Customer health score: 6 signals and how to calibrate

A customer health score is a 0 to 100 rating that sums up, for each account, the chance that it renews, and it only warns of a cancellation months in advance when it combines signals of different kinds and takes its weights from the company’s own cancellation history. Research on the shortcut most often used in place of a score, NPS, points the same way: in a two-year panel of more than 8,000 customers from three industries in the United States, recommend intention alone was not enough to predict customers’ future behavior, and a model with several indicators did better at predicting retention and recommendation (Keiningham, Cooil, Aksoy, Andreassen and Weiner, 2007). For the CS manager and for leadership, a score that works combines six families of signals (product usage, engagement with the team, tickets, payment, sponsor change and customer perception), goes through a test against the accounts that left and ties each band to an action with an owner and a deadline in the CRM. Without that link, the score sits on the dashboard and the CS calendar stays the same.

In B2B churn I proposed a starter score, with six scored signals and three bands. Here I cover the next step, which turns that starting point into a management tool: the indicators for each family, weights calculated from the company’s data, the test before you trust the number, the playbook for each band and the build in the CRM.

What is a health score and which decisions does it support?

A health score, or customer health rating, is a leading indicator: the score moves before revenue moves. Churn rate, GRR and NRR count what has already happened and arrive too late for the account that left, while the score exists so that CS can see the at-risk account while there is still time to act, preferably 90 to 180 days before renewal. NPS and ticket satisfaction feed the score as inputs, because they describe the opinion of whoever answered at one moment. The score supports decisions that today come from memory: where CS spends the week’s hours, which renewals leadership follows closely and which accounts can receive an expansion proposal.

Which signals go into the health score?

Each family of signals answers a different question about the account, and a score built on one family sees only one kind of risk. Pick one or two indicators per family, always from a source that records the date, because a signal without a date blocks the test against the past that I describe below. The table is a starting point for the first version.

Family	Suggested indicator	Common trap
Product usage	Active users over contracted licenses, against the average of the previous three months	Measuring logins instead of the feature tied to the agreed outcome
Engagement with the team	Days since the last meeting with the decision maker or sponsor	Counting the operational user’s emails as relationship
Tickets	Critical tickets open for more than 10 days or reopened in the quarter	Scoring total volume, which rises when the customer uses more
Payment	Invoice overdue by more than 15 days, billing dispute, discount request outside renewal	Applying the same deadline to customers who pay on a group calendar
Sponsor change	Sponsor leaving or replaced, reorganization, merger or acquisition	Finding out when the email bounces
Customer perception	Decision maker’s score in the last survey and their lack of response	Using the average of all respondents

How to measure usage when the company sells services

In consulting, implementation, agencies or recurring services, usage shows up in consumption of the contracted scope: hours used against hours contracted, deliverables approved on time, customer attendance at follow-up meetings and progress on the success criteria agreed in onboarding. The account that contracted 40 hours a month and uses 12 is signaling the same thing as the software customer with half its licenses idle.

Sponsor change: the signal the system does not capture on its own

No integration detects a change in the person who championed the purchase. The account owner logs the change on the day they hear of it, and the company record in the CRM keeps the current sponsor with the date of the last confirmation. Once a quarter the CSM confirms whether that person is still in the role and has the same influence, and an expired confirmation becomes an alert.

NPS in the health score: why it comes with caveats

NPS was born in the article in which Frederick Reichheld, of Bain, argued for a single question about recommendation. He wrote that, in most of the industries he studied, the share of customers willing to refer the company tracked the differences in growth rates among competitors, and noted in the same text that this did not hold in every industry (Reichheld, 2003). The academic attempt to replicate the finding, with longitudinal data from 21 firms and more than 15,500 interviews from the Norwegian Customer Satisfaction Barometer, did not find NPS superior to other measures in the industries Reichheld himself cited as examples (Keiningham, Cooil, Andreassen and Aksoy, 2007). Both academic studies work with consumers, and in B2B the problem grows, because one account has several respondents and the decision maker seldom answers. That is why perception enters as one signal among six, with a weight calibrated like the others: use the score of the decision maker or sponsor and treat their silence in two surveys in a row as weak engagement.

How do you weight each signal and calibrate against past cancellations?

A weight set in a meeting reflects the last cancellation that hurt leadership. A study of churn in non-contractual B2B relationships compared several prediction techniques and concluded that the model-driven approach, also used to design the retention campaign, outperformed the rules of thumb managers commonly use (Tamaddoni Jahromi, Stakhovych and Ewing, 2014). To start, you need the account history and a simple comparison.

Step 1: build the calibration base

List the accounts that had a renewal decision in the last 12 to 24 months and classify the outcome: renewed, reduced scope or cancelled. Add to the cancelled group the reductions that removed 20% or more of the account’s revenue, because that customer gave the same signals as those who left. For each account, mark the signals present in the 180 days before the decision. When I ask for this base in the companies I work with, the hard part is usually the date of the signals, because the CRM keeps the current state of the account and loses the history.

Step 2: compare cancellation with and without each signal

The hypothetical example uses 200 accounts with a decision in the last year, of which 40 cancelled or reduced scope (20%) and 160 renewed. For each signal, compare the cancellation rate of the accounts that had the signal with that of the accounts that did not.

Signal	Cancelled with the signal (of 40)	Renewed with the signal (of 160)	Cancellation with the signal	Cancellation without the signal	Ratio	Weight
Usage drop above 30%	25	20	25 ÷ 45 = 55.6%	15 ÷ 155 = 9.7%	5.7	35
Sponsor change	12	8	12 ÷ 20 = 60.0%	28 ÷ 180 = 15.6%	3.8	20
No meeting with decision maker in 90 days	22	40	22 ÷ 62 = 35.5%	18 ÷ 138 = 13.0%	2.7	15
Critical ticket open for more than 10 days	14	30	14 ÷ 44 = 31.8%	26 ÷ 156 = 16.7%	1.9	10
Invoice overdue by more than 15 days	10	20	10 ÷ 30 = 33.3%	30 ÷ 170 = 17.6%	1.9	10
Decision maker detractor or no response	9	24	9 ÷ 33 = 27.3%	31 ÷ 167 = 18.6%	1.5	10

The weights split 100 points in proportion to the ratios, which add up to 17.5, rounded to multiples of five. The score starts at 100 and loses the weight of each signal present: an account with a usage drop and no meeting with the decision maker lands at 50. A signal present in fewer than ten cancelled accounts swings a lot from one year to the next and keeps the minimum weight. Signals also move together, since a sponsor change tends to pull usage down months later, and a simple sum counts the same risk twice; with a few hundred decisions on record, whoever owns RevOps can fit a logistic regression on the same base.

Step 3: test the score on the snapshot from 90 days before

Calculate the score each account would have had 90 days before the decision, the minimum time CS needs to act, and see how the bands separated those who left from those who stayed. In the hypothetical example, with green from 85 to 100, yellow from 50 to 84 and red below 50:

Band 90 days before	Accounts	Cancelled or reduced	Cancellation rate
Green (85 to 100)	120	14	11.7%
Yellow (50 to 84)	50	8	16.0%
Red (below 50)	30	18	60.0%
Total	200	40	20.0%

Coverage shows that 26 of the 40 accounts that left (65%) were yellow or red with 90 days to go, and the red hit rate, 18 of 30 (60%), justifies the effort of a recovery plan. Yellow separates little from green, 16% against 11.7%, and calls for a different cutoff or a missing signal. Open the 14 green accounts that left: if many of them had no integrated usage data, the problem is recording, and the “no data” band in the playbook deals with it. Calibrate on the older decisions and test on the more recent ones.

Step 4: recalibrate every quarter

At the end of each quarter, compare the score from 90 days before with the outcome of the renewals and redo the table from step 2. Changes in product, customer profile or pricing shift the weight of the signals. Record the version of the rule in the CRM so reports compare accounts calculated the same way.

Health score bands: what CS does in each one

Each band triggers an action with a deadline and an owner, which leadership checks in the portfolio review. Adjust the playbook to the segmentation of your base, because a strategic account in yellow gets more attention than a maintenance account in the same band, as I discuss in account management.

Band	Criterion	What CS does	Deadline	Who joins	Record in the CRM
Green	85 to 100, with signals filled in	Keeps the cadence, reviews value delivered and looks for expansion signals	Quarterly review	CSM, with sales on expansion	Expansion signal becomes an opportunity
Yellow	50 to 84	Diagnoses the signal that pulled the score down and talks to the sponsor	Contact within 5 business days and plan within 15	CSM and CS leadership in the weekly review	Likely cause, next step and date
Red	Below 50	Builds a recovery plan with exit criteria and suspends expansion proposals	Meeting with the decision maker within 10 business days	CS leadership, original salesperson and executives if the contract is large	Plan with owners and dates, reviewed every week
No data	No usage data, or sponsor unconfirmed for 90 days	Completes the data before classifying	10 business days	CSM and CRM admin	Provisional band until the data arrives

In the green band, the value review with the sponsor happens in the ritual described in QBR, and the expansion signal follows the rule between CS and sales in revenue expansion. In yellow, the account is still talking and the problem usually has a single cause. In red, choose the action by the recorded cause: a discount offered to an account that lost its sponsor still leaves it without an internal champion, and the research I gathered in customer retention shows that a poorly targeted retention action can bring the exit forward.

How do you build the health score in the CRM?

The health score lives on the company record in the CRM, where the whole team can see it. Whoever administers the CRM or owns RevOps builds the pieces, and CS leadership owns the definitions.

Piece	What to configure	Who maintains it
Signal properties	One field per signal with a date, such as “usage drop since” and “sponsor confirmed on”	Integrations and account owner
Score and band	Calculated score from 0 to 100, band, heaviest signal and rule version	CRM admin
Manual override	Band set by the CSM, with mandatory justification and 30-day validity	CSM, with leadership approval
Monthly snapshot	Score and band of each account on the first day of the month, stored as history	CRM automation
Lists	Accounts that changed band this week, red accounts without a plan, renewals within 120 days outside green	CRM admin
Alerts	Task for the owner when the account drops a band, with the playbook deadline	CRM automation
Reports	Accounts and revenue by band, cancellation by band from 90 days before, share with no data	CS leadership and RevOps

Signals that come from systems, such as usage, tickets and invoices, arrive through a daily integration and recalculate the score on their own; those that depend on a person, such as a sponsor change, are entered by the CSM, and the weekly review checks who failed to log them. The monthly snapshot looks like a configuration detail, and without it the test in step 3 means rebuilding the history by hand the following quarter. Avoid an alert for every point of movement, because the team learns to ignore notifications within a few weeks. Duplicate contacts and companies without an owner undermine trust in the score, problems I cover in CRM data quality.

Common health score mistakes

A score nobody uses

The symptom is a score calculated in the CRM and a portfolio review run from the CSM’s memory, usually because the band triggers no action or because the team never saw the test against the past. Tie each band to a task with a deadline, open the review with the accounts that changed band and show the team the coverage and hit rate table. In the CS teams I work with, the score starts being used when leadership asks about the red account without a plan in front of the group, week after week.

An NPS-only score

The recommendation score measures the opinion of whoever answered, once or twice a year, and leaves the account that did not answer without a score. The studies by Keiningham and coauthors indicate that several indicators together predict retention better than the single question.

Weights set by opinion

Leadership puts 40 points on tickets because the last lost customer complained about support, and the score starts warning about a risk that has already passed. Redo the weights with the table from step 2 and use leadership’s opinion to choose the candidate signals.

Green accounts for lack of data

When the score starts at 100 and only loses points for recorded signals, an account with no usage integration and no recorded meeting looks healthy. It needs the “no data” band, with a deadline for the CSM to complete the record.

Health score and L&D: what HR needs to know

HR recognizes the need when leadership learns about cancellations from finance, when CS has a score that does not change the calendar or when the team came from support and treats renewal like a ticket. Participants are CS leadership and analysts, onboarding, the salespeople who close recurring contracts and whoever administers the CRM. In the program, the group builds the calibration base with real accounts and practices the conversation with the sponsor in the yellow band, the skill most often missing in people who came from support. To measure, set a baseline for the share of accounts with a score updated this month, the share of red accounts with a plan and the 90-day coverage of the score, and agree with leadership that GRR responds only after a renewal cycle.

30-day roadmap for leadership

1. Week 1: write the definitions. Record what counts as a cancellation and a relevant reduction and which signals go into the first version.
2. Week 1: build the calibration base. List the renewal decisions from the last 12 to 24 months with the signals from the 180 days before each one.
3. Week 2: calculate the weights. Build the with-and-without table for each signal and split the 100 points by the ratios.
4. Week 2: test on the 90-day snapshot. Calculate coverage and hit rate and adjust the cutoffs until yellow separates from green.
5. Week 3: configure the CRM. Create the dated signal properties, the score, the band, the manual override and the monthly snapshot.
6. Week 3: write the playbook. Define action, deadline, owner and record for each band, including “no data”.
7. Week 4: run the portfolio review with the score. Open with the accounts that changed band, with the CRM on screen.
8. Week 4: schedule the recalibration. Book the comparison between the 90-day score and renewals for the end of the quarter.

Running the review from the score and acting on each band are behaviors that leadership and the team build through guided practice on the real portfolio, and that is what I do in the in-company CRM adoption training: I build the calibration base and the band playbook with leadership, configure the properties, lists and alerts with whoever administers the system and prepare CS for the conversation with the sponsor. The company leaves with the CRM best practices guide and the portfolio and customer base management manual, written around your operation.

Frequently asked questions

What is a customer health score?

A health score is a rating, usually from 0 to 100, that sums up the chance an account renews based on observable signals such as usage, engagement, tickets, payment, sponsor and perception. It helps CS prioritize accounts and helps leadership follow renewals before they come due.

Which signals should a B2B health score use?

Start with one or two indicators per family: usage of the product or of the contracted scope, a recent meeting with the decision maker, critical tickets, overdue invoices, sponsor change and the decision maker’s survey score. Each signal needs a dated source, so you can test it against past cancellations.

How do you set the weight of each signal?

Compare the cancellation rate of accounts that had the signal with that of accounts that did not, using renewal decisions from the last 12 to 24 months. Split the points in proportion to those ratios and test the score calculated 90 days before each decision, adjusting the band cutoffs.

Can NPS be used as a health score?

NPS on its own predicts account exit poorly. Academic studies of NPS found better prediction from several indicators together, and in B2B the score depends on who answered. Use the decision maker’s score as one of the signals and treat their silence in consecutive surveys as weak engagement.

How often should you recalibrate the health score?

Recalculate the score with every new signal or at least once a week, and recalibrate the weights every quarter with the outcome of that period’s renewals. Changes in product, customer profile or pricing call for recalibration outside the calendar.

How do you prepare the CS team to use the health score?

Train the team on the real portfolio: building the calibration base, logging signals with dates and holding the conversation with the sponsor in the yellow band. CS leadership starts opening the weekly review with the score’s lists, and HR tracks the share of red accounts with a recorded plan.