Customer Scoring
Customer scoring assigns a numeric value to each account based on attributes like fit, potential value, product usage, and churn risk, helping teams decide where to focus effort.
Key takeaways
- Inputs are normalised to a common range before weighting, or currency values dominate.
- Judgment-set weights explain easily; fitted weights predict better and resist explanation.
- Score crossed with renewal date orders the work queue better than score alone.
- Three or four bands with defined plays beat a continuous number without actions.
- Thresholds fitted on large accounts mislabel healthy small accounts as at risk.
In depth
A customer score is a weighted sum over an account rather than a person. Each input is normalised to a common range so that a contract value in currency and a support-ticket count can be added together at all, then multiplied by a weight and totalled into a number that is usually banded into tiers. The weights come either from judgment, where a team asserts what matters, or from fitting historical accounts against a known outcome such as renewal, expansion or churn.
The score moves with whichever inputs carry the most weight, which makes weight setting the whole design problem. Depth of use across teams usually predicts renewal better than raw login volume; contract size predicts revenue impact but says nothing about risk. Judgment-based weights are transparent and easy to defend but drift from reality; fitted weights track outcomes better and become hard to explain to the account manager who has to act on them. Most teams accept that trade and refit periodically.
In use, the score is a queue rather than a verdict. Customer success sorts by score crossed with renewal date, so a low-scoring account renewing next month outranks a low-scoring account renewing next year. Bands matter more than the decimal: three or four tiers with defined plays beat a continuous number nobody knows how to act on. Where a Pivix quiz tiered an account at capture, that original fit tier can stay on the record as one input among several after the sale.
A score built on one segment misleads on another: usage thresholds derived from enterprise accounts will mark perfectly healthy small accounts as at risk. It also lags by construction, because most inputs are recorded after behaviour has already changed, so a score can look stable through the month a champion quietly leaves. And scores describe correlation, not cause. Acting on a low score without diagnosing why it is low produces the same intervention for accounts with entirely different problems.
Example in practice
How to measure it
Test the score the way you would test a forecast. Take accounts as they were scored some months ago, then compare what actually happened, renewal, expansion or churn, grouped by the band they held at that time. A useful score shows a clear gradient across bands. If the top and bottom bands churn at similar rates, the weights carry no information and a refit is overdue.
Alongside accuracy, watch coverage and stability. Coverage is the share of accounts with enough input data to be scored at all, because unscored accounts silently leave the queue. Stability is how often an account changes band with nothing having happened, since a score that flickers week to week trains the team to ignore it and usually points to one noisy input carrying too much weight.
Common mistakes
The most damaging habit is scoring activity that costs the customer nothing. Logins, page views and email opens all rise when an administrator clears notifications, so an account can look engaged while the people who decide renewal never sign in. Weight actions that require effort and imply value received instead: a workflow built, a report scheduled, a second team onboarded. Then check each weight against actual renewal outcomes before trusting it.
The second is publishing the score without publishing the plays. An account drops from green to amber, the dashboard shows it, and nobody has agreed what happens next, so within a quarter the alert has become background noise. Attach a specific action to each band and to each transition between bands, name the owner, and check whether the play was run at all before concluding the score was wrong.
Frequently asked questions
How is customer scoring different from lead scoring?
Lead scoring predicts the likelihood that a prospect will buy, while customer scoring evaluates existing accounts for expansion, advocacy, or churn risk. One serves acquisition; the other serves retention and growth.
What inputs go into a customer score?
Typical inputs include firmographic fit, product usage depth, contract value, support sentiment, and renewal proximity. The right mix depends on which signals actually correlate with value in your data.
Can quiz tiers feed a customer score?
Yes. The Hot, Warm, or Cold tier a respondent earns in a Pivix quiz captures fit at the start of the relationship. That tier can carry into a post-sale score to prioritize onboarding and success efforts.
How is customer scoring different from a health score?
In most stacks they are the same object under different names, though health scores lean toward churn risk while customer scores often include value and expansion potential too. What matters is which outcome the weights were fitted against. A score built on churn will not rank expansion candidates well, and using one for the other is a common source of disappointment.
Should the score be a number or a colour band?
Publish bands and keep the number underneath. A continuous score implies precision the inputs do not support and invites arguments about whether sixty-eight beats sixty-four. Bands force the useful question, which is what changes when an account crosses a boundary. Keep the underlying number available so trends can still be read over time.
What inputs actually predict churn?
Usually breadth rather than volume: how many teams use the product, how many workflows depend on it, and whether the original champion is still in place. Support sentiment and unresolved escalations add signal. Raw login counts predict poorly on their own, because they rise with administrative activity that has nothing to do with value received.
How often should scores be recalculated?
Often enough to catch a real change, which for most inputs means weekly or daily. The weights themselves should be refitted far less often, perhaps once or twice a year, and only against enough completed renewals to be meaningful. Refitting weights every month chases noise and destroys the team's trust in the ranking.
Can a lead score carry over into a customer score?
The fit portion can. Firmographic fit measured before the sale does not change at signature, so it stays a legitimate input afterwards. What cannot carry over is intent, which describes buying readiness and stops meaning anything once the purchase has happened. Keep the two as separate fields rather than merging them into one running total.
What do I do about accounts with too little data to score?
Give them their own category rather than a default score. An unscored account is usually new, small or lightly instrumented, and assigning it a middling number buries it in the middle of the ranking. Route them by a simple rule instead, such as contract value or days since onboarding, until enough usage data accumulates.