Customer Health Score for Ecommerce: How to Predict Loyalty & Prevent Churn?
Most ecommerce retention problems are not surprising. The signals that a customer is disengaging accumulate over weeks before any formal churn event occurs: purchase frequency drops, email engagement declines, support contacts increase, loyalty activity stalls. The challenge is not the absence of data but the absence of a framework that consolidates those signals into an actionable, real-time view of each customer's relationship health. A customer health score provides that framework.
What is a Customer Health Score?
A customer health score (CHS) is a composite metric that synthesises multiple behavioural, transactional, and engagement signals into a single weighted number, typically expressed on a 0 to 100 scale or as a red, amber, green status. The score functions as a predictive indicator of a customer's likelihood to remain active, return for additional purchases, or disengage and churn.
The concept originated in B2B SaaS, where account health was tracked to predict subscription renewals. Its application to ecommerce is structurally equivalent: in both contexts, the goal is to identify which customers are building a relationship with the brand and which are moving toward exit before the exit event makes intervention too late. AI-enhanced health score models can predict churn 60 to 90 days in advance with 85% accuracy, according to Gainsight's 2024 benchmarking data, giving marketing and customer success teams a meaningful intervention window.
Why Customer Health Scores Matter for Ecommerce?
Aggregate retention metrics tell you that retention is declining. They do not tell you which specific customers are at risk, why they are at risk, or which intervention is appropriate for each segment. A customer health score framework addresses all three dimensions simultaneously.
The commercial stakes are significant. Organisations implementing comprehensive churn prediction strategies based on health scoring typically see a 20% to 30% reduction in customer churn rates. Given that acquiring a new customer costs five to seven times more than retaining an existing one, and that returning customers generate higher average order values than first-time buyers, each prevented churn event carries compounding commercial value that extends well beyond the immediate transaction.
Health scores also enable resource allocation. Without a scoring system, retention effort is applied reactively and inconsistently, typically directed at customers who have already contacted the brand about an issue. With a health score, proactive retention effort can be concentrated on customers whose behavioural trajectory indicates risk before it becomes visible through a service interaction or a cancelled subscription.
Key Signals That Make Up a Health Score?
A well-designed ecommerce health score draws on four signal categories. The weighting of each signal should be calibrated against historical data from the specific brand's customer base, ideally by correlating each signal's value against actual churn outcomes over a defined retrospective window.
Purchase Frequency
Purchase recency and frequency are the foundational signals in any ecommerce health model. A customer who purchased three times in the last 90 days has a fundamentally different health profile from one who last purchased eight months ago. The most predictive metric is not the raw frequency count but the deviation from that individual customer's established baseline. A customer who typically purchases monthly and has now gone 45 days without a transaction is at a higher churn risk than one with a naturally low purchase frequency who is still within their normal pattern.
Rolling 30 and 90-day purchase frequency windows, combined with a recency score based on days since last transaction, form the core transactional layer of the model. These signals should account for 25% to 35% of the overall score in most ecommerce contexts.
Engagement Rate
Engagement signals capture how actively the customer is interacting with the brand outside of transactional moments. Email open and click rates, app session frequency and duration, website visit recency, and responsiveness to loyalty programme communications all contribute to the engagement dimension.
A customer whose email open rate has declined from 60% to 15% over the past two months is signalling reduced interest in the brand relationship, even if their last purchase was recent. Declining engagement consistently precedes declining purchase frequency, making it a leading indicator of risk that the transactional signals alone will surface too late.
Engagement signals typically carry a weight of 20% to 25% in a balanced ecommerce health score.
Support Ticket Volume
Support interactions are a dual-directional signal. A low volume of resolved support contacts is neutral to positive, indicating that the product experience is largely smooth. A high volume of unresolved or escalated tickets, or a rapid increase in contact frequency, is a strong churn predictor. According to Totango's 2025 research, an increase in unresolved or high-severity support interactions is among the most reliable leading indicators of imminent churn.
The sentiment of support interactions adds qualitative depth to the volume signal. Tickets submitted with increasingly frustrated language, repeated contacts about the same unresolved issue, or a pattern of returns and refund requests all indicate a deteriorating relationship that a simple ticket count would not fully capture. Where NLP tooling is available, sentiment scoring of support transcripts adds a meaningful layer to this signal.
Support signal weighting typically sits between 15% and 20% of the composite score.
Loyalty Programme Activity
For brands operating a loyalty programme, programme engagement is one of the most sensitive and commercially specific health signals available. Active earning, redemption activity, and programme communication engagement all indicate that the customer perceives ongoing value in the relationship beyond the transactional. Conversely, a loyalty member who stops earning, allows their balance to approach expiry without redeeming, or disengages from programme communications is exhibiting disengagement behaviour that the transactional signals may not yet reflect.
The loyalty engagement signal is particularly valuable because it is brand-specific in a way that general engagement metrics are not. A customer who is browsing your website less frequently may simply be browsing less overall. A customer who has stopped interacting with your loyalty programme specifically has reduced their investment in the brand relationship, which is a more targeted risk signal.
Loyalty programme activity typically contributes 15% to 20% of the composite score.
How to Build a Customer Health Score Model?
Step 1: Define what "churned" means for your business. Before building the model, the churn event must be defined precisely. In ecommerce, churn is typically defined as no purchase within a defined inactivity window, most commonly 90, 120, or 180 days depending on the category's natural purchase cycle. All subsequent modelling is built around this definition.
Step 2: Select four to six predictive signals. More signals do not automatically produce a better model. Tracking everything creates noise and makes the score harder to interpret and act on. Selecting four to six signals that demonstrably correlate with churn outcomes, based on historical data analysis, produces a more reliable and actionable score than an over-engineered composite of every available data point.
Step 3: Normalise each signal to a common scale. Raw signal values are not directly comparable: a purchase frequency of three per month and an email open rate of 45% cannot be added together meaningfully. Normalising each metric to a 0 to 100 scale, where 100 represents the healthiest observed value and 0 represents the least healthy, creates a common currency for weighted combination.
Step 4: Assign weights based on predictive correlation. Use historical churn data to determine which signals most reliably predicted actual churn events. Signals with higher predictive correlation should receive higher weights. The initial weights should be treated as hypotheses to be validated and refined over time as more outcome data accumulates.
Step 5: Calculate and update scores continuously. A health score that is calculated monthly and reviewed quarterly provides insufficient signal speed to enable timely intervention. Scores should be updated at least weekly, and ideally daily for high-frequency signals such as email engagement and site activity.
The composite score formula takes the form: CHS = (Signal A value × Weight A) + (Signal B value × Weight B) + (Signal C value × Weight C), summed across all included signals, producing a figure between 0 and 100 that is mapped to a health tier (healthy at 75 to 100, at-risk at 40 to 74, critical at 0 to 39, following Totango's widely used tier framework).
Using Health Scores to Trigger Loyalty Interventions
A health score that exists only in a dashboard is an analytical exercise, not a retention tool. The commercial value is realised when score changes automatically trigger specific, pre-defined interventions delivered through the loyalty programme and communication channels.
A customer moving from the healthy to at-risk tier should trigger a personalised loyalty communication that surfaces their current balance, highlights what they are close to earning, and presents a contextually relevant offer tied to their recent purchase history. The offer should feel like a recognition of the relationship rather than a generic retention attempt.
A customer entering the critical tier should trigger an escalated intervention: a higher-value incentive, a personal outreach from a customer service representative for high-LTV segments, or a points bonus event with a short expiry window designed to prompt an immediate re-engagement transaction.
Customers who move from at-risk back to healthy following an intervention provide the validation data for the intervention's effectiveness. Tracking the percentage of at-risk customers who return to healthy status after each intervention type enables ongoing optimisation of both the scoring thresholds and the intervention playbooks.
Customer Health Score Tools
Several tools support customer health scoring in ecommerce contexts, with varying levels of complexity and integration requirements.
Klaviyo enables health-score-adjacent segmentation through its predictive analytics layer, including predicted customer lifetime value, expected next-order date, and churn risk probability. These predictive segments can be used to trigger automated flows that approximate a health score intervention without requiring a separate scoring infrastructure.
Peel Insights and RetentionX are specialist ecommerce retention analytics platforms that generate customer-level health metrics directly from Shopify data, including RFM scoring, cohort retention curves, and at-risk customer identification.
Gainsight and Totango are purpose-built customer health scoring platforms originated in the B2B SaaS context but increasingly applied to high-value ecommerce and subscription commerce customer bases where structured health monitoring is commercially justified.
For brands with data warehouse access and SQL capability, a custom health score can be built directly from order, session, and loyalty event data and surfaced through a business intelligence tool such as Looker, Metabase, or Tableau. This approach requires more technical investment but produces a model precisely calibrated to the brand's own churn patterns rather than adapted from a generic framework.
The appropriate choice depends on data volume, technical infrastructure, and the commercial value of the customer base being monitored. For most ecommerce brands, starting with the predictive analytics features of their existing email or loyalty platform provides the most accessible entry point, with migration to dedicated tooling as the scoring framework matures.
