Run quarterly health score reviews

Put the review in the diary for the first two weeks after each quarter ends. Ninety minutes, the same people every time: whoever owns retention, whoever owns the account relationships, and someone who can pull data. Without a fixed date it slips, and a score nobody checks stops being used.

Send the data pack two days ahead so the meeting is about decisions, not about finding numbers.

Gather four things for every customer: the health score at the start of the quarter, the score at the end, whether they renewed, downgraded or cancelled, and the revenue involved. A spreadsheet is enough. If you use a customer success tool such as Gainsight, Planhat or ChurnZero, export the same columns.

Add a fifth column for the reason, in one line, for every account that left or shrank. Fill it in from the account owner's notes before the meeting.

Sort the accounts that churned by the score they had 90 days before they left. This is the test that matters. If most of them were already red, the score works. If half of them were green, it does not.

Then do the reverse. Look at the red accounts that renewed anyway. A few are normal. A lot of them means you are raising false alarms, and your team will stop believing the colour.

Look at each input of the score on its own. Did falling logins come before cancellations? Did late payments? Did a drop in support tickets, which can mean a customer has given up on you? Keep a simple tally for each signal: how many churned accounts showed it, and how many healthy accounts showed it too.

Raise the weight of signals that appear mostly among the churned. Lower the weight of the ones that appear everywhere. If you are still choosing the inputs, How to monitor customer health covers that step.

Individual saves belong to Proactive outreach to at-risk accounts. This meeting looks for what repeats. Group the losses by plan, industry, company size, onboarding date, account owner and the feature they never used.

One churned account is a story. Four from the same segment is a finding. Write each finding as a sentence a colleague could act on, such as "Customers on the smallest plan who never connect their CRM leave within five months."

Make one to three changes per quarter, and write down what you changed and why. Move a weight, add a signal, retire one that never helped. Big rewrites make it impossible to tell whether the next quarter improved because of the model or because of luck.

Keep a change log in the same sheet. In a year it shows you how the score matured.

Choose two or three numbers and give each an owner. For example: reduce the share of red accounts that cancel from 40 per cent to 30 per cent, get every new customer to a first outcome within 21 days, and contact every account that turned red within five working days.

Base targets on this quarter's actual figures, not on a number from an article. A target you can see yourself hitting gets used.

• Reviewing only the accounts that are red today and never checking old predictions.
• Changing five weights at once.
• Treating the score as the answer instead of a prompt to talk to the customer.
• Skipping the review because the quarter was busy, which is exactly when the score drifts.
• Holding the meeting without account owners, so the reasons behind each churn are guesses.

Two checks. First, most customers who left were already flagged amber or red 90 days before. Second, your team acts on a red account within days without being chased. When both are true, the review has made the score something people rely on, and the retention targets become believable.