How to Build a Weighted Customer Health Score: Categories, Weights and Bands

Almost every customer success team has a health score. Far fewer have one they could rebuild from scratch and defend line by line. The score usually started as a quick formula in a spreadsheet, picked up new inputs over time, and now produces a number that nobody fully understands.

A weighted health score fixes that by making every design choice explicit: which categories count, which metrics feed each category, how much each category matters and where the cutoffs sit. This guide walks through building one step by step, from deciding what the score should predict to testing it against real renewals.

Step 1: Decide What the Score Is For

Before you pick a single metric, write down the question the score answers. For most post-sales teams, it is some version of: "How likely is this account to renew, and is it ready to grow?" That sentence becomes the test for every input. If a metric doesn't help answer it, leave it out.

Two more decisions belong here:

• One score per account. Contact-level and product-level scores can be useful, but the score your team triages from should roll up to the account that holds the contract.
• Who acts on it. A score built for CSMs needs to be explainable in a renewal call. A score built for leadership needs to roll up cleanly into at-risk revenue. Ideally, the same model serves both.

Step 2: Choose Your Scoring Categories

Categories are the main ingredients of the score. Grouping metrics into a handful of categories keeps the model readable: you can say "support is dragging this account down" instead of listing twelve metrics. Quivly's Health Score ships with five default categories, which cover the signals most B2B teams care about:

1. Engagement: how actively the customer talks to you, such as Slack conversations, calls and meeting attendance.
2. Product Usage: whether the product is actually being used, and whether that use is growing or shrinking.
3. Support: ticket volume, severity and how long issues stay open.
4. Revenue: billing and contract health, including payment status, plan changes and renewal timing.
5. Market Signals: what is happening at the customer's company, such as funding, hiring, leadership changes and news.

You don't have to use all five, and you can rename or swap them. A product-led business might split Product Usage into breadth (how many users) and depth (which features). A services-heavy business might add an onboarding or delivery category. Aim for four to six categories. Fewer than that hides too much; more than that becomes hard to explain.

Step 3: Pick and Normalize the Metrics Inside Each Category

Each category needs one or more concrete metrics, pulled from the system where that data lives. Product usage might come from a data warehouse like BigQuery, engagement from Slack or call recordings, revenue from Stripe, and account details from Salesforce or HubSpot.

Prefer trends over raw totals

A raw count rarely tells you much on its own. Two hundred weekly executions is excellent for a small team and alarming for a large one. Percent change over a set window is usually more telling. Quivly's own example configuration pairs each usage metric with its percent change, for instance "V2 Executions" alongside "V2 Executions Percent Change."

Convert every metric to the same scale

Categories can only be weighted fairly if every metric lands on a common scale, typically 0 to 100. Common ways to get there:

• Thresholds: define what "bad," "okay" and "great" look like and map each to a score range.
• Peer percentiles: score each account against similar accounts by size, plan or industry.
• Change from baseline: compare the account to its own history, so a sharp drop scores low even if the absolute number still looks fine.

Be selective. Every metric you add is one more thing someone has to explain when the score moves. If two metrics always move together, keep the one that's easier to understand.

Step 4: Assign Category Weights

Weights tell the model how much each category matters for your business. The composite score is a weighted average: multiply each category score by its weight, then add the results.

A worked example

The live example on Quivly's Health Score page shows one account with the following category weights and scores. Treat the weights as an illustration of the math, not a recommendation for your business.

Notice what the weights do. Product usage is nearly perfect at 98, but because Revenue carries the largest weight and scores 35, the account lands in the middle. That is the model working as designed: in this example, the team has decided that billing trouble matters more than heavy usage.

How to choose your own weights

• Start simple. If you have no history to go on, begin with roughly equal weights or a sensible default, then adjust.
• Look backward. Pull the accounts that churned or downgraded in the last year and ask which categories moved first. Those deserve more weight.
• Keep weights summing to 100%. It makes contributions easy to read and compare.
• Avoid a single dominant category. If one category carries most of the weight, you have built a single-metric score with extra steps.

Step 5: Set Score Bands Your Team Can Act On

A number from 0 to 100 is hard to act on. Bands turn it into a decision. Quivly's default model uses four bands, each spanning 25 points:

• Rescue: 0 to 25
• Protect: 25 to 50
• Sustain: 50 to 75
• Grow: 75 to 100

Even splits are a fine starting point, but the cutoffs should reflect how your team actually triages. A few ways to tune them:

1. Check the distribution. If most of your book lands in one band, the cutoffs are not separating accounts in a useful way.
2. Match capacity. The Rescue band should hold roughly as many accounts as your team can genuinely give urgent attention to.
3. Validate against outcomes. Accounts that churned should mostly have sat in the lower bands in the months before renewal.

Designing what happens inside each band is its own exercise. Here, the goal is simply cutoffs that sort accounts in a way your team agrees with.

Step 6: Handle Gaps, Explain Every Score and Keep It Calibrated

Plan for missing data

Some accounts won't have every data source connected. A model that treats missing data as zero will punish those accounts unfairly. A better approach, and the one Quivly uses, is to exclude categories with no inputs, rebalance the remaining weights automatically and flag scores driven by partial data with a confidence indicator.

Make every change explainable

The most common reason teams stop trusting a health score is that nobody can say why it moved. Every score should show its category contributions, what changed since the last calculation and a plain-English summary of the drivers. Quivly generates these narratives with each claim cited back to its source system, so a CSM can check the evidence before acting. For more on why opacity kills adoption, see why complex health scores lose the team's trust.

Test, version and revisit

• Back-test before you switch. Run proposed weights against historical data and compare which accounts would have been flagged.
• Version every change. Keep a record of each configuration so you can compare results and roll back if a change makes things worse.
• Review on a schedule. Revisit weights and bands as your product, pricing and customer base change, and after any renewal cycle that surprised you.
• Put the score where people work. Writing the score back to a CRM custom field in Salesforce or HubSpot keeps one source of truth across reports and workflows.