The customer health score started in software firms that sell subscriptions. A team with hundreds of accounts needs to know which ones are drifting before the renewal date. A loyalty program faces the same question for millions of members, with no renewal date to warn it.
What is a customer health score?
A customer health score blends a few signals into one grade for how a customer is doing. Definitive Healthcare, a data firm, uses one. It calls the score a predictive metric that shows whether clients are likely to renew. The staff who onboard and train new clients keep it up to date (Definitive Healthcare 10-K).
The score now shows up in annual reports. Lumen Technologies lists it beside Net Promoter Score as a customer experience score. Lumen says bad trends in either one, compared with rivals, could hurt the business (Lumen 10-K).
Most teams show it as points from 1 to 100. Others use letter grades or a green, yellow and red scale (ChurnZero). Whatever the format, the score exists to answer one question fast: which customers need attention first?
What goes into a customer health score?
Most models rest on a few dimensions. A 2024 Harvard Business Review article by Bryan Hochstein and colleagues names three. The first is the quality of the relationship. The second is product usage, and the third is the value the customer has realized. The article walks through how BigCommerce, an ecommerce platform, built and refined its own model. Its lessons include finding the link between health and churn, and a flywheel effect as scoring matures (HBR, 2024).
Gainsight, a customer success software firm, teaches a four-part version called DEAR. The letters stand for deployment, engagement, adoption and ROI. Teams set a threshold and a weight for each metric. A team that wants more feature use can give adoption the most weight (Gainsight).
ChurnZero advises five to seven factors per score, each worth 10 to 20 points on a 100-point scale. It also advises a separate score for each stage, because a new customer uses the product in a different way from one in year two (ChurnZero). The table maps each dimension to a software signal and its loyalty program twin.
| Dimension | B2B software signal | Loyalty program signal |
|---|---|---|
| Usage | Logins and features used per seat | Purchase days and recency against the member’s own cycle |
| Relationship | Contact with the buyer and the sponsor | App visits, email opens and replies to offers |
| Value realized | Goals met in a success plan | Points redeemed and benefits used |
| Sentiment | Survey scores and call notes | NPS or CSAT answers joined to the member record |
| Service | Open tickets and escalations | Complaints, missing points claims and repeat contacts |
| Commercial | Renewal date and seat count | Spend trend, tier progress and share of category spend |
How do you calculate a customer health score?
Start from the outcome the score should predict: renewal for a contract, or more buying for a member. Pick the few signals that moved before past customers left. Score each one from 0 to 100 against a threshold. Multiply each by its weight, then add them up. Last, set the bands, such as green for 70 and up.
For illustration, a hotel program scores members on four signals. Recency against the member’s usual gap between stays carries 40%. The trend in stays per year carries 30%. Redemption and benefit use carry 20%, and service contacts the last 10%. One member scores 90 on recency, 60 on trend, 50 on redemption and 100 on service. The weighted sum is 36 plus 18 plus 10 plus 10, or 74, so she is green. If her next stay slips well past her usual gap, recency falls to 30. Her score drops to 50, which is yellow.
Keep the health score apart from a churn risk score. Planhat describes health as the current state of a relationship across several dimensions, written for a person to read. A churn risk score is one forward-looking odds figure, usually from a model. Planhat says the two diverge more often than teams expect, so it uses both (Planhat).
How do you know a health score works?
Test it against what customers did next. Gainsight’s chief customer officer wrote that the firm’s renewal rates have lined up almost perfectly, year after year, with the scores its DEAR model produced (Gainsight). That is the right test, run by a vendor on its own book.
Method matters. In a 2006 contest, teams from schools and firms built churn models on the same data. The gaps in accuracy between teams could swing a campaign’s profit by hundreds of thousands of dollars. The models lost very little accuracy on data from three months later (Neslin et al., 2006).
Then test whom to act on. Eva Ascarza ran two field experiments and found that the customers at highest risk are not always the best targets. She advises targeting customers by how much they respond to the offer, whatever their risk (Ascarza, 2018). A health score finds who is drifting. A holdout test shows whose behavior an offer can change.
Watch for drift as well. Planhat notes that a score can only be as good as the signals a team can reach. The signals a team wires up on day one tend to stay in the score for good, whether or not they turned out to predict anything (Planhat).
How do loyalty programs score member health?
Most members hold no contract, so health rests on behavior. The usual inputs are recency against the member’s own buying cycle and the trend in frequency. Breadth across categories counts too, along with redemption and service history. Ascarza and Bruce Hardie found that use and renewal reflect one hidden state that shifts over time. They tested their model on a group that charges a yearly fee to buy, and it maps the paths members take before they quit (Ascarza and Hardie, 2013).
Rakuten shows the version for a whole membership. It values its members as active users times cross-use, times lifetime value less the cost to acquire. Cross-use is the share of its users who used two or more Rakuten services in the past 12 months. That share stood at 77.0% at the end of June 2026, and membership value rose 16.1% in a year (Rakuten, Q2 2026).
Use member health to route effort. A high-value member whose score falls gets a service call or a personal offer. A low-value member with a falling score gets a lighter touch. Keep a holdout on each play, so the team learns which drops an offer can reverse.
How GRAVTY handles customer health scores
In GRAVTY, native and external scores sit on the member record as fields that offers can act on. That turns an at-risk list into an audience (customer retention). Member 360 pulls each member’s behaviors and transactions into one view, household included. The patented Visual Rules engine lets the loyalty team write a win-back rule, test it in a simulation and put it live without an IT ticket.
On the airline side, GRAVTY looks for disengagement and churn risk early enough to act on it. One question it answers: which high-value members show the first signs of drifting away (airline loyalty)?
Frequently asked questions
What is a good customer health score?
A good customer health score is one that splits customers who stay from customers who leave. The number itself depends on the scale and the bands you set. Check each band against past renewals or repeat purchases. If green and yellow customers leave at the same rate, move the cut-off or change the inputs.
What is the difference between a health score and a churn score?
A health score shows how a relationship is doing now across several dimensions, while a churn score is a model’s odds that the customer leaves. Planhat notes that an account can look healthy and still be at high risk, for example when its champion departs (Planhat). Use health for daily work and risk for escalation.
Should NPS be part of a customer health score?
Yes, as one input among several. Gainsight’s DEAR model counts survey feedback as part of ROI, next to data on use and adoption (Gainsight). But surveys cover only the customers who reply. Behavior covers everyone, so it should carry more of the weight.
How often should a customer health score update?
Update it whenever its inputs change: daily for usage, and after each survey or ticket for the rest. Review the weights each quarter against who actually left. In a 2006 study, churn models kept nearly all their accuracy three months after they were built (Neslin et al., 2006), so a quarterly review keeps pace.
Who owns the customer health score?
In software firms, the customer success team owns it. At Definitive Healthcare, the staff who onboard and train new clients keep the score (Definitive Healthcare 10-K). In a loyalty program, the CRM or analytics team should own the model. Service and marketing then act on it.
