Stop Scoring Accounts. Start Reading them
The CS industry has a health score addiction.
Every platform sells one. Every team builds one. Weighted metrics, colour-coded dashboards, algorithms that blend usage, NPS, support tickets, and engagement into a single number that’s supposed to tell you whether an account is healthy.
And then you use it for a quarter and realise it’s telling you almost nothing useful.
The green accounts churn. The red accounts renew without a fuss. The amber ones sit there being amber forever, which is about as actionable as a weather forecast that says “might rain.” You end up ignoring the score and going with your gut anyway, which is exactly what the score was supposed to replace.
It’s not that the data doesn’t matter. It’s the opposite. The way most teams package it strips out everything that actually helps you make decisions. A number can’t tell you that your champion just changed roles. A colour can’t tell you that the customer’s CEO mentioned “vendor consolidation” on their last earnings call. A weighted average can’t tell you that the account feels different to how it felt six months ago, in ways that matter but don’t fit neatly into a formula. The more data you try and blend into a perfect score, the more convoluted and distorted that score becomes. You can hide a multitude of churn indicators behind strong, growing and “healthy” usage, for example.
The alternative isn’t ditching the data. It’s better data, gathered more intentionally, with a process that forces you to actually think about each account rather than glance at a dashboard and move on.
9 Data Points. That’s It.
Here’s everything you need to assess an account properly. Nine data points, split between things you can pull from your systems and things that require a human being to observe.
From your systems:
Last executive engagement date. When did someone senior at this account last have a real interaction with you? Not product usage. A conversation. What does “senior” mean to you in this context?
Product breadth. How many of your products or features are they actually using? This tells you embedding depth without needing a complex integration score.
Usage trend direction. Up, flat, or down. You don’t need the absolute numbers. You need to know which way it’s moving. Beyond raw usage, look for trigger signs in product usage. New users, new projects, new activity that wasn’t there before. Look back at your last 10 accounts that churned, what did they have in common in interaction and product usage patterns? Use those patterns to spot trends you would have missed from looking at pure usage alone.
Support ticket sentiment. Any frustrated or competitive mentions in recent tickets? You’re looking for signal, not volume.
Payment history. Any late payments or disputes? This is one of the most underrated early warning signals in CS.
From your conversations:
Can the customer articulate the value they get from you? Not in technical terms. In business terms. Yes, no, or unknown.
Champion status. Named, engaged, still in role? Yes, no, or unknown.
Expansion signal. Are they building new things where you could help? Yes or no. Are you helping (or are they talking to you about what they’re building)? Yes, no or unknown.
Competitive chatter. Any mention of alternatives or pricing concerns? Yes or no.
That’s it. No complex scoring. No weighting. No spreadsheet with thirty columns that nobody updates after the first month.
Five things from your tools. Four things from being a human who talks to customers. And the four conversation-based ones are almost always more predictive than the five automated ones, which should tell you something about where the real signal lives.
The 15-Minute Quarterly Review
The data points above feed into the three questions I covered in the first post in this series: Who Cares (stakeholder health), How Deep (switching cost), and What’s Changing (momentum). Here’s how the whole thing works in practice.
Step 1: Gather (5 minutes). Pull the automated data. Review your recent calls, emails, and notes for the manual data points. If you use call recording or AI summaries, skim those for sentiment.
Step 2: Answer the 3 Questions (5 minutes). Who Cares? Rate stakeholder health as green, amber, or red. How Deep? Rate switching cost as low, medium, or high. What’s Changing? Identify momentum as positive, negative, or flat.
Step 3: Decide and Act (5 minutes). Based on where the account sits, determine your next action. Log it. Commit to it.
Fifteen minutes per account per quarter. For a book of 25 accounts, that’s about six hours of work every 90 days. That is a very small investment for knowing, with real confidence, where your risks and opportunities actually are.
Why Simple Beats Sophisticated
There’s a reason health scores don’t work well, and it’s not that the data is wrong. It’s that the aggregation destroys the signal.
When you blend five or ten inputs into a single number, you lose the ability to see which input matters for this account right now. An account with perfect usage but no executive relationship looks the same as an account with mediocre usage but a brilliant champion. They both score “72” or whatever. But the actions you’d take are completely different, and a number can’t tell you that.
Keeping the inputs separate, and forcing yourself to interpret them rather than letting a formula do it is the whole point. It’s more work than glancing at a dashboard. But it’s dramatically less work than rebuilding stakeholder relationships after a churn event you should have seen coming.
The other advantage of simplicity is that people actually do it. I have never seen a 30-metric health score model that was still being maintained after six months. I have seen plenty of teams sustain a lightweight quarterly review process for years, because it’s quick enough to be realistic and useful enough to be worth the time.
There’s also merit in a model that tells you what you’re missing and where to focus. One of the reasons I like MEDDICC so much is because it’s simple, and if you’re missing an M you know that metrics is a weak spot. This framework works like that, because if you don’t know the answer to a question, the first job on your list should be to go and find that out. People who ask great discovery questions are the ones that know which direction they want to travel in.
Perfection is the enemy
You have to make a call based on the data you have. That’s true in CS. It’s true in sales. It’s true in every forecasting exercise you’ll ever do.
The temptation is always to add more data, more inputs, more sophistication. One more metric and we’ll really understand what’s going on. But the marginal return on additional complexity drops off fast, and the maintenance cost of keeping it all current goes up faster.
Nine data points. Three questions. Fifteen minutes. A clear action for every account.
You will sometimes get it wrong. An account you rated as low risk will surprise you. A deal you were confident about will stall. That’s fine. The goal isn’t to predict the future perfectly. The goal is to be working six months ahead instead of reacting at the last minute, and to have a consistent, repeatable process for knowing where to focus your time.
That’s not a health score. It’s judgment, informed by the right inputs, applied regularly.
