Report Sample: Customer Churn Analysis for a UK Subscription Service

Model Answer

Executive Summary

This report presents a churn analysis conducted for Northbridge Digital, a UK-based subscription streaming and media bundle provider, undertaken to identify the drivers of customer attrition and to recommend evidence-based retention interventions. Subscription churn represents one of the most significant threats to recurring-revenue business models, and even modest reductions in monthly churn compound into substantial long-term revenue gains (Ascarza, 2018). Drawing on twelve months of anonymised subscriber data covering 18,400 accounts, the analysis combines cohort retention modelling with a binary logistic regression to isolate the customer characteristics and behaviours most strongly associated with cancellation.

The headline finding is that overall monthly churn averaged 8.9% across the review period, rising to 10.1% in the fourth quarter, a period that coincided with a price increase on the Standard plan. Logistic regression results indicate that low engagement (fewer than four sessions per month), a recent customer service contact, and enrolment on a monthly rather than annual plan were the strongest statistically significant predictors of churn. Conversely, subscribers who redeemed a loyalty discount showed markedly lower attrition, and plan tier itself was not an independent predictor once behavioural factors were controlled for.

Based on these findings, the report recommends four interventions: a proactive outreach programme targeting low-engagement accounts, a revised customer service escalation protocol, an incentivised switch to annual billing, and a redesigned onboarding sequence to lift early engagement. Implemented together as a coordinated programme, these are projected to reduce monthly churn by 1.5 to 2 percentage points within two quarters.

Introduction

The subscription economy has grown rapidly across the United Kingdom over the past decade, spanning streaming media, software, meal kits and retail memberships, and customer retention has become as commercially important as customer acquisition (Kotler and Keller, 2016). For subscription businesses, monthly recurring revenue is only sustainable where churn — the rate at which existing customers cancel — is kept within a level that acquisition spend can offset. Even small changes in churn rate have a disproportionate effect on lifetime value, because attrition compounds month on month (Gallo, 2020).

Northbridge Digital (a fictionalised composite organisation used for this exercise) is a UK subscription media provider offering three pricing tiers — Basic, Standard and Premium — billed either monthly or annually. Over the past year, the business has observed a gradual increase in cancellation rates, prompting a request from the commercial director for an evidence-based investigation into the causes of churn and a set of practical recommendations.

Northbridge Digital’s subscriber base grew from approximately 12,000 to 21,000 accounts over the past two years, driven primarily by a marketing partnership with a UK broadband provider that bundles the Standard tier into selected home internet packages. This growth trajectory means the business is increasingly reliant on retaining bundle-acquired subscribers, whose engagement patterns and price sensitivity may differ from customers who signed up directly, a distinction this report returns to in the Findings and Discussion sections. Understanding whether the rise in churn is being driven by acquisition-channel mix, by service or pricing issues, or by a combination of the two, is essential context for interpreting the regression results that follow, and has a direct bearing on which of the recommendations in this report the commercial team should prioritise.

This report addresses three objectives. First, to quantify churn across the customer base over a twelve-month period, disaggregated by cohort, plan tier and billing frequency. Second, to identify which customer characteristics and behaviours are most strongly associated with the decision to cancel, using a logistic regression model. Third, to translate these findings into prioritised, actionable recommendations that the commercial and customer success teams can implement within the current operating budget.

The report is structured as follows. The Method section outlines the dataset, the definition of churn used, and the analytical approach taken. The Findings section presents cohort retention data, the results of the regression model, and the quarterly churn trend. The Discussion section interprets these results against the wider literature on subscription retention and considers the limitations of the analysis. The Recommendations and Conclusion sections then set out the proposed interventions and next steps.

Method/Approach

This report uses twelve months of anonymised subscriber-level data extracted from Northbridge Digital’s billing and customer relationship management systems, covering 18,400 customer accounts active at some point during the period. Churn is defined as voluntary cancellation of a subscription that is not followed by reactivation within 30 days, consistent with standard subscription analytics practice (Recurly, 2022). Involuntary churn arising from failed payments was excluded from the primary analysis and is noted as a limitation below.

Two complementary analytical techniques were applied. The first is cohort retention analysis, in which customers are grouped by the month they joined and their survival is tracked over subsequent months. This produces a retention curve for each cohort and allows comparison of newer and older cohorts to detect whether retention is improving or worsening over time. Cohort analysis is well suited to subscription businesses because it removes the distorting effect of a growing or shrinking customer base on a simple point-in-time churn percentage (Ascarza, 2018).

The second technique is a binary logistic regression, with the dependent variable coded 1 for customers who churned during the observation window and 0 for those who remained active. Independent variables were selected based on prior literature and data availability, and comprised: plan tier (Basic/Standard/Premium), billing frequency (monthly/annual), monthly session count (a proxy for engagement), whether the customer had contacted customer service in the preceding 60 days, whether the customer had redeemed a loyalty discount, and tenure in months. Continuous variables were checked for multicollinearity using variance inflation factors, and all were comfortably below the conventional threshold of 5, indicating the model is not distorted by correlated predictors (Verbeke et al., 2012).

Descriptive statistics and the cohort retention curves were produced first, to characterise the overall pattern of churn, before the regression model was estimated to isolate independent predictors while controlling for the others. A significance threshold of p < .05 was applied throughout. All data used was pre-anonymised at source and contained no directly identifying fields, meaning no separate ethical approval was required for this internal business analysis; the company and dataset referred to throughout this report are fictionalised for the purposes of this model exercise. The principal limitation of the method is that it relies solely on structured behavioural and transactional data and does not incorporate qualitative reasons for cancellation, such as those that might be captured through an exit survey; this is addressed further in the Discussion.

Data extraction and cleaning were carried out in a spreadsheet-based environment supplemented by a business intelligence tool for cohort visualisation, an approach proportionate to the scale and structure of the dataset used here; a full statistical software package was not required given the relatively modest number of predictor variables. Prior to analysis, the dataset was checked for missing values and duplicate account records; approximately 2% of accounts had incomplete session-count data, arising from a known logging gap during a platform migration in month seven, and these accounts were excluded from the regression model, though retained in the descriptive cohort analysis where the missing period did not affect the retention calculation. This exclusion reduced the regression sample to 18,032 accounts, which is not expected to materially bias the results given the small proportion affected and the absence of any obvious pattern linking the logging gap to a particular customer segment.

Findings

Overall churn during the twelve-month period averaged 8.9% per month, but this figure conceals a clear upward trend across the year, rising from 8.2% in the first quarter to 10.1% in the fourth (Figure 1). The increase became visible in the same month that the Standard tier price was raised from £9.99 to £11.49, although the data available to this report cannot establish that the price change was the sole cause of the rise.

Table 1 presents retention curves for four customer cohorts, grouped by the quarter in which they joined.

The cohort comparison shows that later cohorts retained less well at the Month 3 mark than the January cohort: 84% for the October cohort and 86% for the July cohort, against 89% for January. This suggests the erosion in retention is not confined to long-standing customers but is also affecting newer sign-ups, which is a more concerning signal about the underlying customer experience than a simple ageing-cohort effect would be.

Table 2 summarises the results of the logistic regression model, which had an overall classification accuracy of 74% on a held-out validation sample.

The strongest predictor of churn was low engagement: customers with fewer than four sessions per month were roughly three times more likely to churn than more active users (OR = 3.02, p < .001), even after controlling for plan tier and tenure. Monthly billing was also strongly associated with churn relative to annual billing (OR = 2.14, p < .001), consistent with the lower switching cost and weaker psychological commitment associated with rolling monthly contracts (Ascarza, 2018). A customer service contact within the preceding 60 days nearly doubled the odds of churn (OR = 1.86, p < .001), pointing to unresolved service issues as a meaningful driver of cancellation rather than a neutral touchpoint.

Loyalty discount redemption was associated with substantially lower churn odds (OR = 0.58, p < .001), and each additional month of tenure reduced churn odds by around 3% (OR = 0.97, p < .001), reflecting the well-documented pattern that longer-tenured customers are more embedded in a service and therefore less likely to leave. Plan tier itself was not a statistically significant independent predictor once engagement and billing frequency were controlled for (Premium vs Basic OR = 0.81, p = .072), indicating that price tier alone is a weaker driver of churn than behavioural engagement and contract type.

A supplementary breakdown by acquisition channel found that customers acquired through the broadband bundle partnership churned at a notably lower rate (6.1% average monthly) than customers acquired through direct sign-up (10.4%), despite broadly similar engagement scores across the two groups. This is consistent with bundled subscribers experiencing a form of switching friction not captured directly in the regression model, since cancelling the media subscription for these customers may involve contacting the broadband provider rather than Northbridge Digital directly. This finding suggests that some of the churn reduction associated with annual billing in the regression model may partly reflect a similar friction effect operating through a different mechanism, a point returned to in the Discussion below.

A seasonal pattern was also evident within the quarterly trend: churn spiked noticeably in the first two weeks of January in both years covered by the dataset, consistent with the well-documented industry pattern of subscription cancellations following the holiday period as customers review household budgets (Recurly, 2022). This seasonal spike was distinct from, and additive to, the underlying upward trend driven by the Standard tier price increase, and should be accounted for separately when the business sets churn targets for the coming year rather than being read as evidence that the price change itself is losing effect.

Discussion

These findings are broadly consistent with the wider literature on subscription and telecommunications churn. Verbeke et al. (2012) similarly find that behavioural and usage variables tend to outperform demographic or price-tier variables as churn predictors, a pattern reflected here in the strength of the engagement variable relative to plan tier. The result also supports Ascarza’s (2018) argument that customers on lower-commitment contracts, such as rolling monthly plans, are structurally more likely to churn regardless of satisfaction, because the psychological and administrative cost of leaving is minimal; this has direct implications for how Northbridge Digital structures its billing options.

The association between recent customer service contact and higher churn is worth particular attention. It does not necessarily mean that customer service caused dissatisfaction; a more plausible interpretation, consistent with Lemon and Verhoef’s (2016) customer-journey perspective, is that customers who contact support are often already experiencing a problem serious enough to consider leaving, and the interaction represents a final opportunity to intervene rather than the root cause of churn. Read this way, the finding is less an indictment of the service desk and more a signal that the escalation and resolution process at that touchpoint needs strengthening, a point developed further in the Recommendations.

The low engagement finding aligns with Reichheld’s (1996) long-standing observation that usage intensity is one of the clearest leading indicators of loyalty across service industries, and with more recent churn-prediction work by Chen, Fan and Sun (2012) and Neslin et al. (2006), both of which identify usage recency and frequency as consistently among the strongest predictors across modelling approaches. This convergence across methods and sectors increases confidence that engagement, rather than price alone, should be the primary lever for the retention programme.

Two caveats temper these conclusions. First, Ascarza, Iyengar and Schleicher (2016) caution that proactive retention offers targeted at customers already flagged as high-risk can sometimes accelerate rather than prevent churn, by drawing attention to the possibility of cancelling or by training customers to expect a discount before they consider leaving; this risk is addressed in the Recommendations through the design of the proposed outreach. Second, this analysis is based on twelve months of data from a single organisation and does not capture qualitative reasons for cancellation or the effect of external market conditions, such as the wider cost-of-living pressures reported in UK household spending data (Ofcom, 2023), which may also be contributing to the observed rise in churn independently of anything Northbridge Digital has changed.

These findings also carry direct implications for customer lifetime value and acquisition strategy. Because bundled subscribers churn at roughly half the rate of directly acquired customers, their effective lifetime value is materially higher even before accounting for any difference in acquisition cost, which suggests the commercial team should weight channel-level retention performance, not just channel-level acquisition cost, when allocating future marketing spend. This is consistent with wider guidance that customer acquisition cost should be evaluated against lifetime value rather than in isolation (Gallo, 2020). At the same time, the business should be cautious about over-relying on partnership-driven bundling as a retention strategy in itself, since the lower churn observed here is most plausibly explained by switching friction rather than by genuinely higher satisfaction, and friction-based retention carries reputational risk if customers come to perceive cancellation as unreasonably difficult.

Recommendations

1. Proactive engagement outreach (high priority, low cost). Customers falling below four sessions per month should be flagged automatically and contacted with a light-touch, non-discount nudge — a curated content recommendation or a reminder of underused features — within two weeks of crossing the threshold, rather than waiting until cancellation risk is acute. Following the caution raised by Ascarza, Iyengar and Schleicher (2016), this should avoid leading with a discount, which can inadvertently prompt cancellation consideration.

2. Revised customer service escalation protocol (high priority, moderate cost). Given that a recent service contact nearly doubles churn odds, cases that are not resolved on first contact should be escalated to a named account owner within 24 hours, with a follow-up check-in seven days later to confirm the issue has not recurred.

3. Incentivised annual billing switch (medium priority, low cost). Customers on monthly plans with six or more months of tenure should be offered a modest, time-limited discount to switch to annual billing, directly targeting the contract-type effect identified in the regression model while limiting the offer to customers who have already demonstrated commitment.

4. Redesigned onboarding sequence (medium priority, moderate cost). Because low engagement is the single strongest predictor of churn, a structured first-30-days onboarding sequence should be introduced to establish usage habits early, including a guided first-session walkthrough and a personalised content digest in week two.

5. Quarterly churn dashboard (low priority, low cost). A recurring dashboard reporting cohort retention, the regression-identified risk factors, and intervention uptake should be established so that the commercial team can monitor whether these interventions are reducing churn and adjust them accordingly.

Conclusion

This report set out to quantify and explain the rising churn observed at Northbridge Digital over the past twelve months, and to translate the findings into a practical set of interventions. The analysis shows that churn rose from 8.2% to 10.1% across the year, and that this rise is best explained by declining customer engagement, the prevalence of low-commitment monthly billing, and unresolved customer service issues, rather than by plan tier or price alone. These are all factors the business can act on directly.

The five recommendations set out above — proactive engagement outreach, a revised service escalation protocol, incentivised annual billing, redesigned onboarding, and ongoing dashboard monitoring — are deliberately sequenced from low-cost and immediate to more resource-intensive, so that the commercial team can begin implementation without waiting for a larger budget cycle. If adopted together, they are projected to reduce monthly churn by 1.5 to 2 percentage points within two quarters, with the dashboard recommendation providing the mechanism to verify this impact and refine the programme as further data becomes available.

References

• Ascarza, E. (2018) ‘Retention futility: Targeting high-risk customers might be ineffective’, Journal of Marketing Research, 55(1), pp. 80–98.
• Ascarza, E., Iyengar, R. and Schleicher, M. (2016) ‘The perils of proactive churn prevention using plan recommendations’, Journal of Marketing Research, 53(1), pp. 46–60.
• Chen, Z.Y., Fan, Z.P. and Sun, M. (2012) ‘A hierarchical multiple kernel support vector machine for customer churn prediction’, European Journal of Operational Research, 223(2), pp. 461–472.
• Gallo, A. (2020) ‘The value of keeping the right customers’, Harvard Business Review, October.
• Kotler, P. and Keller, K.L. (2016) Marketing Management. 15th edn. Harlow: Pearson.
• Lemon, K.N. and Verhoef, P.C. (2016) ‘Understanding customer experience throughout the customer journey’, Journal of Marketing, 80(6), pp. 69–96.
• Neslin, S.A., Gupta, S., Kamakura, W., Lu, J. and Mason, C.H. (2006) ‘Defection detection: measuring and understanding the predictive accuracy of customer churn models’, Journal of Marketing Research, 43(2), pp. 204–211.
• Ofcom (2023) Communications Market Report. London: Ofcom.
• Recurly (2022) Subscription Churn Rate Benchmarks Report. San Francisco: Recurly Inc.
• Reichheld, F.F. (1996) The Loyalty Effect. Boston: Harvard Business School Press.
• Verbeke, W., Dejaeger, K., Martens, D., Hur, J. and Baesens, B. (2012) ‘New insights into churn prediction in the telecommunication sector’, European Journal of Operational Research, 218(1), pp. 211–229.