This is a genuinely different axis from RFM or purchase propensity, both of which are anchored to purchase behavior. CES scores activity that frequently precedes or predicts a purchase decision rather than measuring the purchase itself — a customer can carry a high CES with no recent order (an engaged evaluator or a subscription user mid-cycle) or a low CES immediately after a purchase (someone who bought once and hasn't returned to the product since).
Building it starts with choosing which events actually signal engagement for your specific product — a login means little for a transactional e-commerce brand but is central for a SaaS product; a support contact can signal either genuine engagement or frustration depending on context, and needs to be weighted or excluded accordingly rather than assumed positive.
The score's real value is as a leading indicator rather than a status report. A customer whose CES is declining over consecutive periods is showing disengagement before it shows up in churn or renewal data — which is the window where a retention intervention still has a chance to work, rather than after the cancellation has already happened.
Formula
CES = Σ(Weight₁ × n₁ + Weight₂ × n₂ + ... + Weightₙ × nₙ), normalized to a 0-100 scale
Identify the discrete events that genuinely signal engagement for your product, assign each a weight reflecting its actual signal strength (not just what's easiest to track), and sum weighted occurrence counts per period. Normalize against your own customer base's score distribution rather than an external benchmark — a score of 71 means nothing without knowing what your own engaged customers actually score.
Why Customer Engagement Score matters
CES catches disengagement while it's still a leading indicator rather than a lagging one — a customer's engagement typically declines for weeks or months before a cancellation or non-renewal, and that decline is invisible to purchase-based metrics like RFM until the customer has effectively already left.
Two customers RFM would score identically
Two subscription customers both show identical recency, frequency, and monetary figures — same plan, same tenure, same last-payment date. RFM alone would treat them as equivalent. Their engagement scores tell a different story: Customer A logs in weekly, uses three core features, and has opened recent product emails — a CES in the engaged range. Customer B hasn't logged in for six weeks and stopped opening emails a month ago — a CES trending toward dormant, despite an identical, still-current subscription. Only the CES flags Customer B as a real churn risk before the next renewal date arrives.
Common mistakes
Weighting every event equally
A login and a completed core-workflow action are not equally informative — treating them the same produces a score that's easy to calculate and doesn't actually separate engaged customers from passively-still-subscribed ones.
Assuming support contact is always positive engagement
A support ticket can signal genuine product engagement or genuine frustration, and folding it into the score as a flat positive weight without context muddies the signal rather than sharpening it.
Benchmarking against an external CES range instead of your own base
A '71+ is highly engaged' rule of thumb from another company's product tells you little about your own — normalize against your own customer base's actual score distribution, not a borrowed threshold.
Where we work on this