Churn rising while every plan improves
A subscription business reports that overall monthly churn rose from 4.1% in Q1 to 4.8% in Q2. The board pack headline reads 'churn worsening'. When you break churn down by plan, every plan improved: Basic 6.5% to 6.2%, Plus 3.2% to 3.0%, Pro 1.6% to 1.5%. Over the same period a heavily discounted acquisition campaign shifted the customer mix: Basic went from 40% to 62% of the customer base, Plus from 35% to 26%, Pro from 25% to 12%. Nothing else changed - pricing, product and support were constant.
1. Explain how overall churn can rise while churn falls in every plan. Show the arithmetic.
Read the worked answer
This is a mix effect (Simpson's paradox). Churn is a weighted average of plan-level rates, and the weights moved sharply toward the highest-churn plan.
- Q1: 0.40 × 6.5 + 0.35 × 3.2 + 0.25 × 1.6 = 2.60 + 1.12 + 0.40 = 4.12%
- Q2: 0.62 × 6.2 + 0.26 × 3.0 + 0.12 × 1.5 = 3.84 + 0.78 + 0.18 = 4.80%
Every plan's rate fell, yet the blended figure rose, because Basic went from 40% to 62% of customers. A weighted average cannot be read as a performance measure when the weights are changing this fast.
The headline 'churn worsening' is therefore wrong as stated. The correct statement is: 'churn per plan improved; blended churn rose because the customer base shifted toward the Basic plan after the discount campaign.'
2. What should the board be told, and which metric should lead the pack?
Read the worked answer
Report both numbers, but lead with the one that reflects performance, with the mix explained alongside.
- Constant-mix churn: apply the Q1 mix to the Q2 rates - 0.40 × 6.2 + 0.35 × 3.0 + 0.25 × 1.5 = 2.48 + 1.05 + 0.375 = 3.9%, down from 4.12%. On a like-for-like base, retention genuinely improved by about 0.2pp.
- Blended churn (4.8%) stays in the pack, because it is what actually happened to revenue-bearing customers, but it is labelled as mix-driven and decomposed.
- Add revenue churn and plan-level absolute losses. The Pro plan shrank from 25% to 12% of customers; if Pro carries the highest revenue per customer, the revenue mix shifted in a far worse direction than the customer mix suggests. Customer-count churn alone can hide this.
The pack should show a small waterfall or bridge: Q1 churn → effect of plan-rate improvement → effect of mix shift → Q2 churn. That one visual makes the sentence defensible and stops the same confusion recurring next quarter.
3. Does this change your view of the discount campaign? What would you check before defending or killing it?
Read the worked answer
It reframes the question. The campaign worked on acquisition but changed the composition of the base, so it cannot be judged on new sign-ups alone.
Checks before deciding:
- Are the new Basic subscribers churning faster than existing Basic subscribers? Plan-level rates are pooled across old and new customers; if the discounted cohort churns at 9% against 5.5% for existing Basic, the campaign is buying short-lived customers and even the plan-level improvement masks it. Build a cohort view by sign-up month.
- Revenue, not headcount. A discounted Basic customer may be worth a fraction of a Pro customer. Compute revenue churn and net revenue retention; blended 4.8% customer churn with falling ARPU is a worse outcome than the headline suggests.
- LTV against acquisition cost. Discounted Basic only pays back if retention over its lifetime covers the discount plus acquisition cost.
- Whether Basic customers upgrade. If they do within a few months, the mix effect is temporary and the campaign is fine; if not, you are structurally shifting to a low-value base.
The right output is not 'churn is bad' but a clear statement: acquisition succeeded, retention per plan improved, and the campaign is worth repeating only if discounted-Basic cohorts retain at a rate that covers their lower price.