Growth Marketing skill 03 of 13
Growth Model and Forecasting SKILL.md
The Growth Model skill makes Claude, Codex or another AI agent build cohort-based growth models and forecasts. It combines new users by channel, activation, retention curves, monetisation, expansion, referral and reactivation, calibrates the model on history, projects a baseline, runs sensitivity analysis to rank levers, builds scenarios tied to initiatives and sets input targets with owners.
growth-model/SKILL.md, version 1.0.0, 1,580 words.
A growth model is a simple, explicit description of how a business grows: which inputs (new users by channel, activation, retention, monetisation, referral) combine to produce the outcome, and by how much. It turns "we need to grow 50%" into "we need activation from 22% to 30% and two new channels at this scale", and shows which lever moves the outcome most. Most companies set targets top-down and hope; a model makes targets traceable and trade-offs visible.
This skill builds growth models in a spreadsheet or notebook, calibrates them with historical data, runs sensitivity analysis to find the levers that matter, and produces forecasts with scenarios that teams can own.
Operating principles
- Simple enough to explain. A model the team understands beats a perfect one nobody trusts.
- Cohort-based. New users join in cohorts and decay over time; growth stacks cohorts on top of each other.
- Calibrated on history. Every assumption comes from your data or is labelled as an assumption.
- Sensitivity before targets. Find which input changes the outcome most before assigning goals.
- Living document. Update monthly with actuals and adjust assumptions.
Model building blocks
| Block | Typical formula | Data source |
|---|---|---|
| New users | Sum of channel volumes (paid, organic, referral, partnerships) | Analytics, attribution, ad platforms |
| Activation | New users × activation rate | Product analytics |
| Retention | Activated users × retention curve by month since start | Cohort analysis |
| Monetisation | Active users × paid conversion × ARPU (or orders × AOV) | Billing, commerce |
| Expansion and contraction | Existing revenue × net expansion rate | Billing |
| Referral or loop | Active users × invite rate × invite conversion | Product analytics |
| Reactivation | Lapsed users × reactivation rate | Lifecycle data |
When to use, and when not to
Use it for annual and quarterly planning, setting input targets, deciding between growth investments, fundraising forecasts, and when leadership asks what it would take to hit a goal.
Don't use it for detailed media budget allocation (use the Performance Marketing media-planning-budgeting skill) or financial statements beyond growth drivers.
Inputs to ask for, in one message
- Monthly new users or customers by channel for 12 to 24 months.
- Activation rate by cohort and channel.
- Cohort retention curves (users and revenue).
- Monetisation: conversion to paid, ARPU or AOV, expansion.
- Referral or invite data if any.
- Planned initiatives and their expected effects (to label as assumptions).
- The target and time frame.
Raw SKILL.md file
This is the exact file agents read, including the credit header and licence notice. Save it as growth-model/SKILL.md.
---
name: growth-model
description: "Build cohort-based growth models and forecasts: new users by channel, activation, retention curves, monetisation, expansion, referral and reactivation, calibrated on history, a baseline projection, sensitivity analysis to rank levers, conservative, expected and stretch scenarios tied to initiatives, input targets with owners and monthly actuals versus forecast. Use whenever someone asks for a growth model, growth forecast, what it takes to hit a target, which growth lever matters most, cohort-based revenue forecasts or planning growth for a quarter or year."
license: CC-BY-4.0
metadata:
author: Afeez Adebayo
author_url: https://afeezadebayo.com
author_linkedin: https://www.linkedin.com/in/afeez-adebayo/
author_x: https://x.com/za_icon
source: https://afeezadebayo.com/skills-md/marketing/growth-marketing/growth-model/
category: marketing
subcategory: growth-marketing
version: "1.0.0"
created: "2026-10-10"
updated: "2026-10-10"
---
# Growth Model and Forecasting
A growth model is a simple, explicit description of how a business grows: which inputs (new users by channel, activation, retention, monetisation, referral) combine to produce the outcome, and by how much. It turns "we need to grow 50%" into "we need activation from 22% to 30% and two new channels at this scale", and shows which lever moves the outcome most. Most companies set targets top-down and hope; a model makes targets traceable and trade-offs visible.
This skill builds growth models in a spreadsheet or notebook, calibrates them with historical data, runs sensitivity analysis to find the levers that matter, and produces forecasts with scenarios that teams can own.
## Operating principles
1. **Simple enough to explain.** A model the team understands beats a perfect one nobody trusts.
2. **Cohort-based.** New users join in cohorts and decay over time; growth stacks cohorts on top of each other.
3. **Calibrated on history.** Every assumption comes from your data or is labelled as an assumption.
4. **Sensitivity before targets.** Find which input changes the outcome most before assigning goals.
5. **Living document.** Update monthly with actuals and adjust assumptions.
## Model building blocks
| Block | Typical formula | Data source |
|---|---|---|
| New users | Sum of channel volumes (paid, organic, referral, partnerships) | Analytics, attribution, ad platforms |
| Activation | New users × activation rate | Product analytics |
| Retention | Activated users × retention curve by month since start | Cohort analysis |
| Monetisation | Active users × paid conversion × ARPU (or orders × AOV) | Billing, commerce |
| Expansion and contraction | Existing revenue × net expansion rate | Billing |
| Referral or loop | Active users × invite rate × invite conversion | Product analytics |
| Reactivation | Lapsed users × reactivation rate | Lifecycle data |
## When to use, and when not to
Use it for annual and quarterly planning, setting input targets, deciding between growth investments, fundraising forecasts, and when leadership asks what it would take to hit a goal.
Don't use it for detailed media budget allocation (use the Performance Marketing `media-planning-budgeting` skill) or financial statements beyond growth drivers.
## Inputs to ask for, in one message
1. Monthly new users or customers by channel for 12 to 24 months.
2. Activation rate by cohort and channel.
3. Cohort retention curves (users and revenue).
4. Monetisation: conversion to paid, ARPU or AOV, expansion.
5. Referral or invite data if any.
6. Planned initiatives and their expected effects (to label as assumptions).
7. The target and time frame.
## Workflow
### Step 1. Choose the shape
SaaS: users or accounts, activation, retention, conversion, ARPU, expansion. E-commerce: new customers, repeat purchase curves, AOV. Marketplace: supply and demand sides with liquidity. Consumer app: installs, activation, retention, monetisation, virality.
### Step 2. Calibrate on history
Build the model for the past 12 months and compare predicted with actual outcomes month by month. Adjust until it fits reasonably.
### Step 3. Project the baseline
If nothing changes, where does the outcome land? This is the honest starting point for targets.
### Step 4. Run sensitivity analysis
Change each input by a realistic amount (for example +10% relative) and record the effect on the outcome after 12 months. Rank inputs by impact.
### Step 5. Build scenarios
Conservative, expected and stretch, each a set of input assumptions tied to named initiatives, with evidence labels.
### Step 6. Set input targets
Give each team a target for the input they own, consistent with the expected scenario.
### Step 7. Track actuals
Monthly: actual inputs versus assumptions; update the forecast.
## Sensitivity example (12-month impact on revenue)
| Input | Change | Revenue impact |
|---|---|---|
| Month-3 retention | +10% relative | Largest (compounds across cohorts) |
| Activation rate | +10% relative | Large |
| Paid conversion | +10% relative | Medium |
| New users (paid) | +10% | Medium, with rising cost |
| Invite rate | +10% | Small now, grows if loop strengthens |
Illustrative ranking; retention often matters most because it compounds.
## Spreadsheet structure
```
Tab 1: Assumptions (each with source and evidence label)
Tab 2: Cohorts (rows = sign-up month, columns = months since sign-up)
Tab 3: Channels (volumes and costs by month)
Tab 4: Outputs (active users, paying customers, revenue, CAC, payback)
Tab 5: Scenarios and sensitivity
Tab 6: Actuals vs forecast
```
## How cohorts stack into growth
| Sign-up month | Month 0 | Month 1 | Month 2 | Month 3 |
|---|---|---|---|---|
| Jan cohort (1,000 activated) | 1,000 | 620 | 510 | 470 |
| Feb cohort (1,100 activated) | | 1,100 | 680 | 560 |
| Mar cohort (1,200 activated) | | | 1,200 | 740 |
| Total active | 1,000 | 1,720 | 2,390 | 1,770 + Apr cohort |
Each month's active base is the sum of every cohort still active. Improving retention lifts every column at once, which is why it usually beats acquisition in sensitivity tests.
## Assumption register
| Assumption | Value | Source | Evidence label | Owner | Review date |
|---|---|---|---|---|---|
| Activation rate | 24% → 30% | Last 6 months 22-25%; onboarding test planned | Assumption (test pending) | Growth PM | Monthly |
| Month-3 retention | 47% | Cohort data Jan-Jun | Data | Analyst | Monthly |
| New channel volume | 400 users/month | Pilot results | Test (small) | Acquisition lead | After 8 weeks |
## Forecast accuracy check
Each month, compare forecast and actual for the outcome and each input. Track the error. If one input is consistently wrong, fix that assumption rather than adjusting everything. A model that is within about 10% on the outcome after three months is usually good enough to plan with.
## Communicating the model
- Lead with the baseline ("if nothing changes") before showing targets.
- Show which assumptions drive the gap between baseline and goal.
- Name the initiatives behind each assumption and who owns them.
- Present ranges, not single numbers, for the stretch scenario.
## Edge cases and judgement calls
| Situation | What to do |
|---|---|
| Little history (new product) | Use benchmarks from similar products, labelled; update monthly |
| Big seasonality | Model seasonal indices per month |
| Pricing change planned | Model before and after separately; watch conversion and churn |
| Marketplace | Model supply, demand and match rates; liquidity is usually the constraint |
| Investors want aggressive numbers | Show the stretch scenario with the assumptions it requires |
## Handoffs to related skills
- Metric tree and inputs: `north-star-metric-tree`
- Retention curves: `cohort-retention-analysis`
- Funnel rates: `funnel-analysis`
- Channel volumes: `acquisition-channel-testing`
- Loop contributions: `growth-loops`
- Reporting actuals: `growth-reporting`
## Worked example
A subscription fitness app wanted to double revenue in 12 months. The calibrated model showed the baseline reaching about 1.4 times current revenue. Sensitivity analysis found month-3 retention was the strongest lever, followed by trial-to-paid conversion; doubling paid acquisition alone would add less and raise CAC. The expected scenario combined a retention initiative (personalised plans after week two), a trial conversion experiment programme and one new channel, each with input targets and owners. Monthly tracking showed retention improving faster than conversion, and the forecast was updated to reflect it.
## Output format
```
# Growth model: <company>, <date>
## Model shape and assumptions (with sources and labels)
## Calibration against history
## Baseline projection
## Sensitivity ranking
## Scenarios (conservative, expected, stretch) with initiatives
## Input targets and owners
## Tracking plan for actuals
```
## Common mistakes
- Top-down targets with no link to inputs.
- Ignoring cohort decay, so forecasts overshoot.
- Unlabelled assumptions presented as facts.
- Treating acquisition as the only lever.
- Never updating the model with actuals.
## Questions to ask before you model
- What outcome and time frame does the model need to answer?
- Which inputs have reliable history?
- Which initiatives are planned, and what evidence supports their expected effect?
## Self-check before you answer
- Is the model cohort-based and calibrated on history?
- Is every assumption sourced or labelled?
- Did I run sensitivity before setting targets?
- Are input targets owned and tracked monthly?
## Further reading
- Reforge, growth models: https://www.reforge.com/blog
- a16z, consumer metrics and growth: https://a16z.com/
---
Created by Afeez Adebayo, Marketing, SEO & Growth Expert
Website: https://afeezadebayo.com
LinkedIn: https://www.linkedin.com/in/afeez-adebayo/
X: https://x.com/za_icon
Source: https://afeezadebayo.com/skills-md/marketing/growth-marketing/growth-model/
Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). You may copy, share and adapt this skill, including for commercial use, as long as you credit Afeez Adebayo and link to the source URL above. Keep this notice in any copy or adapted version.
Workflow
Step 1. Choose the shape
SaaS: users or accounts, activation, retention, conversion, ARPU, expansion. E-commerce: new customers, repeat purchase curves, AOV. Marketplace: supply and demand sides with liquidity. Consumer app: installs, activation, retention, monetisation, virality.
Step 2. Calibrate on history
Build the model for the past 12 months and compare predicted with actual outcomes month by month. Adjust until it fits reasonably.
Step 3. Project the baseline
If nothing changes, where does the outcome land? This is the honest starting point for targets.
Step 4. Run sensitivity analysis
Change each input by a realistic amount (for example +10% relative) and record the effect on the outcome after 12 months. Rank inputs by impact.
Step 5. Build scenarios
Conservative, expected and stretch, each a set of input assumptions tied to named initiatives, with evidence labels.
Step 6. Set input targets
Give each team a target for the input they own, consistent with the expected scenario.
Step 7. Track actuals
Monthly: actual inputs versus assumptions; update the forecast.
Sensitivity example (12-month impact on revenue)
| Input | Change | Revenue impact |
|---|---|---|
| Month-3 retention | +10% relative | Largest (compounds across cohorts) |
| Activation rate | +10% relative | Large |
| Paid conversion | +10% relative | Medium |
| New users (paid) | +10% | Medium, with rising cost |
| Invite rate | +10% | Small now, grows if loop strengthens |
Illustrative ranking; retention often matters most because it compounds.
Spreadsheet structure
Tab 1: Assumptions (each with source and evidence label)
Tab 2: Cohorts (rows = sign-up month, columns = months since sign-up)
Tab 3: Channels (volumes and costs by month)
Tab 4: Outputs (active users, paying customers, revenue, CAC, payback)
Tab 5: Scenarios and sensitivity
Tab 6: Actuals vs forecastHow cohorts stack into growth
| Sign-up month | Month 0 | Month 1 | Month 2 | Month 3 |
|---|---|---|---|---|
| Jan cohort (1,000 activated) | 1,000 | 620 | 510 | 470 |
| Feb cohort (1,100 activated) | 1,100 | 680 | 560 | |
| Mar cohort (1,200 activated) | 1,200 | 740 | ||
| Total active | 1,000 | 1,720 | 2,390 | 1,770 + Apr cohort |
Each month's active base is the sum of every cohort still active. Improving retention lifts every column at once, which is why it usually beats acquisition in sensitivity tests.
Assumption register
| Assumption | Value | Source | Evidence label | Owner | Review date |
|---|---|---|---|---|---|
| Activation rate | 24% → 30% | Last 6 months 22-25%; onboarding test planned | Assumption (test pending) | Growth PM | Monthly |
| Month-3 retention | 47% | Cohort data Jan-Jun | Data | Analyst | Monthly |
| New channel volume | 400 users/month | Pilot results | Test (small) | Acquisition lead | After 8 weeks |
Forecast accuracy check
Each month, compare forecast and actual for the outcome and each input. Track the error. If one input is consistently wrong, fix that assumption rather than adjusting everything. A model that is within about 10% on the outcome after three months is usually good enough to plan with.
Communicating the model
- Lead with the baseline ("if nothing changes") before showing targets.
- Show which assumptions drive the gap between baseline and goal.
- Name the initiatives behind each assumption and who owns them.
- Present ranges, not single numbers, for the stretch scenario.
Edge cases and judgement calls
| Situation | What to do |
|---|---|
| Little history (new product) | Use benchmarks from similar products, labelled; update monthly |
| Big seasonality | Model seasonal indices per month |
| Pricing change planned | Model before and after separately; watch conversion and churn |
| Marketplace | Model supply, demand and match rates; liquidity is usually the constraint |
| Investors want aggressive numbers | Show the stretch scenario with the assumptions it requires |
Handoffs to related skills
- Metric tree and inputs:
north-star-metric-tree - Retention curves:
cohort-retention-analysis - Funnel rates:
funnel-analysis - Channel volumes:
acquisition-channel-testing - Loop contributions:
growth-loops - Reporting actuals:
growth-reporting
Worked example
A subscription fitness app wanted to double revenue in 12 months. The calibrated model showed the baseline reaching about 1.4 times current revenue. Sensitivity analysis found month-3 retention was the strongest lever, followed by trial-to-paid conversion; doubling paid acquisition alone would add less and raise CAC. The expected scenario combined a retention initiative (personalised plans after week two), a trial conversion experiment programme and one new channel, each with input targets and owners. Monthly tracking showed retention improving faster than conversion, and the forecast was updated to reflect it.
Output format
# Growth model: <company>, <date>
## Model shape and assumptions (with sources and labels)
## Calibration against history
## Baseline projection
## Sensitivity ranking
## Scenarios (conservative, expected, stretch) with initiatives
## Input targets and owners
## Tracking plan for actualsCommon mistakes
- Top-down targets with no link to inputs.
- Ignoring cohort decay, so forecasts overshoot.
- Unlabelled assumptions presented as facts.
- Treating acquisition as the only lever.
- Never updating the model with actuals.
Questions to ask before you model
- What outcome and time frame does the model need to answer?
- Which inputs have reliable history?
- Which initiatives are planned, and what evidence supports their expected effect?
Self-check before you answer
- Is the model cohort-based and calibrated on history?
- Is every assumption sourced or labelled?
- Did I run sensitivity before setting targets?
- Are input targets owned and tracked monthly?
Further reading
- Reforge, growth models: https://www.reforge.com/blog
- a16z, consumer metrics and growth: https://a16z.com/
On this page
- When to use this skill, and when not to
- Example prompts
- Example output
- What the agent needs from you
- How this skill hands off to related skills
- How to run Growth Model and Forecasting step by step
- Situations this skill handles
- What makes this skill different
- Practical tips from real projects
- Frequently asked questions
- Changelog
When to use this skill, and when not to
Use it for annual and quarterly planning, setting input targets, choosing between growth investments, fundraising forecasts and whenever leadership asks what it would take to hit a goal.
A growth model turns “grow 50%” into “raise activation from 22% to 30% and add two channels at this scale”, and shows which lever moves the outcome most. Because cohorts stack on each other, retention usually matters more than teams expect, which sensitivity analysis makes visible.
Don’t use it for media budget allocation (use the Performance Marketing Media Planning skill) or full financial statements.
Example prompts
- “What would it take to double revenue in 12 months? Build the model.”
- “Which growth lever matters most for us?”
- “Calibrate our forecast on the last 12 months of cohorts.”
- “Build conservative, expected and stretch growth scenarios.”
- “Set input targets for each growth team from the model.”
Example output
The skill returns a fixed structure. A shortened example for an imaginary site:
# Growth model: example fitness app, 10 October 2026 ## Baseline (no change) About 1.4x revenue in 12 months ## Sensitivity (+10% relative, 12-month revenue impact) 1. Month-3 retention (largest) 2. Trial-to-paid conversion 3. Activation 4. Paid acquisition volume (higher CAC) ## Expected scenario (2x) Retention initiative + conversion experiments + one new channel, each with input targets and owners
Every assumption is sourced or labelled, and the baseline is shown before any target, so stretch goals are traceable to specific inputs.
What the agent needs from you
The skill asks for 12 to 24 months of new users by channel, activation by cohort, cohort retention curves for users and revenue, monetisation data, referral data if any, planned initiatives and their expected effects and the target and time frame.
How this skill hands off to related skills
The model connects strategy and execution. Inputs come from North Star Metric, Cohort Retention, Funnel Analysis, Channel Testing and Growth Loops, and actuals are tracked in Growth Reporting.
How to run Growth Model and Forecasting step by step
- Install the skill as
growth-model/SKILL.md. - Export monthly new users by channel and activation by cohort.
- Add cohort retention for users and revenue, plus monetisation.
- Share the target and planned initiatives.
- Review calibration, baseline, sensitivity and scenarios.
- Assign input targets and track actuals monthly.
Situations this skill handles
Models take different shapes. The skill handles these situations:
| Situation | What the skill does |
|---|---|
| SaaS | Accounts, activation, retention, conversion, ARPU, expansion |
| E-commerce | New customers, repeat purchase curves, AOV |
| Marketplace | Supply, demand and match rates |
| Consumer app | Installs, activation, retention, monetisation, virality |
| Big seasonality | Seasonal indices by month |
| Investors want aggressive numbers | Stretch scenario with explicit assumptions |
What makes this skill different
Most growth targets are set top-down and hoped for. This skill builds a cohort-based model calibrated on your history, shows the baseline plainly, ranks levers by sensitivity and ties every scenario to named initiatives with evidence labels.
It stays alive after planning: monthly actuals versus forecast show which assumptions drift, so the model improves each month instead of becoming a forgotten spreadsheet.
Practical tips from real projects
- Show the baseline first.
- Run sensitivity before setting targets.
- Label every assumption.
- Update monthly with actuals.
A worked example: a subscription fitness app wanted to double revenue in 12 months. The calibrated model showed the baseline reaching about 1.4 times current revenue. Sensitivity analysis found month-3 retention was the strongest lever, followed by trial-to-paid conversion; doubling paid acquisition alone would add less and raise CAC. The expected scenario combined personalised plans after week two, a conversion experiment programme and one new channel, each with input targets and owners. Monthly tracking showed retention improving faster than conversion, and the forecast was updated accordingly.
Cohort stacking explains why retention wins. Each month’s active base is every cohort still active added together; improving retention lifts every column at once, while extra acquisition only adds one new row. A modest retention gain often outweighs a large acquisition push over a year.
The spreadsheet structure keeps the model readable: assumptions with sources and labels, cohorts by sign-up month and months since start, channel volumes and costs, outputs (active users, paying customers, revenue, CAC, payback), scenarios and sensitivity, and actuals versus forecast. An assumption register records each value, its source, evidence label, owner and review date.
Communicating the model matters as much as building it: lead with the baseline, show which assumptions drive the gap to the goal, name the initiatives and owners behind each and present ranges rather than single numbers for stretch scenarios.
For marketplaces the model usually reveals liquidity, the share of supply or demand that finds a match, as the true constraint, which changes where the team should invest first.
Model blocks are simple: new users as the sum of channel volumes, activation as new users times activation rate, retention as activated users times the retention curve, monetisation as active users times paid conversion times ARPU, expansion as existing revenue times net expansion, referral as active users times invite rate times invite conversion, and reactivation as lapsed users times reactivation rate. Each block draws on a specific data source, so the team knows where every number comes from and who maintains it.
Forecast accuracy is checked every month. Compare forecast and actual for the outcome and each input, and track the error. If one input is consistently wrong, fix that assumption rather than adjusting everything. New products with little history start from labelled benchmarks of similar products and replace them with real data as cohorts mature, while businesses with big seasonality add monthly seasonal indices so growth isn’t confused with calendar effects. Planned pricing changes are modelled before and after separately, with conversion and churn watched closely, because pricing often moves both at once.
Frequently asked questions
What is a growth model?
A simple, explicit description of how inputs such as new users, activation, retention, monetisation and referral combine to produce growth, with numbers from your own data.
Why cohort-based?
Users join in cohorts and decay over time; each month’s active base is the sum of every cohort still active. Ignoring that makes forecasts overshoot.
Which lever usually matters most?
Often retention, because improvements compound across every cohort. Sensitivity analysis shows the ranking for your business.
How accurate should the model be?
Within about 10% on the outcome after three months is usually good enough to plan with. Track errors by input and fix the assumptions that drift.
Can I build one with little history?
Yes, using labelled benchmarks from similar products, updated monthly as your data grows.
Spreadsheet or code?
Either. A spreadsheet is easier to share; notebooks suit larger, more complex models.
Changelog
- 1.0.0, 10 October 2026: first release.
Use, adapt and share this skill under CC BY 4.0: credit Afeez Adebayo and link to this page.
All Growth Marketing skills
- Growth Audit
- North Star Metric and Metric Tree
- Growth Experiment Backlog and Cadence
- Experiment Design for Growth
- Experiment Readouts and Analysis
- Acquisition Channel Testing
- Growth Loops
- Referral Programmes
- Funnel Analysis
- Cohort Retention Analysis
- Tracking Plan and Event Taxonomy
- Growth Reporting and Attribution
