A common engagement problem

Sessions get shorter every week

The short answer: vary the reward. Here is how to recognise this problem in your funnel, why it happens, and how the best consumer apps design it away.

How to recognise it

  • Session duration trending down across cohorts
  • The core loop identical on every visit
  • Engagement propped up by notifications alone

Why does engagement decay over time?

A loop that pays out identically every time gets fully learned, and fully learned means boring. The hundredth session delivers exactly the hundredth-session experience: no possibility of pleasant surprise, nothing to anticipate.

Habituation is biology, not betrayal. The product did not get worse; the user's brain simply stopped paying attention to a pattern it has fully priced in.

The fix is not more content. It is uncertainty, applied carefully: anticipation of a possible reward engages harder than certainty of a known one.

How do you fight loop fatigue? Vary the reward at the edges

Keep the core reliable, and vary the edges: an occasional bonus, a surprise acknowledgement, a result presented with an unexpected flourish, a rare find mixed into the routine.

The placement rule is absolute: variability belongs where outcomes are low-stakes and repeated. The moment you randomise something users depend on, delight becomes distrust.

Small beats big: a tiny unexpected pleasure at unpredictable intervals outperforms a large scheduled one, because the unpredictability is doing the work.

And watch the ethics line: variable reward is the slot-machine mechanism. Aimed at genuine value it keeps a product fresh; aimed at compulsion it burns users, and eventually brands.

The building blocks that solve it

These come from the Engagement stack of the Product Design Playbook. Each building block is five cards: the tactic and the psychology behind it, a Make It Yours prompt card, and three real app examples.

  • Variable Reward: Predictable core, surprising edges.
  • Small Quirk: An occasional flourish keeps the routine from going grey.

Who does this well

Duolingo: Chest rewards and bonuses vary constantly. The lesson mechanics never do.

TikTok: The entire product is a variable reward schedule: the next swipe might be the best one, so the next swipe happens.

Questions founders ask

Isn't variable reward manipulative?

The mechanism is neutral; the aim decides. Varying how genuine value arrives keeps a product interesting. Engineering compulsion toward empty value is exploitation, and users increasingly name it.

The test: would the user, shown exactly how it works, feel served or farmed?

What can we safely randomise?

The edges: bonuses, celebrations, discoveries, small delights.

Never the core: results users depend on, saved data, prices, anything mission-critical. Randomness there is not delight, it is unreliability.