Retention is usually the highest-return place to apply statistics in a business, because a small improvement compounds across every future period. Four techniques do most of the work, and none requires advanced modelling.
1. Cohort analysis - the single most useful technique
Group customers by when they joined, then track each group's retention over time. A simple table: rows are joining months, columns are months since joining, cells are the percentage still active.
Why it beats a headline retention rate: an aggregate churn figure mixes together customers who joined at very different times under different conditions. Cohorts separate them, so you can see whether customers acquired since your pricing change behave differently from earlier ones - which an average will hide completely.
What to look for: where the curve flattens (your durable core), and whether recent cohorts are better or worse than older ones.
2. Segmentation that actually predicts behaviour
Most segmentation is demographic and useless. Segment by behaviour instead:
- Recency, frequency and monetary value (RFM) - simple, powerful, and buildable in a spreadsheet
- Acquisition channel - frequently the strongest predictor of retention, and often ignored
- First-purchase category
- Onboarding completion
Then compare retention across segments. Nearly every business finds one acquisition channel producing customers who churn at double the rate of another - which changes marketing allocation immediately.
3. Measuring churn properly
Three traps to avoid:
- Define it precisely. For subscriptions it is straightforward. For non-contractual businesses, "churned" is a judgement (no purchase in 12 months?) - pick a definition, document it, and never change it silently.
- Use rates, not counts. More cancellations while growing is not necessarily a worsening churn rate.
- Watch the denominator's timing. Churn calculated on end-of-period customers differs from start-of-period; be consistent or your trend is an artefact.
4. Testing retention interventions properly
The most common retention mistake: send a win-back offer to lapsing customers, observe that many return, and credit the campaign. Some of them would have returned anyway - this is regression to the mean plus no control group.
The fix: hold back a random control group who receive nothing. The difference between the groups is the real effect. This feels wasteful and is the only way to know whether you are spending money on customers who were coming back regardless.
What the analysis usually reveals
Three findings recur across businesses:
- Early experience dominates. What happens in the first 30 days predicts long-term retention more than anything later.
- Channels differ enormously in customer quality, and cost per acquisition alone misleads badly as a result.
- A small share of customers generates most of the profit, and they are frequently not the ones the business focuses on.
The commercial arithmetic
Model it: a five-point improvement in annual retention, compounded over the customer lifetime, usually dwarfs an equivalent investment in acquisition. Working that number out for your own business is often the argument that unlocks the budget. Also worth calculating alongside your acquisition costs by channel.
The Statistics for Business course at London School of Business UK covers cohort analysis, segmentation and controlled testing applied to customer data. Enquire today.