Are users coming back?

The question

Of the people who sign up in a given week, what share create deals again in the weeks after, and is that getting better? Retention needs history: the project needs two completed periods (two weeks, for weekly cohorts) before anything shows, and more is better.

1. Choose the habit

Retention compares two events: one that puts someone into a cohort, and one that counts as coming back.

• Starting event: Signed up. Each person joins the cohort of the week they first did it.
• Returning event: Deal created. Any event would count a single pageview as “coming back”; a real action measures a habit.

Pick the returning event with your team. It is the definition of an active user, and every retention number you quote depends on it.

2. Set the definition

1. Open Retention.
2. Click Edit definition.
3. Set Starting event to Signed up and Returning event to Deal created.
4. Under Count someone as retained if they…, choose Came back in that week. (The stricter Came back every week since drops people for good after one quiet week.)
5. Keep Hide cohorts under 20 people on, so a tiny cohort can’t swing the picture.
6. Press Apply.

3. Read the cards

Keep Weekly cohorts selected. Four cards sit above the grid:

• Week 1 retention and Week 4 retention, averaged across the cohorts old enough to have them.
• Where the curve flattens: the week after which it loses less than 1 point a week. After that, the people left are your core.
• Median cohort size, with the smallest and largest, so you know how much to trust the averages.

4. Read the grid

In the Grid, each row is a cohort: the week it started, how many people, and the share who created a deal in W0, W1, W2 and onwards. The All cohorts row on top blends them. Darker cells mean more people came back.

• Read down a column to see whether things are improving: is W1 higher for recent cohorts than for older ones?
• Read along a row to see how one cohort fades.
• Blank cells are weeks that haven’t happened yet. The current week is left out until it closes, so a young cohort never looks like a drop.

5. Read the curve

Switch the view to Curve. Each cohort becomes a line: the share still creating deals, by week since sign-up.

• A curve that levels off means you have found people who stay.
• A curve that keeps falling towards zero means everyone eventually leaves.
• Newer cohorts’ lines sitting above older ones means retention is improving.

6. See who came back

1. Switch back to Grid and click a W4 cell of a recent cohort.
2. Choose Retained to list who came back, or Not retained for who didn’t.
3. Click a person to open their profile and read what they did.

7. Compare by campaign

1. Click Break down by and choose UTM source.
2. Compare the W1 and W4 columns between sources.

A breakdown splits cohorts by a property of each person’s first starting event, so here by the campaign of the visit in which they signed up. You get the top 5 values, and the rest are grouped as Other. The cards show the highest and lowest values, and the curve draws one line per value.

8. Compare segments

Are people who import contacts in their first days more likely to stay? Make that a segment and compare:

1. Open People › Segments and click New segment.
2. Add did Contacts imported at least 1 time, in the last 90 days. Name it Imported contacts and click Save segment.
3. Back in Retention, click Compare segments, tick Imported contacts, and apply.

A breakdown and a segment comparison replace each other. Retention doesn’t compare to the previous period, so to compare time periods, read newer and older cohort rows side by side.

9. Keep it in view

• On a board, click ＋ Add tile and add a Cohort grid (5, 8 or 12 cohorts) or a Big number for Week-1 retention.
• In Alerts, click New alert and watch Retention · Week 1 retention, so you hear when it leaves its usual range.

Common mistakes

• Any event as the returning event. A single pageview counts as coming back, which flatters the numbers. Use a real action.
• Changing the definition without telling anyone. Every board tile, alert and teammate sees the change at once.
• Reading the newest cohorts’ later weeks. They haven’t happened yet; blank is not zero.
• Averages from tiny cohorts. Check Median cohort size. If cohorts are small, use Monthly cohorts or a more common starting event.
• “Not enough history yet”. The project needs two completed periods. Click Switch to daily cohorts to see something sooner.
• Expecting a link to keep your view. The daily, weekly or monthly choice and Grid or Curve aren’t in the page address, so a shared link opens weekly, as a grid.

Next

Some campaigns bring people who stay, others don’t. Find out which: Where do our best users come from?