Home / Guides

How to do a churn analysis that changes what your team does

By Draft First, October 2026

Most churn analyses end as a pie chart: 40% "budget", 25% "product fit", 20% "other". Nobody changes anything, because "budget" is usually the polite reason, not the real one. A useful churn analysis asks a different question: what could we have seen earlier, and what will we do differently?

Step 1: Gather the lost accounts

Pull every account that cancelled or shrank meaningfully in the last two quarters. For each one, note the size, how long they were a customer, the owner, and the reason recorded at the time.

Step 2: Find the real reason

For each account, read the last six months of tickets, emails and call notes. Ask:

• When did things first start going wrong?
• Did the champion or sponsor change?
• Was there an unresolved issue or broken promise?
• Did they ever reach the outcome they bought for?

If you can, ask the customer. A short, honest exit call often tells you more than six months of data. There's a template for that in churn risk email templates.

Step 3: Look for the warning signs you missed

For each lost account, check what your health score said 90 days before they left. Was it Red? Then the problem is response, not detection. Was it Green? Then your score is measuring the wrong things.

Account        Left     Score 90 days before   First real sign
Marlow Retail  Sep 30   Green                  Champion left in May
Finch Foods    Aug 31   Red                    Usage drop in April, no plan made
Coastline      Jul 31   Yellow                 Never reached first value

Step 4: Group into patterns

Look for three or four patterns, not twenty reasons. Common ones: never reached first value, champion left, an unresolved product gap, a silent sponsor.

Step 5: Turn each pattern into one change

• Never reached first value: tighten the onboarding checklist.
• Champion left: add champion changes to your health score.
• Product gap: bring the evidence to product in your voice of the customer report.

Using AI for the reading

For each account in [churned list], read [tickets, emails, call notes]
from the 6 months before cancellation. Give: the first sign of trouble
with a date and source, whether the champion or sponsor changed, any
unresolved issue or promise, and whether they reached their stated goal.
Quote only from sources. Mark anything uncertain [CONFIRM].

You still decide what the patterns mean. The agent just saves you reading a thousand tickets.

Questions

How do you do a churn analysis?

Gather lost accounts, find the real reasons in their history, check what your health score said earlier, group patterns and change one thing per pattern.

What are the most common reasons customers churn?

Never reaching first value, a champion leaving, unresolved product gaps and a disengaged sponsor are common in B2B.

How far back should a churn analysis look?

Two quarters of lost accounts, and six months of history for each account.