Customer churn analysis by onboarding and segment

Customers who skipped onboarding canceled three times as often, but that doesn't prove why.

The company and data are fictional.

Quillboard's customers at 90 days, one dot each. Onboarded means they created a project and invited a teammate in their first 14 days.
  • Not onboarded 16 of 27

  • Onboarded 6 of 31

  • Canceled by day 89
  • Still subscribed at day 90

Where should retention effort go?

Lena Fischer leads customer success at Quillboard, which sells project software to agencies and consultancies. She has one quarter of retention effort to place, and everyone blames plan or industry.

A customer churn analysis compares customers who canceled, or churned, with those who stayed, over a set number of days.

The churn dashboard

This is Lena's Vista dashboard. Which chart looks different? Then set Churned within to 30 and 60 days, and watch By onboarding and By plan.

Skip the interactive dashboard

Loading the interactive dashboard. The figures below don't depend on it.The interactive dashboard didn't load. The figures below don't depend on it.

Why not a plan or industry campaign?

Plan, size and industry barely change who leaves: at 90 days each splits churn by about 10 points. Onboarding splits it by 40. The plan that churns more also flips with the window.

Most who skip onboarding and leave go by day 60: 15 of 16. So anything Lena tries has to land before day 60.

Who had canceled by each point, and the plan with the higher churn.

WithinNot onboardedOnboardedHigher-churn plan
30 days6 of 271 of 31Team 19%
60 days15 of 272 of 31Studio 35%
90 days16 of 276 of 31Studio 42%

How to run a churn analysis

Lena builds a Strata flow from three exports: subscriptions, customer records and product activity. Follow cus_1003, who canceled after 16 days.

  1. Bring in plan, size and industry.

    A Join matches each subscription to a customer record. One has no match and drops out.

    Join

    cus_1003 has a match.

    SubscriptionsRows
    Exported64 subscriptions
    With a customer record63 customers
  2. Keep customers with no activity record.

    A second Join sets them aside and a Union brings them back. Lose them and churn looks healthier than it is.

    Join, Union

    cus_1003 has no activity record, and stays.

    Overall churnRate
    With them38%
    Without them33%
  3. Decide who counts, and mark onboarding.

    Everyone needs 90 days of history and must still be there at day 14, because the comparison is about those 14 days.

    Formula, Filter

    cus_1003 counts, as not onboarded.

    CustomersCount
    Too recent to judge3
    Left before day 142
    Left to compare58 customers
  4. Count who canceled in the window.

    A Formula reads Churned within from the dashboard.

    Formula, Vista Source

    cus_1003 counts at every setting.

    Churned withinCanceled
    30 days7
    90 days22

Lena tests before she spends

Engaged customers may both onboard and stay, and this data can't separate them. The gap also flips if the onboarding deadline moves to 10 days, as How we built it shows. So Lena doesn't announce a cause or commit a budget.

The test is the teammate invite. Customers who started a project but invited no one canceled at 60%, about as often as those who did nothing.

Who does what. At best, onboarding saves about 11 customers.

WhoDoes what
Customer successTalks to five people from each group
ProductTests a simpler teammate invite in the first 14 days, with a comparison group

Before retention, check where they came from. Sam's CAC by channel is the same company, on a separate sample.

How we built it

Subscriptions

64 subscriptions, with fields found in a Stripe subscription.

Required columns: idcustomerplan_nicknamestart_datecanceled_at

Customers

63 customers from your CRM.

Required columns: customercompany_sizeindustry

Activation

50 activation records from product analytics.

Required columns: customerfirst_project_atteammate_invited_at

See the steps run

Select the Join with activation and open Output Left: 13 customers with no record, on their way back through the Union. 12 reach the comparison; the rules exclude the other 1.

Skip the interactive flow

Loading the interactive flow. The steps above don't depend on it.The interactive flow didn't load. The steps above don't depend on it.

Left to right: the three exports, two Joins, the Union, the rules and Filter, then the window Formula the dashboard controls.

The formulas in this flow

ColumnExpression
OnboardingCASE WHEN CAST(first_project_at AS DATE) < start_date + INTERVAL 14 DAY AND CAST(teammate_invited_at AS DATE) < start_date + INTERVAL 14 DAY THEN 'Complete' ELSE 'Incomplete' END
ChurnedCASE WHEN CAST(canceled_at AS DATE) < start_date + params.window * INTERVAL 1 DAY THEN 1 ELSE 0 END
Churned, as a percentchurned * 100.0

Highest and lowest churn in each comparison at 90 days. A point is the plain difference between two percentages.

ComparisonHighestLowestGap in points
OnboardingNot onboarded 59%Onboarded 19%39.9
PlanStudio 42%Team 33%8.6
Company size11-50 44%1-10 33%10.4
IndustryAgency 42%Consulting 32%10.4

The 14-day line was set in advance

The gap depends on where the onboarding line falls. Move the deadline to 10 days and it reverses. Try enough lines and one will tell the story you wanted, so Lena wrote down 14 days before she looked.

The same customers at 90 days. Only the deadline moves.

Onboarded withinNot onboardedOnboarded
7 days21 of 53, 40%1 of 5, 20%
10 days16 of 45, 36%6 of 13, 46%
14 days16 of 27, 59%6 of 31, 19%
21 days16 of 25, 64%6 of 33, 18%

What the data can't show

Groups of 16 to 33 are too small to rule out even a 30-point gap, so plan, size and industry aren't cleared either. The onboarding gap has a rough 95% range of 17 to 63 points on 58 customers.

If the whole gap were cause, about 11 of the 27 customers would have stayed, roughly 5 to 17.

At 30 days there are only 7 cancellations, so read nothing into the plan chart at that setting.

Onboarded here equals invited a teammate. Customers who started nothing canceled at 58%.

Count from day zero and the 2 early leavers land in the not-onboarded group, which lifts its churn from 59% to 62%.

The method doesn't suit annual contracts, where nobody can leave at day 60, or churn from failed payments. Don't mix up canceled_at and ended_at: one is when they asked to leave, and the other can be months later.

Definitions

Churned
Canceled from day 14 up to the end of the window, counted from the subscription start.
Onboarded
Created a project and invited a teammate in the first 14 days.
Window
How many days after starting we count.
Compared customers
Those with 90 days of history who didn't cancel in the first 14.
How it ties out
16 without onboarding and 6 with it make 22 cancellations. Kept out: 3 customers with under 90 days, 2 customers canceled in the first 14 days and 1 subscription with no customer record.

FAQ

What is a customer churn analysis?
A comparison of customers who canceled with those who stayed, over a fixed window, to see what the leavers have in common. It suggests what to test next.
How do you calculate churn rate?
Divide the customers who canceled in a period by those who could have canceled in it. Saying exactly who could have canceled is most of the work.
How many customers do I need?
There is no clean number, but with a few dozen you can only trust large gaps, and only for one or two factors. Test any finding on a new cohort before you act on it widely.
Why exclude people who cancel in the first 14 days?
Because the comparison is about what people did in those 14 days. Anyone who left first never had the chance, so counting them blames the wrong thing.

Build this comparison from your exports

Test your own first two weeks. Local runs are free.