growth analytics
A growth PM would focus on business metrics, revenue, user growth — funnel optimization. What percentage of people who land on the homepage actually end up converting?
You have been given the “growth” mandate. Maybe the CEO said “double our MAU by March.” Maybe your VP created a growth squad and put your name on it. Either way, you now own a number that only goes up if you understand where users come from, why they stay, and what brings them back.
Most PMs in this seat reach for dashboards. They stare at aggregate line charts. They celebrate when the line goes up and panic when it goes down. They never ask the question that actually matters: which users, acquired when, doing what?
Growth analytics is not about tracking more. It is about decomposing the number into pieces small enough to act on. Funnels show you where users drop. Cohorts show you when they drop. Growth loops show you how to make the whole system compound. Retention curves show you whether you have a product worth growing at all.
Funnel analysis: find the leak before you pour more water
A funnel is a sequence of steps a user must complete to reach a desired outcome. Landing page visit to signup. Signup to first transaction. First transaction to second transaction. Each step has a conversion rate. Each conversion rate is a lever.
Here is the mistake almost every growth PM makes: they try to optimize the top of the funnel first. More traffic, more ads, more social media spend. This is exactly backwards. If your funnel leaks at step three, every user you pour in at step one drains out before reaching value. Fix the narrowest point first, then widen the top.
Growth review at a Bangalore-based fintech. The team launched a UPI-based savings product two months ago.
Growth PM: “We're getting 12,000 app installs per week from our Google Ads campaigns. Our CAC is Rs 85.”
Head of Product: “How many of those 12,000 complete KYC?”
Growth PM: “About 3,200. Around 27%.”
Head of Product: “And how many make their first deposit?”
Growth PM: “...640. About 5% of installs.”
Head of Product: “So we are spending Rs 10.2 lakh per week on ads to get 640 depositors. That is Rs 1,594 per depositing user. Our ARPU is Rs 120 per month. It takes us 13 months to break even on one user — if they even retain that long.”
The room recalculated in silence. The funnel was not an acquisition problem. It was a KYC completion problem.
The cheapest growth lever was not more ads. It was fixing the KYC flow that was killing 73% of users before they reached value.
How to build a useful funnel
Step 1: Define the steps. Not every click is a funnel step. Only include steps that represent meaningful progression toward the outcome. For an e-commerce app: visit, add to cart, initiate checkout, complete payment. Not: visit, scroll, click category, view product, click variant, add to cart, view cart, edit cart, initiate checkout, enter address, select payment, confirm, complete. Too many steps and you cannot see the pattern.
Step 2: Measure conversion between each pair of adjacent steps. This is your drop-off rate. A 60% drop between “add to cart” and “initiate checkout” is a different problem than a 60% drop between “initiate checkout” and “complete payment.” The first suggests browsing behavior. The second suggests a broken checkout flow or payment gateway failure.
Step 3: Segment the funnel. The aggregate funnel hides the story. Segment by acquisition channel (organic vs. paid vs. referral), by device (Android vs. iOS — this matters enormously in India where low-end Android devices dominate), by geography (tier-1 vs. tier-2 cities), by time (weekday vs. weekend). The segment that performs worst is where your biggest opportunity is.
Step 4: Set conversion benchmarks and monitor weekly. Not monthly. Not quarterly. Weekly. Funnel degradation happens fast — a bad app update can tank activation in 48 hours — and you need to catch it before a full cohort is lost.
Cohort analysis: the single most important skill in growth analytics
If you learn one analytical technique from this entire manual, make it cohort analysis. Aggregate metrics lie. Cohorts tell the truth.
I call this The Cohort Lens — a diagnostic discipline. When any metric misses target, your first move is to segment by cohort (sign-up week, acquisition channel, geography, device) BEFORE deciding to iterate or kill. The feature that “failed” in aggregate might work perfectly for one segment. That segment is your beachhead. Without the Cohort Lens, you kill features that are working and keep features that are dying — because the average hides both.
A cohort is a group of users who share a common characteristic within a defined time period. The most common: users who signed up in a given week or month. You track their behavior over time — do they come back in week 1, week 2, week 4, week 8?
Reading a cohort retention table
Here is a simplified cohort retention table. Each row is a signup cohort (the month users joined). Each column is how many months later you are checking.
| Cohort | Month 0 | Month 1 | Month 2 | Month 3 | Month 6 |
|---|---|---|---|---|---|
| Oct | 100% | 40% | 28% | 22% | 14% |
| Nov | 100% | 38% | 25% | 19% | — |
| Dec | 100% | 35% | 21% | — | — |
| Jan | 100% | 30% | — | — | — |
Read it row by row: each cohort’s retention over time. Read it column by column: whether newer cohorts are performing better or worse than older ones at the same age.
In this table, month-1 retention is declining: 40%, 38%, 35%, 30%. Each new cohort retains worse than the one before. This is a product problem, not a marketing problem. Something about the product experience is degrading for newer users — maybe the onboarding changed, maybe new acquisition channels are bringing lower-intent users, maybe a feature broke.
The power of cohort analysis is that it separates when users joined from how they behaved. Without it, aggregate MAU can look healthy while every underlying cohort is decaying.
Types of cohorts that matter
Acquisition cohorts (most common): grouped by signup date. Shows you retention trajectory over time.
Behavioral cohorts: grouped by an action they took. “Users who completed onboarding in under 3 minutes” vs. “users who took more than 10 minutes.” If the fast-onboarding cohort retains at 2x the rate, you know exactly what to optimize.
Feature cohorts: grouped by feature usage. “Users who used the search feature in their first week” vs. “users who did not.” This tells you which features drive long-term retention — which is different from which features are popular.
Channel cohorts: grouped by acquisition source. Organic search users vs. Instagram ad users vs. WhatsApp referral users. In India, this analysis routinely reveals that WhatsApp referrals retain 2-3x better than paid social, because referral users arrive with trust already established.
Growth loops: the shift from linear to compounding
A funnel is linear. Users enter at the top, some fraction exits at the bottom, and you have to keep feeding the top to maintain output. A growth loop is circular. The output of the product feeds back as an input for new user acquisition.
This is not a theoretical distinction. It is the difference between companies that scale and companies that plateau.
The linear model: Spend Rs 100 on ads. Get 10 users. 2 retain. Spend Rs 100 again. Get 10 more users. 2 more retain. Your user base grows arithmetically. The moment you stop spending, growth stops.
The loop model: Spend Rs 100 on ads. Get 10 users. 2 retain. Those 2 users each invite 1 friend. 2 new users arrive for free. Of those, 1 retains and invites another. Your user base grows geometrically. Even when you reduce ad spend, growth continues from the loop.
Three types of growth loops
Content loops. Users create content as part of using the product. That content gets indexed by search engines or shared on social platforms. New users discover it, sign up, create their own content, and the cycle repeats. Example: Stack Overflow. A developer asks a question. An expert answers it. Google indexes the answer. Another developer finds it via search, signs up, asks their own question.
Viral loops. The product becomes more valuable when shared. The act of using it naturally involves other people. Example: UPI payments in India. To send money, you need a recipient. That recipient downloads the app. Now they can send money to others, who also download the app. PhonePe and Google Pay grew through this mechanic — not because users loved sharing the brand, but because the core use case required it.
Paid loops. Revenue from existing users funds acquisition of new users, and the LTV-to-CAC ratio allows profitable reinvestment. Example: a SaaS company with Rs 5,000 LTV and Rs 1,500 CAC can reinvest Rs 1,500 from every retained customer into acquiring two more. The loop works as long as the unit economics hold.
How to identify your loop
Ask two questions. First: what output does my product generate that other people see? Content, transactions, shared links, notifications to non-users, public profiles, exported reports. Second: can that output bring a new user in without my marketing team doing anything?
If the answer to both is yes, you have a loop. Measure it. If the answer is no, you are running a funnel business — which is fine, but you need to be honest about it and manage your acquisition costs accordingly.
Not every product has a natural loop. Enterprise software sold to procurement teams does not go viral. B2B tools with NDAs do not generate public content. Do not force a loop where none exists. Focus instead on making your funnel as efficient as possible.
Retention curves: the shape tells the story
A retention curve plots the percentage of a cohort that remains active over time. The shape of that curve tells you almost everything you need to know about your product’s health.
The death curve. Retention drops steeply and keeps dropping toward zero. No flattening. This means users try the product, get some initial value (or none), and leave. Most products that fail have this shape. If your month-6 retention is below 5% for a consumer app or below 70% for a B2B SaaS, you have a product problem, not a growth problem. Stop spending on acquisition and fix the core experience.
The flattening curve. Retention drops initially, then levels off at some percentage. This is what you want. The early drop is natural — some users will always churn. But the flattening means you have found a core group who derives ongoing value. The higher the floor, the better the product. A consumer app that flattens at 25% month-over-month retention is strong. A B2B SaaS that flattens at 85% is healthy.
The smile curve. Retention drops, flattens, and then starts increasing. This is rare and extremely valuable. It means users who initially churned are coming back — often because the product reached critical mass (more friends joined), because a feature they wanted was finally shipped, or because a seasonal trigger re-engaged them. Meesho’s retention curve in tier-2 Indian cities showed this pattern when social commerce reached density — resellers who dropped off came back when enough of their buyer network was active.
Using retention curves to make decisions
Retention curve benchmarks: A healthy consumer app in India flattens at 8-15% by month 3. If your curve has not flattened by month 3, you do not have product-market fit for the cohort you are acquiring. See PM Benchmarks for benchmarks by product type.
The correct response is not “let’s improve retention with push notifications.” The correct response is to talk to the users who did retain, understand what makes them different, and either build for that segment or find a way to move more users into that behavioral pattern.
If your curve flattens but at a low number (say 5%), you have a niche product. Either you are okay with that niche (and your unit economics work), or you need to broaden the value proposition to retain a larger share.
If your curves are flattening at different levels for different cohorts, you are learning. The cohort that retains best tells you what is working. Compare their behaviors, their acquisition channels, their onboarding paths, and replicate what makes them sticky.
Pick a product you work on (or a product you use daily). Build a retention table using whatever data you have access to — even rough estimates work for this exercise.
- Define your cohort: users who signed up in a specific month.
- Define “active”: what action counts as the user being retained? Be specific — not “opened the app” but “completed at least one core action.”
- Fill in retention at month 0 (always 100%), month 1, month 2, month 3, month 6 if you have the data.
- Plot it. Does it flatten? Where? At what percentage?
- Now segment: split the cohort by acquisition channel or by a key behavior in week 1. Do the curves look different?
The gap between your best-performing segment and your worst-performing segment is your growth opportunity. Narrowing that gap — by fixing onboarding, matching users to the right entry point, or cutting unprofitable acquisition channels — is higher-impact work than any top-of-funnel campaign.
Putting it together: the growth analytics operating system
These four tools — funnels, cohorts, growth loops, retention curves — are not separate analyses you run once. They are a system you operate continuously.
Weekly: Review the funnel. Check conversion at each step. Compare to last week. If any step degraded, investigate immediately.
Bi-weekly: Pull cohort retention. Compare the latest cohort to the one before it. Is the product getting better for new users or worse? If worse, trace back to what changed — new acquisition channel, product update, onboarding change.
Monthly: Evaluate your growth loop metrics. If you have a content loop, measure content creation rate and organic traffic from user-generated content. If you have a viral loop, measure K-factor (invites sent x conversion rate). If you have a paid loop, check LTV-to-CAC ratio by channel.
Quarterly: Plot the retention curves for the last four cohorts on the same chart. The shape of those curves, and whether they are improving or degrading over time, is the single best indicator of whether your product is on the right trajectory.
The PM who understands these four tools does not need to memorize frameworks or chase dashboard numbers. They can decompose any growth problem into its components, identify the binding constraint, and propose a fix that is grounded in what the data actually shows.
Test yourself
You are the growth PM at a food delivery app operating in tier-2 Indian cities. The CEO shows you a chart: MAU has grown 40% over the past quarter. But the board is concerned because burn rate has also increased 60%. The CEO asks you to figure out whether the growth is healthy.
You have access to the full analytics stack. Where do you start your investigation?
your path
Next steps:
- Metrics That Matter — the frameworks for choosing which metrics to track in the first place
- Experimentation — how to run A/B tests on the funnel and retention improvements you identify
- Measuring Outcomes — connecting growth metrics back to business results
Meesho's PM is reviewing Q2 growth analytics. New seller GMV is up 31%, but repeat seller GMV—sellers who have been active for more than 90 days—is flat. The CEO wants to celebrate the new seller growth in the all-hands. The PM thinks this is a warning sign.
The call: How do you frame this for the all-hands, and what does the data actually tell you?
Meesho's PM is reviewing Q2 growth analytics. New seller GMV is up 31%, but repeat seller GMV—sellers who have been active for more than 90 days—is flat. The CEO wants to celebrate the new seller growth in the all-hands. The PM thinks this is a warning sign.
The call: How do you frame this for the all-hands, and what does the data actually tell you?