How to find out why people sign up and never come back

They tell you how many people left and roughly when. They don't tell you why.

Benchmarks are a weak guide. Amplitude's own growth team wrote that "there aren't any super well-established benchmarks" and that "the best retention benchmark you can use is your app's previous performance" (Amplitude). Mixpanel's 2017 report found that median products kept around 30% of people after the first week, and the figure fell to the low teens by week 8 (HackerNoon). That article was sponsored by Mixpanel and measures weekly retention, so treat it as a rough picture.

The early days matter most. ThriveStack, a vendor, says SaaS sign-ups who don't activate within the first 7 days are 2.5 to 3 times more likely to churn (ThriveStack). The page doesn't name a study for that number, so use it as a hint to look at week one, not as a fact to quote.

Guessing gives you a list of fixes with no way to rank them. Pricing, a confusing first screen, a missing feature and bad timing all look the same in a chart: a line that goes down.

The same idea shows up in email. MailerLite suggests asking "a simple question: why are you leaving?" on its unsubscribe page, with five standard reasons plus an optional "Other" box where people can write their own (MailerLite). Kit's version is a single choice from five options, and it is off by default (Kit). Those are for people who leave on purpose. People who signed up and drifted away never click anything, so you have to go and ask.

Keep it short and send it from a person, not a no-reply address. Pick people who signed up at least a week ago and did little or nothing since.

1. Say who you are in one line. "I'm the person who built [product]. You signed up last month and I noticed you haven't been back."
2. Ask one question about a moment, not an opinion. "What were you hoping to do when you signed up?" is easier to answer than "Why didn't you like it?" and it doesn't assume they disliked anything.
3. Make a one-line reply enough. Say that a few words is fine.
4. Don't attach a link to a long survey. A reply to an email is less work than a form. Our post on survey response rates covers why a short invitation from a real sender does better.
5. Don't offer a discount in the same email. It changes who replies and what they say.

For wording that people actually answer, see how to write open-ended survey questions.

Read all of them once without sorting. Then make a short list of groups from what people wrote, not from what you expected. Two groups usually come up first: people who didn't get what they wanted from the product, and people who wanted it but couldn't work out how to get started. Add groups as the replies call for them, such as "wrong tool for what I needed" or "signed up to look, no plan to use it".

Then label each reply with one group, count them, and reread the replies under the biggest count. The label is a shortcut. The sentences are the evidence. Our post on analyzing open-ended responses walks through the method in more detail.

A worked example

The numbers below are invented to show the method. They are not from a real survey.

Say you sent the email to 120 people and 30 replied. Sorted by hand:

The shares are worked out as replies divided by 30, rounded. The top group says the first screen is the problem, not the feature list. Reading those 11 replies might show that most got stuck at the same step, which tells you exactly what to fix first. With 30 replies, treat the counts as a direction. A gap of a few replies is noise.

Also keep in mind who answered. People who reply are not a random sample of everyone who left, so a count of replies tells you what some people said, not how many of all your signups felt the same way.

There are a few honest ways to do this.

• By hand. Send the email from your own inbox and keep the replies in a spreadsheet. It costs nothing and you read every word. It gets slow at a few hundred replies, and sorting by eye drifts.
• A general AI chat such as ChatGPT. Paste the replies in and ask for groups. It's quick, but a summary can merge two different problems into one, so reread the replies behind each group.
• A survey tool. Google Forms, Typeform and Tally all collect answers. The Google Forms comparison shows where they differ from Enform on reading the text answers.

Enform writes the survey from a description of what you want to ask, you share one link, and it reads every reply and counts what people said. Every count opens to the replies behind it, so you can check a group against the sentences. It's a better fit once replies pile up, and a plain email is enough when you have fewer than twenty. Details are on the pricing page, and the onboarding feedback template is a starting point if you'd rather send a link.