User engagement score

A user engagement score is a single composite number that combines the actions a user takes into one weighted measure of how deeply they use a product. Rather than tracking a dozen separate usage signals, the score rolls them into one figure. A user who logs in daily, uses three core features, and invites teammates scores higher than one who logs in once a month and touches a single feature.

The score exists to make engagement comparable and actionable. Raw event counts are hard to reason about, because frequency, depth, and breadth all move independently and no single one captures the full picture. A composite score collapses them into a number you can rank users by, segment on, and watch trend over time. It answers a practical question: which users are getting real value, and which are drifting toward the exit.

The score is only as good as the weights behind it. Weights should reflect how strongly each action predicts a user staying, expanding, or eventually churning. An action that correlates with long-term retention earns a high weight. A vanity action that anyone does once and never again earns a low one. Done well, the score becomes a leading indicator that moves before retention rate or churn rate catch up.

The score is a weighted sum of the actions a user takes in a period. Each tracked action has a weight, the action counts are multiplied by their weights, and the products are added together. Many teams then normalise the result to a fixed scale, such as 0 to 100, so scores are easy to read and compare. The arithmetic is straightforward. The judgement is in choosing which actions to include and how to weight them.

1. 1

   Tracked actions

   The set of user behaviours that feed the score, such as logins, core actions completed, features used, and collaborators invited. Include actions that signal value and leave out noise that does not separate engaged users from drifting ones.

2. 2

   Action weights

   A value for each action reflecting how strongly it predicts a user staying or expanding. Derive weights from data where you can, by checking which actions correlate with retention, rather than setting them by gut feel alone.

3. 3

   Measurement window

   The period over which actions are counted, such as the last 7 or 30 days. A rolling window keeps the score current and lets it fall when a user goes quiet, which is exactly the signal you want for an at-risk alert.

4. 4

   Normalisation

   An optional final step that rescales the weighted sum to a fixed range so scores are comparable across users and over time. Normalising to 0 to 100 makes thresholds and segment cut-offs easy to set and explain.

A worked example shows the weighting at work. Suppose logins carry a weight of 1, completing a core action carries 5, and inviting a collaborator carries 10. A user with 8 logins, 4 core actions, and 1 invite scores 8 plus 20 plus 10, which is 38 before normalising. A user with 12 logins but no core actions and no invites scores just 12, even though they logged in more often. The weights ensure the score rewards the behaviours that actually predict retained value rather than raw activity.

An engagement score is a rollup, which makes it easy to watch and hard to act on. When the score falls, the number alone does not say which behaviour cooled off. A metric tree reverses the rollup, decomposing the score back into the components it was built from and attaching each one to the team that owns it.

The first level splits the score into frequency, depth, and breadth. Frequency is how often users return, the heartbeat of engagement. Depth is how much they do per visit, the sign they are completing real work rather than glancing. Breadth is how many distinct features or workflows they touch, the indicator of how embedded the product is. A score can dip because any one of these slipped, and each slip has a different cause and a different fix.

KPI Tree connects each branch to the action and team that moves it, with RACI ownership so the frequency branch sits with lifecycle, the depth branch with the team that owns the core workflow, and the breadth branch with the team driving cross-feature adoption. When the score drops, the change is pushed to the owner of the component that moved, not broadcast as a number nobody owns. The verified impact loop then checks whether their intervention actually lifted that component on the next reading. That is how the gap between a falling score on a dashboard and a team that knows what to do about it gets closed.

There is no universal benchmark for an engagement score, because the score is defined by weights and actions you choose, so two products almost never compute it the same way. The useful benchmarks are internal: the distribution of scores across your own base and how a score predicts retention. The ranges below describe a score normalised to 0 to 100 and the retention behaviour each band tends to show.

The benchmark that proves the score works is its correlation with what happens next. If users in the lowest band churn far more often than users in the highest, the score is doing its job as a leading indicator. If the bands do not separate retention, the weights are wrong and the score needs recalibrating before anyone trusts it for an alert.

Improving the score means lifting the components beneath it, not gaming the number. Because the score is a weighted sum, the highest-leverage move is usually to raise the component the tree shows is weakest for a given segment, rather than pushing the action that is already strong. Diagnose which of frequency, depth, or breadth is dragging, then act on that one.

1. 1

   Weighting by what is easy to count

   Logins and page views are simple to track but weak predictors of value. If the heaviest weights sit on the easiest signals, the score rewards activity over outcomes and stops predicting retention. Tie weights to behaviours that correlate with staying.

2. 2

   Never recalibrating the weights

   A score calibrated at launch drifts as the product changes and new features arrive. Weights that fit last year can be stale today. Revisit them on a schedule and after major releases so the score keeps tracking real value.

3. 3

   Treating the composite as the whole story

   A single score hides which component carries it, so two very different accounts look identical. Always keep the score decomposable into frequency, depth, and breadth rather than reporting the rollup alone.

4. 4

   Comparing scores across different definitions

   Because every team weights its own score, comparing your number to another product or to an old version of your own formula is meaningless. Compare scores only within one consistent definition, and re-baseline whenever the formula changes.