Reflections on Two Years of Solving Churn

If you’re working on churn or retention, this resource was built for you.

It distills everything I’ve learned over the past two years from pattern-matching data across companies like Notion, Calendly, Superhuman, and many others.

There are three places where I’ve built my churn muscle:

• A churn software: it probably has the largest churn dataset on the internet, with billions of dollars worth of protected revenue.

• My experience: solving churn at Streamline and with several consulting clients.

• Our partners: unicorns and startups alike that trust us to improve their retention.

Together, these experiences have given me insights on churn that you won’t find anywhere else on the internet today.

👇 Raj, who is the VP of Product at Mozilla said this:

---

The Argument Against Churn

The first thing to remember is that churn is solvable, at least to a certain extent.

A lot of it depends on your product and target audience, of course. But a lot of it is still within arm’s reach. Some things work for all businesses. Some things only work for a few.

But, churn is solvable.

If it were not, companies would not be building retention-focused products either.

A lot of times, we feel that churn is not possible to solve. I would like you to read this article with the mindset that it can be solved.

The Basics

The churn and retention world comes with plenty of jargon, and for good reason.

Take a look at the terms and try to fill in any gaps you might have. Just like a good yoga or pilates class starts with theory, this one will too.

Then, we'll move on to tactics.

Tip: If you can’t finish the article in one go, bookmark it so it’s easy to find later.

1. Voluntary churn:

This is the kind of churn where people raise their hand to quit. They may click on the "cancel subscription" button, ask customer support to refund their last payment, or cancel the subscription via Google Pay or their bank without interacting with your software. Voluntary churn typically makes up around 60% of total churn.

2. Involuntary churn:

This is the second type of churn. It happens when people quit silently and unknowingly. For example, a card declines because the bank returns a "do not honor" error. Sometimes the card expires and the customer needs to update it. This makes up a large portion of your churn numbers, but you will not see a breakdown of this in most platforms. The easiest way to calculate it is to subtract voluntary churn from total churn.

3. Revenue churn:

Revenue churn measures churn by looking at the amount of money lost. For example, if you lost $1M this month and $800k last month, that is your revenue churn.

4. Subscriber churn:

This measures churn by counting customers lost. Every customer is treated equally, whether they are large or small.

Measuring both is useful. It helps answer whether you are losing a few big customers or many small ones, so you can decide what to fix.

If you lose one large customer, the strategy you need is different. A simple bank retry will not be enough.

These four metrics help you understand who you are losing and why.

5. Retention cohort chart

This is a chart that plots subscribers by their cohorts and months. There are typically four ways to read it.

• Horizontal: how does a cohort perform over time compared to other cohorts?

• Vertical: how do cohorts perform in their starting month compared to another month?

• Diagonal and outliers

5. Retention Curves

The above chart is typically complex to read, so it is converted into a curve chart.

The goal for most businesses is to have a smile curve. The worst is the declining chart.

6. Layered cake chart

A cake chart looks something like this. It is those smile curves stacked together. And what you want to see is this growing. You can see how each cohort contributes to total revenue over a period of time.

When it does not, it looks like this.

You can see how each cohort contributes to total revenue over a period of time.

As Elena Verna shares, "The main point is that after 6 or 7 years in business, as much as 80% of revenue comes from existing accounts."

A cake chart can demonstrate this well. You can break these charts down further (for example, by geography) to figure out if a certain market retains better than another. If so, you can prioritize retention efforts in the weaker market.

7. Net Revenue Retention

This is where things get a bit more complex. Net revenue retention takes into account expansion and contraction revenue.

So if you had an MRR of $100 at the start, of which $10 churned, but you expanded existing clients by $5 and downgraded existing clients by $2, your net revenue retention would be:

= (100 - 10 + 5 - 2)/100

= 93%

Net basically accounts for everything.

8. Gross Revenue Retention

Gross does not account for everything. It is stricter. It ignores upsells but punishes you for any contractions. With the above example, your gross revenue retention is 88%.

Typically, when people refer to Gross and Net churn, they refer to revenue churn, not subscriber churn.

9. Net Subscriber/Logo Retention

This measures customers lost, upgraded, and downgraded.

10. Gross Subscriber/Logo Retention

The same idea as above, but stricter. It accounts for customers lost and downgraded but ignores upsells. It keeps you humble, honestly. But it is typically not used for investor reporting where you want to look good.

11. Hard vs Soft Declines

Ok, do you remember we covered involuntary churn a little while ago? There are actually two sub-parts of involuntary churn.

Hard decline is when the card permanently declines. For example, the card expired, was stolen, or was blocked by the issuer.

Soft decline is when the issue is temporary. The payment failed but the bank tells you that it is temporary, so you can retry later. An example is "insufficient funds."

There are lots of decline codes. Some banks do not return correct codes either and may put it under "do not honor."

It involves the entire financial infrastructure, so it is no surprise that things can get complex.

For now, knowing that involuntary churn happens due to hard and soft declines is helpful.

You can measure this by looking at the failed payment codes returned by your payment provider. Export all failed payments with their codes, then group them by error code.

You can google whether the code is a hard decline or a soft decline.

Break things down by geography (Japan, India, USA), card issuer (Visa, Amex), and other details if you want to see interesting patterns in your data.

12. Monthly vs annual churn

This is where things get interesting. Did you know that a 5% monthly churn equals 46% annual churn? And at 7%, you lose closer to 75% of all customers annually?

Looking at monthly churn numbers can be relieving, but converting them to annual churn paints a realistic and honest picture of where you are. It will also be more motivating.

And if you reduce monthly churn by just 1%, you could improve acquisition by around 10%. It is a good lever to work on, provided you can make it work.

13. First term-churn

This is the number of people who cancel after the first billing interval. For example, a monthly subscriber pays and cancels immediately, or cancels before their billing cycle renews.

This is your first-term churn. It is not ARR until it recurs.

14. Expansion MRR

Revenue growth from existing subscribers.

15. Contraction MRR

Revenue lost from existing subscribers who downgraded to a cheaper paid plan.

16. Customer Health Score

This grades each customer as either at risk or healthy. A simple way to calculate this is to look at product engagement data. Have they used the features? Telecom companies use hundreds of datapoints, such as whether a customer is on a family plan, to decide whether someone is at risk or healthy.

Knowing the health score of each customer is what helps build a great retention strategy. Churn is a lagging indicator, so having a health score brings directional data closer. This allows you to move faster and test things.

There are many data science projects on Kaggle and videos on YouTube that show you how to calculate this.

17. Time to churn

The average time from signup to cancellation. This is what you need to beat. If you increase LTV by even one month, what is the impact on the bottom line?

17. Reactivation rate

What percentage of users cancel and then come back? Creating a cohort chart is more helpful than a single-point chart.

You could go deeper into this and breakdown using any of the above methods (like churn reason, subscription tier, churn timing - first time vs returning customer)

If you made an offer such as a discount at the point of cancellation, you would want to measure:

• second churn rate (how many people churn again)

• time to second churn

• reactivation discount dependency (how many stay after the discount ends, so you can create a discounting strategy)

I'm assuming that you'll measure all of the obvious retention metrics like DAU, WAU, MAU, feature adoption rates, that are required for your product.

---

Creating a churn reduction strategy

The first thing to do is understand your goals.

• How soon do you want to move the needle?

• How many resources do you have?

• What sort of buy-in do you have from the rest of the company?

• What kind of retention results would be a stretch goal for you?

You'd have to look at the data to be able make a goal. Being able to reduce churn by 20% is typically a good goal to have. But it depends on the business scale and existing investment. You can DM me if you'd like some directional advice.

---

Deploying churn reduction tactics

Not all tactics will help you. There have been times when I have run marketing projects that had no real impact, and other times when they completely changed the trajectory of the company.

That is the nature of the job. We are not surgeons. People do not expect perfection each time. If you do hit home runs every time, the risk is that you are not taking big bets.

Simple Cancel Flow

You can create advanced cancel flows but it doesn't take a lot to get started. Stripe lets you redirect users to a custom link on cancellation. This is what OpenAI uses for feedback collection alongside Qualtrics.

Failed Payment Retries

When cards fail due to reasons like insufficient funds, you can simply retry the card without emailing the customer.

Mastercard and Visa have some limits how often you can retry.

Retries are usually very impactful, and require zero engineering lift to turn on.

Cluster Feedback

Dump the CSV in any AI tool and ask it to cluster (example prompts). Assume some inaccuracies.

Usage Data on Cancellation

Show what users will lose at point of cancellation. Canva does this well showing how many pro features the user has used.

Block Features

Sometimes the buyer and user of the product are two different people. Display in-app paywalls with card-updater on the same page. Offer grace periods instead of revoking immediately.

Bitly shows this, but it's actually quite buggy.

You can a pause wall if the user has paused the subscription.

Framer redirects to Stripe.

Toggle On Auto-Renew

Users turn off auto-renew. Find creative ways to encourage them again (read Elena's article). Some companies roll over credits if you don't let your subscription cancel. Many companies will show a banner at the top, which can be a bit intrusive.

Trial Extensions

If you detect low usage or a drop off after a trial, extend it automatically. If it's a paid trial, offer an extension at point of churn.

If it's a free trial, consider other ways.

Friction Logs

Run friction logging with new hires to capture points of friction/errors.

Ben Williams shares how to effectively run it. You could hire folks to run it or ask new hires to do so.

I ran one for Dropbox that their team found helpful.

Refund Offers

When people ask for refunds due to budget reasons, make an offer.

Chess.com does this when students churn.

Multiple Senders

Send dunning emails via different senders to maximize deliverability and open rates. Prevents email threading.

Grandfathered Users

Show grandfathered users or folks on a discounted plan the dollar values of what they would lose if they cancel. People often appreciate a disclaimer and will retain.

Segmented Cancel Flows

Segment cancel flows based on user's age, tenure, region, plan etc. A/B test and make different offer types to identify best for each segment.

Abandoned Cancellations

Monitor people that visit the cancellation page but don't cancel. They're either dissatisfied if tenure is long or want to see how easy it is to cancel if they perform this within the first few hours of signing up.

Confirmation Step

Add a confirmation step, "Are you sure?" instead of immediately cancelling. This goes a long way and people often change their minds. Comply with laws like FTC's click to cancel so your flows aren't annoying to cancel.

Pauses

Let people pause their subscription and pick a custom pause length. Test different lengths to see what maximizes revenue for you. Pauses extend LTV by ~5.5 months (source).

Expiring Card Reminders

Do you send emails to users whose cards are about to expire? If so, test without it. Card tokenization has made it a thing of the past in some regions and can hurt more than it can help.

Localized Cancel Flows

If you have an international audience, localize the cancel flow in their language. Your brand marketers will love it too. Sometimes, required by law.

Add a Sunk Cost

Give users something that prevents a cancellation. Like unused credits roll over if they stay subscribed. Having a sunk cost prevents cancellation.

Map Feedback to MRR Lost

Continuously map feedback to MRR lost so you can prioritize feedback by revenue whether it's feature gaps, bugs, or competitors.

Churn Risk Model

Run a churn risk model. Train it based on dozens of variables (tenure, LTV, features used) so that the model returns whether a user is likely to churn or stay. View tutorial, research paper.

Hidden Plans

If users don't use all your features, let them downgrade to a lite plan at the point of cancellation. No need to clutter up your pricing page with this.

Canva, Adobe, Spotify all offer daily, weekly, monthly and yearly plans in-app but not on their pricing page.

Skip Logins

Dunning emails and SMS links should take people to the card updater page. Without forcing a login. You're fighting procrastination, more so with consumers, so each step is a drop off point.

Charge Early

Charge 30 days prior to renewal for yearly renewals (if it makes sense for your business). Website builders do this.

Multi-Year Subscriptions

Offer longer than 1 year subscriptions. Domain registrars offer 2-3 year contracts because they know it's an impulse purchase most of the time.

Prorated Refunds if Underused

Automatically reduce invoice if product is underused. Slack doesn't charge for inactive users and it proved well. Some companies that cater to seasonal businesses auto discount subscriptions for a portion of the year.

Walmart was about to churn from Slack because managing access was difficult. That's what led Slack to devise a customer-friendly billing policy.

“I’m a huge believer in less friction. Of all the places you can remove the friction, the purchase decision would be a big one.” — Stewart Butterfield

Reactivation Campaigns

For products with an on-and-off use case or a low switching cost, reactivation campaigns help. Make a great offer and reduce friction best you can (eg, no logins required). Typically hard to pull off.

Assemble Trigger Moments

Remind people that they need to use you, outside of your product. Oreo used Got Milk and the entire desserts range.

Force Network Effects

If your product has some sort of a network effect, don't gate it. If it doesn't, try to add one creatively. Eg, sharing your daily workout wasn't done before Strava decided to make it cool.

Gifts

Swag retains. Hard to scale. Tools: loopandtie or ongoody.com but you can also make it artsy like Figma.

Transform A Drop Off Point

If your free tier has a limit that causes people to drop off (eg, Dropbox' 2GB storage), let them raise this limit by helping you. Like a referral program.

Secondary Products

If your core product has natural churn, layer on a secondary product. People use Zillow to buy homes. You don't purchase homes often. To offset that, Zillow launched Zillow Zestimate to increase retention.

New Products

If your core product has natural churn and a definite day when they don't need you anymore, get users to switch. Bumble launched Bumble for Work and Friends. Consider using a different brand though.

Login Method Reminder

Show last used login method so people can quickly remember which method they used to sign up. Example can be found here.

[more tactics coming soon]

Also, a huge thank you to everyone who said nice things about this article. I am beyond grateful.

Sapph Yip who was kind enough to tell me that Mind the Product featured in their newsletter that goes out to 150k people. Mind the Product is a communuity for PMs that picked it up organically, and I only found out after they published it. I'm super impressed with their curation muscle.

Aman Kumar (Growth at Rippling) went all in.

Taylor O'Brien is a product leader in the monetization space and led Involuntary Churn @ HBO Max. He focused on prevention + recovery of subscription payment failures. He's got a rich portfolio of work on his personal site.

Marc Sirkin who runs a boutique consulting shop and has done leadership roles at SEMRush, PwC, Microsoft amongst others.

Adam Sanghera (Software Engineer). He's got a neat personal site too :)

Natalie A. Thomas, who is a Research and Strategy Director. Having a designer say good things about a churn article is a high honor!

Also, my incredible marketing professor (Sourjo Mukherjee) from 6 years ago at Essec Business School.

Steven Macdonald is the founder at OKRs Tool, and thinks this is best article he's ever read on churn. Wild.

We've got Pedro, Ruhi, Jolene, and Fyodor too!

Chris Todd (Marketing Ops, Strategy) and Kerry Campion (PMM), who are in Slack, ready to upskill themselves.

Susan does Growth Product Marketing at Google and holds a host of different titles. She shared it with her audience and I can't thank her enough for the vote of confidence.

And so did Devprakash, who is a UX designer out of Germany.

P.S - If you found this article via word of mouth, please tell me who shared it so I can properly thank them :)