Exit surveys that people actually answer: one plain question beats a grid

People who are leaving have no reason to give you ten minutes. They are leaving.

SurveyMonkey's guide to exit survey completion lists three reasons people quit partway: the survey is longer than they were told, it asks for information the company already has, and they doubt their answers are confidential (SurveyMonkey). That guide is written for employee exits, but the reasons carry over to anyone who is leaving a product or a service. It also advises stating the real length up front and putting the question you most need on the first page.

The cheapest fix is the same every time. Cut the survey until it is short enough that nobody wonders how long it will take.

A grid is a table of statements with a row of bubbles for each. It looks efficient. It also invites people to click down one column to get to the end.

Milieu tested this by randomly splitting people between two versions of a 20-statement agreement question. In the single-question version, 4% gave the same answer to every statement. In the grid version, 10% did (Milieu). The grid group had 332 people and the single-question group had 512, and Milieu is a survey company reporting on its own test, so read it as one data point.

Academic work points the same direction. A 2013 experiment by Couper and colleagues found that splitting a grid into smaller grids reduced missing data, though not in every version they tried (Couper et al.). The samples were opt-in web panels, so it doesn't prove how people leaving your product will behave.

The pattern is that the bigger the grid, the more people just get through it. That is a bad fit for an exit survey, where an honest answer is the whole point.

Ask one question, in words, about what happened:

• "What was going on when you decided to cancel?"
• "What made you stop using it?"
• "What is one thing that would have changed your mind?"

Pick one. Each asks about a moment, which is easier to answer than an opinion about the product.

Keep it short on purpose. Pew Research Center looked at open questions across more than 40 of its panel surveys. Closed questions were skipped 1% to 2% of the time. Open questions averaged about 18%, and questions that asked for multiple sentences were skipped more often (18% median) than ones that asked for a word or a phrase (13% median) (Pew). Pew's panel is its own, so the exact numbers will differ for you. The direction is useful: a written answer costs more than a click, so ask for a line, not an essay.

You will lose some people either way. A reply from a fifth of the people who leave, in their own words, still tells you more than a full grid answered by clicking down one column.

Where the person already is. Put it on the cancel screen or the last page of the flow, before they confirm, so they don't have to open an email later.

Response numbers from vendors vary, so treat them as rough. Userpilot, citing Refiner's 2025 report on more than 1,300 in-app surveys, gives about 27.5% for in-app surveys on web apps, and cites Delighted's own benchmark of around 15% for email surveys (Userpilot). Both come from companies that sell survey tools.

If you can only reach people by email, send it from a real person's address, say how long it takes, and make the reply one question long.

One-line replies are the easy case. You can label them quickly because each one says one thing.

1. Read all of them once without labelling.
2. Give each a short label for what happened: "price rose at renewal", not "pricing".
3. Count the labels. A reply with two reasons gets two labels.
4. Read every reply under the top three labels again, and split any label that turned out to be two things.

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

Say 40 people answered "What was going on when you decided to cancel?"

"Stopped needing it" is the biggest count, but it is also the vaguest. Reading those 13 replies again might show that 9 of them stopped after a project ended and 4 never got started. Those need different responses. The first is normal and you can ask them to come back. The second is a setup problem you can fix.

This is the step a grid never gives you. A grid can tell you that ease of use scored low. Words tell you which part of the setup stopped them.

Enform does this counting at this step: it reads every reply, counts what people said, and every count opens to the replies behind it, so rereading the top counts takes one click.

There are three honest ways to do this.

• By hand in a spreadsheet. Paste the replies into Google Sheets and label them yourself. It costs nothing and works well under about 100 replies. It gets slow as the replies grow.
• A general AI chat such as ChatGPT. Fast for a first set of labels. Give each reply an ID so you can check the labels against the text, and still read the replies under the counts you act on. Our post on whether ChatGPT can analyze survey responses covers what to check.
• A survey tool. Google Forms and Typeform both collect the replies. See how Enform compares on the Google Forms and Typeform pages.

Enform fits when you want the survey written from a description, one link to share, and the reading and counting done for you. It doesn't fit if you need the survey embedded in your cancel flow with your own code. If you want a ready question set, start from the customer churn survey template, and see pricing for what is included.