Product & engagement analytics · Python
Engagement is a behaviour, not a demographic.
I analysed 700 mobile users sorted into five engagement tiers to answer one product question: what separates a light user from a power user? The answer is clear — who a user is (age, gender, device, OS) predicts engagement almost not at all. What they do — above all, how many apps they adopt — predicts it almost perfectly.
700 users · 5 tiers
Python · pandas
Correlation & segmentation
The numbers at a glance
271min
avg. app use per day
0.98
apps-installed ↔ engagement (top driver)
0.00
age ↔ engagement (no effect)
What actually drives engagement
Correlation of each factor with a user's engagement tier
How strongly each factor moves with engagement (1 = perfect, 0 = none). Behaviour dominates; demographics don't register.
Number of apps installed is the single strongest signal of engagement (0.98) — closely followed by battery, app-time, screen-time and data, which all move together. Age barely registers at all (−0.00): how old someone is tells you nothing about how engaged they are.
The engagement ladder
Apps installed across the five tiers
Every behavioural metric climbs steadily from tier 1 (light) to tier 5 (power). Switch the metric to see the same staircase each time.
Metric
Where it doesn't move — demographics
Average engagement tier by segment
The whole population averages tier 3.0. Cut it any demographic way and every segment lands within a whisker of 3.0 — the bars barely move.
Cut by
Flat. No demographic segment separates from the 3.0 average by more than a fraction of a tier. Targeting users by who they are won't grow engagement — you have to change what they do.
Power users vs light users
Power users tier 5 · 136 users
Apps installed89
App time / day541 min
Data / day1,975 MB
Light users tiers 1–2 · 282 users
Apps installed23
App time / day97 min
Data / day331 MB
What the data says to do
Grow the app portfolio
Apps installed is the #1 correlate of engagement (0.98). Power users hold 89 apps vs light users' 23 — a 3.9× gap.
Make app discovery a core onboarding goal — breadth of adoption is the engine.
Stop targeting demographics
Age, gender, OS and device model all sit at ~3.0 engagement with near-zero correlation. Demographic segments don't behave differently.
Segment by behaviour (tier), not by who users are — it's where the signal is.
Move the middle up
Tiers 2–3 are 41% of users and sit just below the power band. They're the cheapest tier to lift — already habitual, not yet maxed.
Nudge mid-tier users toward one or two more high-value apps to climb a tier.
Watch the whole bundle
Screen time, battery and data all rise together with engagement (0.95–0.98). Engagement shows up across every signal, so any one can proxy it.
Track a simple composite; you don't need all five metrics to spot a rising user.
The retention lever
Adoption breadth, not demographics, is the growth dial
The clearest path to higher engagement — and the retention that rides on it — is getting users to adopt more relevant apps and build daily habit. Because demographics don't separate users, spend targeting budget on behavioural triggers (onboarding, app discovery, habit loops) rather than audience segments. The 41% of users sitting in tiers 2–3 are the highest-return group to move.