acronym, but it is the stage that keeps the business alive, so it is tracked from day one.
The core idea: one conversion rate per step
A funnel table holds one count per stage, side by side for each channel. The basic tool is the step conversion rate: the number of users at a stage divided by the number at the stage before it. In the lab the five rates are:
- Sign-up rate = sign-ups ÷ visitors
- Activation rate = activated users ÷ sign-ups
- Week-4 retention = users still active in week 4 ÷ activated users
- Referral rate = referrers ÷ users retained in week 4
- Visit-to-paid rate = paying users ÷ visitors (the whole funnel in one number)
Each rate answers one question about one stage. That makes the rates easy to own: the marketing team looks at the sign-up rate, the product team at activation, the lifecycle team at retention. The rule that makes the numbers comparable is simple: always divide by the stage directly before. A rate that skips a stage mixes two questions in one number and hides where the loss happens.
The overall rate has a neat relationship to the step rates. If every stage is a subset of the one before it, the visit-to-paid rate equals the product of the step rates in between. When one step improves by 20 %, the overall rate improves by 20 % too, provided nothing else changes. This is why growth teams look for the biggest leak first: the same relative effort moves the overall result by the same amount at every stage, but the leaks differ a lot in how easy they are to fix.
A fully worked example: Kestrelwick, March cohort
Here are the counts for the eight channels. Every user in the cohort is counted once, in the channel through which the account was created.
| Channel | Visitors | Sign-ups | Activated | Retained week 4 | Referrers | Paying |
|---|---|---|---|---|---|---|
| Organic search | 18,400 | 1,656 | 745 | 358 | 54 | 61 |
| App store search | 9,600 | 1,920 | 1,056 | 549 | 99 | 104 |
| Paid social | 22,500 | 1,575 | 520 | 177 | 16 | 21 |
| Paid search | 7,800 | 858 | 412 | 190 | 25 | 36 |
| Creator partnerships | 6,400 | 832 | 449 | 220 | 42 | 37 |
| Referral invites | 3,100 | 1,085 | 705 | 409 | 110 | 86 |
| Newsletter swaps | 4,200 | 462 | 236 | 118 | 17 | 22 |
| Community forums | 2,000 | 240 | 139 | 78 | 19 | 15 |
| Total | 74,000 | 8,628 | 4,262 | 2,099 | 382 | 382 |
In total 74,000 visitors produced 8,628 sign-ups, 4,262 activated users, 2,099 users who were still active in week 4, 382 referrers and 382 paying users. The lab calculates the five rates for each row and for the total row. Here is the result:
| Channel | Sign-up rate | Activation rate | Week-4 retention | Referral rate | Visit-to-paid |
|---|---|---|---|---|---|
| Organic search | 9.0 % | 45.0 % | 48.1 % | 15.1 % | 0.33 % |
| App store search | 20.0 % | 55.0 % | 52.0 % | 18.0 % | 1.08 % |
| Paid social | 7.0 % | 33.0 % | 34.0 % | 9.0 % | 0.09 % |
| Paid search | 11.0 % | 48.0 % | 46.1 % | 13.2 % | 0.46 % |
| Creator partnerships | 13.0 % | 54.0 % | 49.0 % | 19.1 % | 0.58 % |
| Referral invites | 35.0 % | 65.0 % | 58.0 % | 26.9 % | 2.77 % |
| Newsletter swaps | 11.0 % | 51.1 % | 50.0 % | 14.4 % | 0.52 % |
| Community forums | 12.0 % | 57.9 % | 56.1 % | 24.4 % | 0.75 % |
| Total | 11.7 % | 49.4 % | 49.2 % | 18.2 % | 0.52 % |
Look at Figure 1. Referral invites converts 2.77 % of its visitors into paying users, while Paid social converts only 0.09 %. The overall rate is 0.52 %, shown as the dashed line. Only 5 of the eight channels are at or above it, because the large channels with lower rates pull the total down.
Figure 2 shows the same data as a flow. Of 74,000 visitors, 11.7 % sign up. Of the sign-ups, 49.4 % activate. Of the activated users, 49.2 % are still active in week 4, and 18.2 % of those have invited somebody. Two stages lose about half of the users: the step from sign-up to activation and the step from activation to week 4. Those are the first places where a growth team would look for experiments.
The formulas, step by step
All five rates are divisions, so one formula per column is enough. The formulas below are written for row 4 and copied down to row 12, the total row, so the same formula also gives the overall rates.
- Sign-up rate (H4):
=C4/B4. Sign-ups (column C) divided by visitors (column B). Copy it down to H12. - Activation rate (I4):
=D4/C4. Activated users divided by sign-ups. Notice that the denominator moves one column to the left compared with the first formula: always the previous stage. - Week-4 retention (J4):
=E4/D4. Users retained in week 4 divided by activated users. - Referral rate (K4):
=F4/E4. Referrers divided by users retained in week 4. - Visit-to-paid rate (L4):
=G4/B4. Paying users divided by visitors, the first and the last stage. - Format the five columns as percentages. For the total row, use the totals in row 12 (
=SUM(B4:B11)and so on); do not calculate it as the average of the column above.
Because the references are relative and the layout is regular, you can write each formula once, press Enter and drag the fill handle down. In the lab the Coach checks one cell per task, so choose a row from the task text, compare your result with the check and then copy the formula to the other rows.
Why the total row must use totals, not an average of rates
A frequent mistake is to fill the total row with =AVERAGE(H4:H11). That gives every channel the same weight, whether it brought 2,000 visitors or 22,500. The correct total rate divides the total of the numerator by the total of the denominator, for example =C12/B12.
The sign-up rates in the lab show the difference clearly. The plain average of the eight sign-up rates is 14.75 %, but the true overall sign-up rate is 8,628 ÷ 74,000 = 11.66 %. The gap arises because two channels with high sign-up rates (Referral invites at 35 % and App store search at 20 %) bring only 12,700 visitors together, against 22,500 for Paid social, yet they count twice as much in the average.
How to read the result: volume is not quality
The raw counts and the rates tell different stories. Paid social is the biggest source of visitors in the cohort (22,500 visitors, 30.4 % of all of them), but it delivers only 21 of the 382 paying users, 5.5 %. Its sign-up rate of 7.0 % and its visit-to-paid rate of 0.09 % are the lowest of the cohort. Organic search brings 18,400 visitors and 61 paying users, 16.0 % of the total.
At the other end, Referral invites is small (3,100 visitors) but efficient: 35 % of its visitors sign up, 65 % of the sign-ups activate and 2.77 % of the visitors end up paying. People who arrive on a friend's invitation already trust the product, which is why this channel performs so well. It cannot simply be bought in larger quantities, though, because the visitors depend on the existing users.
Two cautions apply. First, channels attract different audiences. Cold paid traffic is expected to convert worse than visitors from a friend's invitation; a low rate is not automatically a reason to cut a channel. What matters is the cost per activated user and per paying user, which the next lessons add. Second, small channels have noisy rates: a rate based on a few dozen users can change a lot from month to month. Do not make big decisions from the referral rate of a channel with 17 referrers.
Finding the biggest leak
For each channel, look for the step with the lowest rate relative to a sensible expectation. For Paid social the weak points are the sign-up rate (7.0 %) and the week-4 retention (34.0 %). The overall funnel loses the most users between sign-up and activation (50.6 % of the sign-ups never activate) and between activation and week 4 (50.8 % of the activated users are gone).
Comparing the same step across channels is as informative as comparing the steps of one channel. If one channel activates 65 % of its sign-ups and another 33 %, the app is the same for both; the difference lies in who arrives and what they were promised before arriving. That points to the landing page, the ad copy or the onboarding for that channel, which are precisely the things an experiment can change.
Common mistakes and how to spot them
- Dividing by visitors in every column. The result is a cumulative rate, not a step rate. If the activation rate of a channel looks tiny, check that column I divides by sign-ups and not by visitors.
- Upside-down divisions.
=B4/C4gives visitors per sign-up, a number above 1. A rate between 0 and 1 means the smaller number is on top. - Averaging rates in the total row. Use ratios of totals, as shown above.
- Mixing time windows. If the visitors are from March and the paying users include people who signed up in earlier months, the rates are meaningless. A cohort table follows the same people through all stages.
- Changing definitions. If activation meant a finished budget last quarter and a linked bank account this quarter, the two quarters cannot be compared.
- Reading a rate without its count. A 35 % rate on 3,100 visitors and a 35 % rate on 30 visitors are not equally reliable.
Quick reference: formulas in this lesson
| Rate | Formula | Example in the lab |
|---|---|---|
| Sign-up rate | sign-ups ÷ visitors | =C4/B4 |
| Activation rate | activated ÷ sign-ups | =D4/C4 |
| Week-4 retention | retained ÷ activated | =E4/D4 |
| Referral rate | referrers ÷ retained | =F4/E4 |
| Visit-to-paid rate | paying ÷ visitors | =G4/B4 |
| Total row | ratio of the column totals | =C12/B12 |
Real-world uses
- A monthly growth review. Paste the stage counts per channel from your analytics tool and your product database into the sheet. The five rates show within minutes which stage changed since last month.
- Choosing what to test. The step with the largest loss, and the channel with the most users at that step, is the natural place for the next experiment. Later lessons turn this into a scored backlog.
- Setting owners and targets. Each rate belongs to one team. A target such as activation above 50 % for the paid channels is easier to follow than a vague goal for the whole funnel.
- Explaining the business. A funnel with real counts and rates is a concise way to tell an investor, a manager or a new colleague where the users come from and where they are lost.
What you will build in the rest of this course
Lesson 1 is free. Lessons 2 to 20 make up the paid course. Each has its own live lab with the same fictional company, Kestrelwick, and builds on the funnel you calculated today:
- The North-Star Metric: Input Metrics, Week-over-Week Growth, the Compound Rate and Weeks to a Goal. You build Kestrelwick's north-star metric from two input metrics, measure its weekly and compound growth and estimate the weeks needed to reach a goal.
- Acquisition Quality: Cost per Sign-Up, Cost per Activated User, Channel Shares and Ranking. You compare nine acquisition channels by cost per sign-up and cost per activated user, calculate each channel's share of activated users, rank the channels and chart the result.
- Activation: Defining the Aha Moment with IF and AND, the Activation Rate and the Retention Gap. You apply a three-part activation definition to 16 new households, calculate the activation rate with COUNTIFS and compare the week-4 retention of activated and non-activated households.
- Retention: Cohort Tables, Mixed References, Weighted Retention per Month and Retention Curves. You turn a triangle of cohort counts into retention rates with a mixed reference, calculate the weighted retention per month, estimate a young cohort's month-6 users and chart the curves.
- Referral: Share of Inviters, Invite Conversion, the K-Factor and the Amplification Factor. You break Kestrelwick's referral loop into the share of households that invite and the invite conversion, calculate the K-factor per month and overall, and turn it into an amplification factor.
- Revenue: Paid Share, ARPPU, ARPU, Subscriber Churn and a Simple Lifetime Value. You calculate the paid share, ARPPU and ARPU per month for Kestrelwick's subscription, the subscriber churn over ten months and a steady-state lifetime value per subscriber.
- Sizing the Leak: Multiplying Step Rates, Activation Targets with XLOOKUP and Extra Paying Users. You multiply the step rates of ten segments into paying users, look up an activation target per platform, and calculate how many extra paying users each segment gains by reaching it.
- Growth Accounting: New, Resurrected and Churned Users, the MAU Identity, Quick Ratio and Churn Rate. You rebuild monthly active households from new, resurrected and churned users, calculate the quick ratio and churn rate per month and the quick ratio and net growth for the year.
- User States: Classifying New, Retained, Resurrected, Churned and Dormant Users with IFS and COUNTIFS. You classify 18 households into growth-accounting states with IFS, count resurrected and churned households with COUNTIFS and calculate the quick ratio and month-to-month retention.
- The Experiment Backlog: RICE Scores, Ranking, Balance across AARRR Stages and a Top-3 Rule. You score 14 experiment ideas with RICE, rank them, check how the backlog is spread across the AARRR stages and highlight the three ideas to run next.
- Planning a Test: Absolute Effect, Sample Size per Variant, Weeks to Run and a Run-or-Rethink Verdict. You translate relative lifts into absolute effects, estimate the sample per variant with a rule of thumb, convert it into weeks of traffic and decide which tests are worth running.
- Reading a Test: Relative Lift, z-Score, Two-Sided p-Value and a Ship, Revert or Inconclusive Decision. You read ten finished experiments: relative lift, z-score and two-sided p-value, a three-way decision with nested IF, and a rule that highlights the variants to ship.
- Experiment Cadence: Win Rate to Date, Velocity Trend with SLOPE, FORECAST.LINEAR and Expected Wins. You measure Kestrelwick's experiment system over twelve weeks: the win rate to date with an expanding range, the launch trend with SLOPE and FORECAST.LINEAR, and the expected wins of the next quarter.
- Compounding Wins: Growth Factors, PRODUCT, the Compounded Lift and Small Weekly Improvements. You combine ten shipped wins into a cumulative multiplier with PRODUCT, compare the compounded lift with the plain sum and calculate what a small weekly improvement becomes in a year.
- Growth Loops: Simulating Loop Sign-Ups, Churn and Active Households Month by Month. You simulate a referral loop for twelve months, calculate loop sign-ups, churn and the active base from the previous row, measure how much of the growth the loop delivers and chart the result.
- Channel Unit Economics: CAC, Lifetime Value, the LTV to CAC Ratio and Payback Months. You calculate CAC, lifetime value, the LTV to CAC ratio and the payback period for eight channels and rank them in a chart.
- Scenario Grids: Mixed References, COUNTIF Targets, INDEX-MATCH Lookups and a Heat Map. You build a grid of paying households for 30 combinations of activation and retention, count the scenarios that reach a target, look up a chosen scenario, solve for the retention needed and shade the grid as a heat map.
- Trusting a Test: Expected Split, Chi-Square, the SRM p-Value and a Status Rule. You check ten experiments for a sample ratio mismatch: expected group size, chi-square, p-value with CHISQ.DIST.RT, a status flag and a rule that marks the tests to distrust.
- Final Project: A Growth Review of the Year with Growth Accounting, Activation, Win Rate, a Forecast and a Trend Chart. You combine the course into one dashboard: MAU identity, quick ratio, activation rate, the win rate of the experiment system, a linear forecast of next month's sign-ups and a trend chart.
Summary
The AARRR framework splits growth into acquisition, activation, retention, referral and revenue. Each stage gets one step conversion rate that divides by the stage before it, and the overall visit-to-paid rate divides paying users by visitors. In the March cohort 74,000 visitors became 382 paying users, an overall rate of 0.52 %, and the channel with the most visitors was not the channel with the most value. The total row must be a ratio of totals, never an average of rates.
Now scroll up to the lab and build the five rates yourself. It takes about fifteen minutes, the Coach checks each step, and you will end up with a funnel table you can reuse with your own channel data.
Frequently asked questions
What does AARRR stand for?
Acquisition, Activation, Retention, Referral and Revenue. The framework, nicknamed the pirate metrics, sorts user numbers into these five stages so that a team can see which stage loses the most users and assign one owner and one conversion rate to each stage.
How do I calculate a step conversion rate?
Divide the number of users at a stage by the number of users at the stage directly before it. For example, the activation rate is activated users divided by sign-ups. In a spreadsheet this is one division with relative references that you copy down to every row.
What is the difference between a step rate and the overall rate?
A step rate compares neighbouring stages, such as sign-ups to visitors. The overall visit-to-paid rate divides the last stage by the first. When every stage is a subset of the previous one, the overall rate equals the product of all step rates between them.
Why should the total row not be an average of the channel rates?
An average gives every channel the same weight, so a small channel counts as much as a large one. The correct total rate divides the sum of the numerator column by the sum of the denominator column, which weights each channel by its size.
How do I define activation for my own product?
Pick the first action that shows a user has received real value and that predicts coming back, for instance finishing a first project or linking a first account. Check with retention data that users who do it stay longer than those who do not, and write the definition down so it stays stable.
Do I need special software for this lab?
No. The lab runs in your browser with a built-in spreadsheet and the Coach checks each formula instantly. The same formulas work in any spreadsheet program, and the stage counts can come from your own analytics or product data.
Keep going: Growth Experiments with AARRR Pirate Metrics: North-Star Metric, Funnel Rates, Growth Accounting, Experiment Backlog & Compounding Loops
This free lab is lesson 1 of 20. Unlock the other 19 hands-on labs, the final exam and your certificate for a one-time $9.90.
- The North-Star Metric: Input Metrics, Week-over-Week Growth, the Compound Rate and Weeks to a Goal
- Acquisition Quality: Cost per Sign-Up, Cost per Activated User, Channel Shares and Ranking
- Activation: Defining the Aha Moment with IF and AND, the Activation Rate and the Retention Gap
- Retention: Cohort Tables, Mixed References, Weighted Retention per Month and Retention Curves
- Referral: Share of Inviters, Invite Conversion, the K-Factor and the Amplification Factor
- Revenue: Paid Share, ARPPU, ARPU, Subscriber Churn and a Simple Lifetime Value
- Sizing the Leak: Multiplying Step Rates, Activation Targets with XLOOKUP and Extra Paying Users
- Growth Accounting: New, Resurrected and Churned Users, the MAU Identity, Quick Ratio and Churn Rate
- User States: Classifying New, Retained, Resurrected, Churned and Dormant Users with IFS and COUNTIFS
- The Experiment Backlog: RICE Scores, Ranking, Balance across AARRR Stages and a Top-3 Rule
- Planning a Test: Absolute Effect, Sample Size per Variant, Weeks to Run and a Run-or-Rethink Verdict
- Reading a Test: Relative Lift, z-Score, Two-Sided p-Value and a Ship, Revert or Inconclusive Decision
- Experiment Cadence: Win Rate to Date, Velocity Trend with SLOPE, FORECAST.LINEAR and Expected Wins
- Compounding Wins: Growth Factors, PRODUCT, the Compounded Lift and Small Weekly Improvements
- Growth Loops: Simulating Loop Sign-Ups, Churn and Active Households Month by Month
- Channel Unit Economics: CAC, Lifetime Value, the LTV to CAC Ratio and Payback Months
- Scenario Grids: Mixed References, COUNTIF Targets, INDEX-MATCH Lookups and a Heat Map
- Trusting a Test: Expected Split, Chi-Square, the SRM p-Value and a Status Rule
- Final Project: A Growth Review of the Year with Growth Accounting, Activation, Win Rate, a Forecast and a Trend Chart