Customer health score: what it is and how to calculate one
Short answer: A customer health score is one number that rates how likely an account is to stay, built from signals such as how fast the customer replies, whether meetings happen and how much of the product they use. To calculate one, turn each signal into a 0 to 100 score, give each a weight so the weights total 100%, then multiply and add. This page shows that working on a made-up account, gives a template, and says where a health score stops and churn prediction starts.
What is a customer health score?
A customer health score is a single rating of how healthy a customer account is, usually 0 to 100, built from several signals and used to decide which accounts need attention first. It's a judgement made from data, so two teams can score the same account differently.
Customer success teams use the score to rank accounts, spot change early and decide where to spend limited time. Gainsight describes its health scorecard as "the best way to get a complete picture of your customer's satisfaction and likelihood to renew"1. A score is a hypothesis about which signals matter, and it earns trust only after you check it against accounts that renewed and accounts that left.
Sources: Gainsight, checked 26 October 2026.
1 gainsight.com/glossary/customer-health-score
What metrics go into a customer health score?
A score can draw on six groups: product usage, feature adoption, sentiment, engagement, relationships and customer value. Start with a few signals you can collect.
Gainsight's guide, "Essential Considerations For Your First Customer Scorecard", suggests starting with six areas: product usage measured against entitlements, feature adoption, customer sentiment, engagement (for example call attendance and follow-through on executive business reviews), relationships (for example whether there is an active executive sponsor) and customer value1. It also advises keeping the data focused to avoid noise.
The table maps those groups to three places a CS team can usually read data from:
| Source | Example signals | What it tells you |
|---|---|---|
| Reply time, silence in business days, sentiment of recent messages | Whether the customer is still talking to you, and how they feel | |
| Calendar | Meetings held against cancelled, executive attendance | Whether the relationship is being kept up on both sides |
| Product usage | Active users against licensed seats, use of the key features | Whether the customer is getting value |
Sources: Gainsight, checked 26 October 2026.
1 gainsight.com/glossary/customer-health-score
How do you calculate a customer health score, step by step?
Pick the outcome, choose your signals, write a 0 to 100 rule for each, weight them to total 100%, then multiply each score by its weight and add the results. Test the result against accounts whose outcome you already know.
- Pick the outcome. Decide what the score should warn about, such as a lost renewal or a downgrade. Everything else is judged against this.
- Choose signals you can get. A handful is plenty. Use only data you can pull every week.
- Write a rule that turns each signal into 0 to 100. For example, "customer silent for 0 to 3 business days scores 100". Write the rule down so anyone can reproduce the number.
- Give each signal a weight. The weights must total 100%. Forrester notes that "Not all metrics carry the same weight in every organization"2, so expect to adjust them.
- Multiply and add. Score times weight gives the points for each signal. The points add up to the health score.
- Test it, then set your own bands. Forrester's fourth step is to validate the scorecard against historical accounts that were retained or lost2. Choose your own cut-offs from that test. Don't copy someone else's.
- Review and refine. Compare the score with what customers actually do, and change the rules where they disagree. Forrester also suggests separate score profiles for segments such as high-touch and low-touch accounts2.
Sources: Forrester, checked 26 October 2026.
2 forrester.com/blogs/creating-a-customer-health-score-in-five-steps
What does a worked example look like in numbers?
For the made-up account below, the weighted scores add up to 52.6 out of 100, and email loses the most points. The signals, rules and weights are illustrative only.
Illustrative method only. This is NOT Kelpie's scoring. The account, the numbers, the rules and the weights below are all made up to show the arithmetic. They are not Kelpie's own weights, which aren't published. The example weights email at 40%, calendar at 30% and product usage at 30%.
Made-up example account: "Example Co", 30 licensed seats, one executive sponsor.
First, the rules, written down so the numbers can be reproduced:
| Signal | Source | Rule (made up) |
|---|---|---|
| Reply time | Customer's typical reply time in business days: 1 or less scores 100, 2 scores 70, 3 scores 40, 4 or more scores 10 | |
| Silence | Business days since the customer last wrote: 0 to 3 scores 100, 4 to 7 scores 50, 8 or more scores 0 | |
| Sentiment | Last 5 customer emails: positive 100, neutral 60, negative 0, then average | |
| Meetings held | Calendar | Meetings held ÷ meetings scheduled in the last 60 days × 100 |
| Exec attendance | Calendar | Held meetings the executive sponsor attended ÷ meetings held × 100 |
| Active users | Product usage | Users active in the last 30 days ÷ licensed seats × 100 |
| Key feature use | Product usage | Key features used in the last 30 days ÷ key features chosen × 100 |
Now the working for Example Co. Points are score × weight.
| Signal | What we observed | Score (0 to 100) | Weight | Points |
|---|---|---|---|---|
| Reply time | 3 business days | 40 | 15% | 40 × 0.15 = 6.00 |
| Silence | 6 business days since last email | 50 | 15% | 50 × 0.15 = 7.50 |
| Sentiment | 2 positive, 2 neutral, 1 negative | (100 + 100 + 60 + 60 + 0) ÷ 5 = 64 | 10% | 64 × 0.10 = 6.40 |
| Meetings held | 3 of 4 scheduled meetings held | 3 ÷ 4 × 100 = 75 | 15% | 75 × 0.15 = 11.25 |
| Exec attendance | Sponsor at 1 of 3 held meetings | 1 ÷ 3 × 100 = 33 (rounded) | 15% | 33 × 0.15 = 4.95 |
| Active users | 18 of 30 seats active | 18 ÷ 30 × 100 = 60 | 15% | 60 × 0.15 = 9.00 |
| Key feature use | 2 of 4 key features used | 2 ÷ 4 × 100 = 50 | 15% | 50 × 0.15 = 7.50 |
| Health score | 100% | 52.60 out of 100 |
By source: email gives 19.9 of a possible 40 points, calendar 16.2 of 30 and product usage 16.5 of 30. Email lost the most (20.1 points), so that is where to look first.
A what-if shows how to read it. If Example Co's reply time improved to one business day, that signal would score 100 and add 9 points (60 × 0.15), taking the total to 61.6. Whether 52.6 or 61.6 counts as healthy depends on your own test against past accounts. This example sets no bands, because yours should come from your own data.
What does a customer health score template look like?
A template is a table with one row per signal and columns for source, how to measure it, the 0 to 100 rule, the weight, the score and the points. Copy the table below into a spreadsheet and fill in the blanks.
| Signal | Source | How to measure | Rule (0 to 100) | Weight (%) | Score | Points |
|---|---|---|---|---|---|---|
| Reply time | Business days to reply | (yours) | ||||
| Silence | Business days since last customer message | (yours) | ||||
| Sentiment | Tone of the last few customer emails | (yours) | ||||
| Meetings held | Calendar | Held ÷ scheduled, last 60 days | (yours) | |||
| Exec attendance | Calendar | Meetings with the sponsor ÷ meetings held | (yours) | |||
| Active users | Product usage | Active users ÷ licensed seats | (yours) | |||
| Key feature use | Product usage | Key features used ÷ key features chosen | (yours) | |||
| Total | 100 | (sum of points) |
In a spreadsheet, set Points to Score × Weight ÷ 100 and the health score to the sum of the Points column. Check that the weights add up to 100. Add a column for your own bands once you have tested the score against past accounts. Keep one copy of the rules next to the sheet so the number can be reproduced.
How do health scores relate to churn prediction?
A health score says how an account looks now, using rules you write, while churn prediction estimates the chance an account will leave, usually with a model trained on past customers. A health score is the simpler first step, and both depend on good data.
Rules-based scoring, like the worked example, is easy to explain and quick to start. You choose the signals and weights, so you can say why a number moved. The cost is that the weights are your judgement until you test them.
Machine-learning churn models learn the weights from history. Vendors describe what they need. Pendo says its Predict product "connects your CRM and product usage data"1. Pecan says you need "historical, event-level data from your warehouse", though you can also upload CSV or Excel files2. Neither page gives a minimum amount of history, so ask any vendor how much past churn their model needs.
Four limits apply whichever route you take:
- Bad inputs give bad scores. Zendesk's own page says "Inaccurate or incomplete data will lead to unreliable churn scores"3.
- A prediction flags risk without giving the reason. Zendesk's page notes that spotting at-risk customers does not by itself reveal why they are leaving3.
- A model needs past examples of customers who left. A small team with few churned accounts may have too little history to train one.
- Data you do not collect cannot count. If the warning sign is in chat or a support desk and you do not feed it in, no score will see it.
Sources: vendor pages, checked 26 October 2026.
1 pendo.io/product/pendo-predict ·
2 pecan.ai/solution/customer-churn ·
3 zendesk.com/service/churn-prediction-software
What are the best customer churn prediction tools?
Pendo Predict, Pecan, Zendesk, ChurnZero and Gainsight are among the tools that rank for churn prediction searches, and they differ mainly in the data they start from. Apart from Zendesk's starting plan price, none showed a price on the pages we checked. Kelpie scores account health from email, calendar and product usage and does not claim to be a statistical churn model.
| Tool | What the vendor's page says it does | Data it starts from | Published price |
|---|---|---|---|
| Pendo Predict | Flags at-risk and expansion-ready customers and alerts reps in Salesforce and Slack1 | CRM and product usage data1 | Not listed on the page1 |
| Pecan | Platform that "identifies at-risk accounts weeks in advance"2 | Historical event-level data from a warehouse, or CSV and Excel uploads2 | Tailored quote at demo2 |
| Zendesk | Its QA AI reviews support conversations to estimate churn risk3 | Support interactions3 | Plans start at $25 per agent/month (the page does not say this covers churn prediction alone)3 |
| ChurnZero | Lists "Renewal forecasting" as a feature4 | Account, usage, support and survey data5 | Not published4 |
| Gainsight | Lists "Predict churn and address risk" under its Retention use case6 | Not stated on the pricing page | Not published ("Request Pricing")6 |
| Kelpie | Scores account health from email, calendar and product usage, with the working shown7 | Email, calendar, your meeting notes and usage by CSV7 | USD 125 a seat a month7 |
Sources: vendor pages, checked 26 October 2026.
1 pendo.io/product/pendo-predict ·
2 pecan.ai/solution/customer-churn ·
3 zendesk.com/service/churn-prediction-software ·
4 churnzero.com ·
5 churnzero.com/features/customer-health-scores ·
6 gainsight.com/pricing ·
7 Kelpie details as of 24 Oct 2026. Usage sources: kelpie.pro/trust
Treat the table as a snapshot from 26 October 2026, and confirm any figure on the vendor's own page before you rely on it. More vendors' price pages are covered in customer success software pricing.
What is the best customer health score software?
No single tool is best. Gainsight, ChurnZero, Vitally, Planhat, HubSpot, Totango (now Odie) and Kelpie all sit in this space, and the right pick depends on team size, the data you have and whether you want a published price. Kelpie lists its price. Gainsight, ChurnZero and Vitally showed no amount on the pages we checked, and HubSpot's Service Hub pricing page lists per-seat amounts, so check which tier includes health scores.
| Tool | Health scoring on the pages we checked | Published price |
|---|---|---|
| Gainsight | "Health Scorecards" included in the Essentials and Enterprise plans1 | Not published ("Request Pricing")1 |
| ChurnZero | Lists "Customer health scoring"2; scores are built from "the metrics you choose", including usage, support history and NPS, with separate scores by segment or lifecycle stage3 | Not published2 |
| Vitally | Builder with "threshold rules, weightings, and previews"4 | Not published ("Request Pricing")5 |
| Planhat | Health scores not mentioned on its home or pricing pages6 7 | Not published (quote-based)7 |
| HubSpot | Customer Success Workspace health scores with points per rule and a status label from thresholds you set. Needs Service Hub Professional or Enterprise8 | Not checked for this page |
| Totango (now Odie) | Health scores not mentioned on its home page9 | Not published9 |
| Kelpie | Scores from email, calendar and product usage, with the working shown10 | USD 125 a seat a month10 |
Sources: vendor pages, checked 26 October 2026.
1 gainsight.com/pricing ·
2 churnzero.com ·
3 churnzero.com/features/customer-health-scores ·
4 vitally.io/features/health-scores ·
5 vitally.io/pricing ·
6 planhat.com ·
7 planhat.com/pricing ·
8 knowledge.hubspot.com/help-desk/customize-a-health-score-in-the-customer-success-workspace ·
9 totango.com ·
10 Kelpie details as of 24 Oct 2026
"Not mentioned" means only that the page we fetched did not say it. Planhat and Totango may well score health elsewhere on their sites. The larger platforms cover far more than scoring, and some, such as Gainsight, also sell customer education and communities1. If you want that breadth, a scoring-focused tool won't replace it. For the comparisons, see Gainsight alternatives and ChurnZero alternatives.
What should an account health dashboard show?
A good account health dashboard shows each account's score, how it has moved, and what moved it, so a CSM can decide who to contact today. A score with no explanation sends people hunting through other tools to find the reason.
Five things to look for in any dashboard:
- Each account's score, and whether it is rising or falling. A number alone says little.
- Which signal moved the score: silence, a cancelled meeting or a drop in usage.
- The rules behind the number. If you cannot see how it was built, you cannot argue with it.
- A preview before the rules go live. Vitally says you can preview how rules affect real accounts before they go live1.
- Access for managers, who should be able to view the dashboard without paying for a full seat.
Kelpie's dashboard shows each account's score and the working behind it. Managers who only view the dashboard are free.
Sources: Vitally, checked 26 October 2026.
1 vitally.io/features/health-scores
How does Kelpie score customer health?
Kelpie scores each account from email, calendar and product usage, shows the working behind every score, and is recommended for customer success teams of up to 15. It costs USD 125 a seat a month, where a seat is a mailbox Kelpie reads.
- Three inputs. Email, calendar and product usage. Kelpie also uses your meeting notes. Usage goes in as a CSV upload today. Segment, Mixpanel, Amplitude and PostHog are coming, and none is live yet.
- Public holidays. Kelpie counts Monday to Friday. A public holiday still counts.
- The working is shown. Each score shows what moved it, including whether usage or the conversation did.
- Where you use it. Kelpie's own API distils the mail. ChatGPT, Claude or Grok, on any plan including the free ones, is only the window where you read and write. The connector between them reads and writes.
- Price. USD 125 a seat a month, at the same price however many emails the mailbox gets. No platform fee, no minimum and no credits, so you know the bill upfront. Managers who only view the dashboard are free.
Kelpie's own weights aren't published, and the worked example above is not them.
When is Kelpie not the right choice?
Kelpie is not the right choice if you need a statistical churn model, a CS team bigger than the 15 Kelpie is recommended for, a security certificate or live usage feeds today. The details follow.
- Kelpie scores account health from email, calendar and product usage. We do not claim it produces a statistical churn forecast. Statistical churn models need a lot of churn history, which small teams usually don't have, so Kelpie scores health from signals such as engagement instead.
- Kelpie is recommended for CS teams of up to 15, so a bigger team may be better served by the larger platforms above.
- Kelpie is early, and its trust page says "no SOC 2 or ISO certification". If procurement requires a certificate, that rules Kelpie out for now.
- You bring usage in by CSV for now, and the direct integrations are not live.
- Chat, your CRM and your support desk are outside what Kelpie reads, per its trust page. A warning sign that appears only there will be missed.
- Kelpie focuses on account scoring. If you want one system with a wider reach, look at the larger platforms above.
Frequently asked questions
What is a good customer health score?
There is no universal good number, because every team builds its own scale. Test your score against accounts that renewed and accounts that left, then set your own bands from that. A cut-off copied from another company may mean nothing for your customers.
How do you calculate a customer health score?
Turn each signal into a 0 to 100 score with a written rule, give each a weight so the weights total 100%, multiply each score by its weight and add the results. The worked example above shows the arithmetic for a made-up account, which came to 52.6 out of 100.
What is the difference between a health score and churn prediction?
A health score rates how an account looks now using rules you write. Churn prediction estimates the chance an account leaves, usually with a model trained on past customers. A health score is the simpler starting point.
How many metrics should a health score use?
Start with a small number you can collect every week. Gainsight advises keeping the data focused to avoid noise. Add signals later once the score has proved useful.
Can you build a customer health score in a spreadsheet?
Yes. The template above is a table you can copy into a spreadsheet: one row per signal, with the rule, weight, score and points. The hard part is pulling the signals in every week, which is where software helps.
Does Kelpie use product usage data?
Yes. Kelpie scores from product usage as well as email and calendar. Usage goes in as a CSV today, and Segment, Mixpanel, Amplitude and PostHog are coming.
What does Kelpie cost?
USD 125 a seat a month. A seat is a mailbox Kelpie reads, at the same price however many emails it gets. There is no platform fee and no minimum. Managers who only view the dashboard are free.
Is Kelpie SOC 2 certified?
No. Kelpie is early, and it is not SOC 2 certified. What is and is not built is listed on Kelpie's trust page.
See it on your own accounts
Bring a few of your own accounts to a 30-minute walkthrough. Kelpie scores them and shows the working, so you can check the numbers against what you already know.
More: Kelpie home · Gainsight alternatives · ChurnZero alternatives · Customer success software pricing · Kelpie inside ChatGPT, Claude or Grok · Trust