Startup Customer Health Scores: A Simple Model
Most customers do not leave without warning. Logins drop, support tickets pile up, the champion goes quiet. A customer health score pulls those signals into one simple number or color so your team can spot trouble weeks before a cancellation email. You do not need a data team to build one; you need a handful of honest signals, a spreadsheet and a habit of acting on what it shows.
Key takeaways
- A customer health score combines a few signals, such as usage, support experience and relationship strength, into one easy-to-read rating.
- Start with 4 to 6 signals you can measure reliably. Simple and trusted beats clever and ignored.
- Use clear bands, such as green, yellow and red, and attach a specific action to each one.
- Check the score against reality every quarter and adjust the weights when it misses churn or flags healthy accounts.
Customer health scores are common in software companies, but the idea works for any business with recurring customers: agencies on retainers, subscription boxes, managed service providers and B2B service firms. The goal is the same everywhere. You want to know which customers are thriving, which are drifting and where a phone call this week could save a relationship.
Early September is a good moment to set one up. Many teams are heading into the final quarter with renewals and budget conversations ahead, and a working score gives you a few months of signal before those decisions land. This guide walks you through a simple model you can build in an afternoon and improve over time.
What a customer health score is and is not
A health score is a summary indicator, usually a number from 0 to 100 or a color band, that estimates how likely a customer is to stay, grow or leave. It is built from signals you already have: product usage, support history, payment behavior, survey feedback and the quality of your relationship.
It is not a prediction you can bank on. Think of it like a dashboard warning light. It tells you to look under the hood; it does not tell you exactly what is wrong. The real value comes from the conversations and fixes it triggers.
Why simple models win at startups
- People understand them. If your team cannot explain why an account is red, they will not trust the score.
- The data is limited. With dozens or a few hundred customers, complex statistics can mislead you.
- They are easy to change. You will learn which signals matter, and a simple model lets you adjust quickly.
Choose your signals
The best signals are the ones that genuinely reflect whether a customer is getting value. Pick from these categories, and choose only what you can measure consistently for every account.
Usage and adoption
How often customers log in or use the service, how many of their seats are active, and whether they use the features tied to the outcome they bought you for. For a service business, this might be meetings attended, requests submitted or deliverables approved.
Support experience
The number of open tickets, how many are urgent and whether the same issue keeps coming back. A few tickets can be a healthy sign of engagement; a steady stream of frustrated ones is not.
Relationship strength
Do you have an engaged champion? Have you met a decision maker recently? Did your main contact leave the company? Relationship changes are often the earliest warning of churn.
Sentiment
Survey results, comments in check-in calls and the tone of recent emails. A simple question such as "How likely are you to recommend us?" gives you a repeatable data point.
Commercial signals
Late payments, requests to downgrade, questions about contract terms or a pause in expansion conversations.
Common mistake
Using only product logins. Some customers log in daily and still leave because the champion moved on or the budget was cut. Others log in rarely because your product runs quietly in the background. Mix usage with relationship and sentiment signals to get a fair picture.
Build the model step by step
Here is a straightforward way to turn signals into a score.
- List your signals. Pick 4 to 6 from the categories above and write a one-line definition for each, such as "active seats as a share of paid seats, last 30 days."
- Score each signal from 0 to 10. Define what 0, 5 and 10 mean in plain language so different people score the same way.
- Assign weights. Give each signal a weight so the total adds up to 100%. Put more weight on the signals you believe matter most for your customers.
- Calculate the score. Multiply each signal score by its weight, add them up and multiply by 10 to get a number from 0 to 100.
- Set bands. For example, 70 or above is green, 40 to 69 is yellow and below 40 is red. These cut-offs are a starting point, not a rule.
- Attach actions. Decide what happens for each band, who does it and how fast.
- Test it on history. Score a few customers who left and a few who renewed happily. If the model would have flagged the churned ones, you are on the right track.
A customer health score template
Copy this into a spreadsheet. Each row is a customer; each signal column holds a 0 to 10 score. The weights and customers below are made-up examples for illustration.
| Customer | Usage (30% weight) | Support (20% weight) | Relationship (25% weight) | Sentiment (15% weight) | Payments (10% weight) | Health score | Band | Next action |
|---|---|---|---|---|---|---|---|---|
| Example Co A | 9 | 8 | 8 | 9 | 10 | 87 | Green | Ask for a case study or referral |
| Example Co B | 6 | 4 | 5 | 6 | 10 | 58 | Yellow | Book a check-in call within 2 weeks |
| Example Co C | 3 | 3 | 2 | 4 | 5 | 31 | Red | Founder call this week and recovery plan |
The formula for each row is: (usage x 0.30 + support x 0.20 + relationship x 0.25 + sentiment x 0.15 + payments x 0.10) x 10. Add a "Last month" column and a "Change" column so you can see which accounts are slipping, even if they have not changed color yet.
Hypothetical example
Imagine Sana, who runs customer success at a 15-person startup selling inventory software to independent retailers. She scored 80 accounts with the five signals above. Seven came out red. When she looked closer, four of those had lost their main contact in the last two months, something the team had not noticed. She raised the relationship weight, started a "new champion" outreach email for any account whose contact changed, and set a monthly review. Over the next quarter, two of those red accounts moved back to yellow after onboarding sessions with their new contacts. (Names and numbers are made up for illustration.)
Turn scores into action
A health score that nobody acts on is just a pretty spreadsheet. Give each band a clear playbook.
Green accounts
Keep them happy and learn from them. Ask for reviews, referrals or case studies, explore expansion where it genuinely helps the customer, and note what they have in common so you can repeat it with others.
Yellow accounts
Investigate within a set time. Book a check-in, ask what has changed, offer training or a quick win and agree a next step. Many yellow accounts just need attention and a reminder of the value they are getting.
Red accounts
Escalate fast. A senior person, often a founder at an early-stage startup, should reach out personally. Listen first, then agree a short recovery plan with clear owners and dates on both sides.
Pro tip
Hold a 30-minute health review every week or two. Look only at accounts that changed band or dropped sharply. Short, regular reviews keep the score alive and stop red accounts from sitting unnoticed.
Tools that can help
A spreadsheet is enough to start. As you grow, customer health tools such as HubSpot, Vitally and ChurnZero can pull usage and support data in automatically and alert your team when scores change. Check current features, pricing and privacy policies before connecting customer data, and make sure any tracking you add is covered by your privacy notice, especially for users in the UK and EU, where GDPR applies.
Keep the model honest
Your first model will be imperfect, and that is fine. What matters is that you improve it with evidence.
- Review every quarter. Compare the scores with what actually happened. Did customers who left show up as red or yellow beforehand? Were any green accounts surprises?
- Adjust one thing at a time. Change a weight or a signal definition, then watch for a quarter. Changing everything at once makes it impossible to learn.
- Ask the front line. Account managers and support staff often know which signals the score misses.
- Segment if needed. Small customers and large customers may behave very differently. Separate weights for each segment can make the score fairer.
- Watch for bias. If a signal depends on someone's opinion, write clear definitions so scores do not depend on who fills them in.
If your company sets quarterly OKRs, a health-related key result, such as reducing the number of red accounts, can keep the work on the agenda without becoming a vanity metric.
Your customer health score checklist
- Pick 4 to 6 signals you can measure for every customer
- Write a plain definition and a 0 to 10 scale for each signal
- Assign weights that add up to 100%
- Build the spreadsheet and score every active customer
- Set green, yellow and red bands with a named action for each
- Test the model against customers who left and customers who renewed
- Hold a short health review every week or two
- Review weights and signals every quarter
Frequently asked questions
How many customers do I need before a health score is useful?
Even a few dozen customers can benefit, because the score organizes what you already know and makes trends visible. With very few customers, you may simply review each account by hand, using the same signals as a checklist.
Should customers see their own health score?
Usually not as a raw number, because it reflects your internal view and can feel like a judgment. Instead, share the useful parts, such as adoption progress or goals reached, during regular check-ins.
What if my signals are mostly opinions?
That is normal early on. Define each opinion-based signal clearly, such as "met a decision maker in the last 90 days, yes or no," and add measurable data like usage or payment history as soon as you can collect it.
How is this different from a net promoter survey?
A recommendation survey is one sentiment signal. A health score combines several signals, including behavior and relationship strength, so it can catch risk even when a customer has not answered a survey.
Related reading
The bottom line
A customer health score does not need to be complicated to be valuable. Choose a few honest signals, score them consistently, weight them sensibly, set clear bands and, above all, attach an action to every color. Then check the model against reality each quarter and refine it.
Build your first version this week with the template above. Within a month, you will know which customers need a call today, and you will have a much calmer view of the renewals ahead.
