Nine Entertainment · All mastheads · FY22–FY24

Pricing the renewal
before the cancellation decision

Around a quarter of subscribers cancelled the moment their promotion ended. A churn-propensity model tiered every subscriber 1–10 and priced their renewal accordingly — until its precision problems forced the right call: abandon model-led targeting for randomised testing.

~1 in 4

of subscribers cancelled the moment their promotion ended

5-figure

additional subscribers retained in the modelled scenario

6-figure

projected additional annual revenue

1–10

churn-propensity score segmenting every subscriber

Role: CRO & Experimentation Lead — retention pricing strategy across all mastheads
Programme scale: 3 risk tiers, D-6 transactional eDM + SMS renewal comms, Starter vs Premium Digital churn patterns
Key decision: Pivoted from a 5–10% precision-limited propensity model to randomised A/B testing on flash-sale cohorts
Status: Modelled at a six-figure annual revenue uplift; retention pricing adopted over acquisition spend

Contents

1. The Problem
2. The Approach
3. Modelled Outcome
4. The Pivot That Matters
5. Key Insights

01The Problem

Promotion-acquired subscribers were not surviving the end of their promotion: around a quarter of subscribers cancelled the moment their promotion ended, and fewer than half remained after their first year. Price sensitivity among promotion-acquired subscribers peaks exactly at renewal — the moment the renewal mechanic meets them with full price.

02The Approach

A churn propensity model scoring each subscriber 1–10, segmenting at-risk subscribers into three risk tiers and treating each differently at renewal:

Risk tier	Score	Renewal treatment
Low	1–4	Full price renewal
Intermediate	5–7	Stepped mid-level discount
High	8–10	Extended, stronger discount to retain longer

Transactional eDM and SMS comms went out at D-6 before renewal carrying the personalised discount, so the offer arrived before the cancellation decision rather than after it.

03Modelled Outcome

If 30–40% of no-engagement and low-engagement subscribers are saved, the model retains a five-figure number of additional subscribers — a projected six-figure uplift in annual revenue.

04The Pivot That Matters

With 5–10% precision issues in the propensity model, the decision was to stop relying on it as a targeting engine and move to randomised A/B testing on flash-sale cohorts instead. A model with a 5–10% precision problem is a bad basis for a pricing decision; randomisation answers the same question without that exposure.

In hindsight, probabilistic scoring should have been validated against a randomised hold-out from the start — and the 50–79% retention discounts, which were priced commercially, should have carried their own randomised structure so the discount depth itself was a tested variable.

05Key Insights

Retention pricing is more cost-effective than acquisition — retaining a subscriber at a 50–79% discount beats losing the full-price revenue entirely.

It also has to be segment-specific: Starter Digital and Premium Digital have genuinely different churn patterns, and a single renewal funnel cannot serve both.