Churn rate analysis: find where losses changed

Churn rate analysis identifies which customer groups explain a change in departures and whether the change comes from customer mix or weaker retention within those groups. Begin with a loss count that reconciles to the starting customer base. Then investigate the largest contributions to the change before choosing an intervention.

Keep the definition fixed

Use the same customer unit, departure rule, and observation period in both comparisons. The renewal, retention, and churn definitions own the denominator question; the churn glossary gives the concise definition. This workflow investigates movement after that definition has been agreed.

For the illustrative US service company below, churn means a starting customer ends paid service during the month. June 2025 and June 2026 are measured separately using their own opening customer rosters. The customer base can change throughout the intervening year; these are independent monthly snapshots. New customers joining within either June are outside that month's starting cohort.

Export one row per account and month: A period label, B account ID, C segment at period start, D starting-active flag, and E lost-during-period flag. Add start date, product, acquisition cohort, and cancellation reason as further columns. Freeze the segment at the start so cancellation does not move customers into a different group retrospectively.

Deduplicate account IDs within each period. Check account merges, reactivations, missing statuses, and customers who buy multiple products before counting customers as distinct departures. For a nonsubscription business, use the separate ledger-based churn method to establish the inactivity rule first.

Build a table that reconciles

The company consistently uses two nonoverlapping tenure segments: new and established. Every starting account belongs to exactly one segment.

June 2025 losses reconcile as 20 + 16 = 36; June 2026 losses reconcile as 48 + 12 = 60. Both starting bases total 1,000. June 2026 has 24 more departures, and the overall rate rises by 6.0% minus 3.6% = 2.4 percentage points.

Set the segment label in H2 and the period label, such as "June 2025" or "June 2026", in I1. In a spreadsheet containing the exported columns described above, count starting customers with:

=COUNTIFS($A$2:$A$2001,I$1,$C$2:$C$2001,$H2,$D$2:$D$2001,1)

Count losses from that same starting population with:

=COUNTIFS($A$2:$A$2001,I$1,$C$2:$C$2001,$H2,$D$2:$D$2001,1,$E$2:$E$2001,1)

Divide the second result by the first; show an unavailable rate when the starting count is zero. Microsoft's COUNTIFS documentation confirms that the function counts rows meeting all the supplied criteria. Covirage is not affiliated with Microsoft.

Separate customer mix from rate changes

The new segment produces 48 minus 20 = 28 additional losses, while established customers produce 12 minus 16 = four fewer. That explains the net increase of 24, but it does not explain why it happened.

The number of new starting customers doubled between the two June snapshots. Apply June 2025 segment rates to June 2026 customer counts to hold retention behavior constant:

The calculation is 400 × 10.0% + 600 × 2.0% = 52 losses. At unchanged segment rates, the June 2026 customer mix would therefore produce 52 / 1,000 = 5.2% churn.

The total change splits into:

Mix effect = Held-rate June 2026 losses - June 2025 losses = 52 - 36 = 16

Within-segment rate effect = Actual June 2026 losses - Held-rate June 2026 losses = 60 - 52 = 8

The reconciliation is 16 + 8 = 24 additional departures. With equal starting bases, those effects also equal 1.6 and 0.8 percentage points, which sum to the 2.4-point overall rise. Percentages throughout are rounded to one decimal place.

This is an arithmetic attribution using June 2025 rates as the reference, not a causal conclusion. With different total customer counts, separate the effect of base size as well; do not call every increase in expected losses a mix effect.

Investigate the group that deteriorated

The next review belongs with the owner of onboarding and the manager responsible for the new-customer experience. The eight losses above the held-rate calculation are the signal to investigate. The larger new-customer base also creates more expected departures even when its rate is unchanged.

Split the new segment by product or acquisition cohort, using the same starting roster. Compare customers at the same tenure and with equally completed observation windows. Review the cancellation evidence behind any concentrated change before deciding whether service, pricing, customer fit, or another factor warrants action.

Keep dimension tables separate: adding product and tenure totals counts customers repeatedly. Small groups also produce unstable percentages: NIST's proportion-interval guidance explains why uncertainty needs attention when sample sizes or event counts are small.

Use churn reduction for the intervention and measurement workflow. If the investigation concerns supplied scores rather than observed loss drivers, use churn rate prediction to compare predictions with completed outcomes.

Where it goes wrong

• Segment totals do not tie. Missing labels or overlapping segments can make the driver table disagree with the company loss count.
• Rates are averaged equally. Total churn divides total departures by total starting customers, rather than averaging segment percentages.
• Tenure windows differ. Recent customers can look safer simply because they have had less time to leave.
• Labels change after departure. End-of-period segmentation can obscure which starting group actually lost the customer.
• Association becomes a cause. A product or source associated with departures needs investigation before it justifies a policy change.

Analyze your customer losses in Covirage

Bring your starting rosters, dated departures, and customer attributes as files or a Google Sheet. Covirage maps the columns; its tools compute and check the cohort comparisons, and the external AI model explains the results with every figure traced to its rows. Create a free Covirage account and bring a sample of your own export.

See renewal watch for the supporting business context.