We know that intermittent rewards keep people pulling a lever, but what happens when the lever is a spreadsheet, a codebase, or a set of sales calls? A recent workplace experiment found that inserting a short, unpredictable reward every ninety seconds raised persistence on a dull task by 34% by late afternoon — a bigger effect than most managers would predict from something so small. The question worth asking is not whether this works, but why the timing and the randomness matter more than the size of the reward.

The mechanics of unpredictable reward

Behavioural psychology has understood since the 1950s that the pattern of reinforcement shapes behaviour more powerfully than its magnitude. B.F. Skinner's work on schedules of reinforcement identified the variable-ratio schedule — rewards delivered after an unpredictable number of responses — as producing the highest and most persistent rates of behaviour. The classic explanation is that you cannot predict which response will pay off, so you keep responding. What is often missed is the corollary: the interval schedule, where a reward arrives after an unpredictable amount of time rather than an unpredictable number of actions, produces steadier, less frantic persistence. That distinction matters enormously for knowledge work.

A 90-second randomised interval sits in an interesting middle ground. It is short enough that the next reward always feels imminent, but irregular enough that you cannot simply wait for it. The unpredictability does two things at once. First, it prevents the habituation that kills fixed rewards — a bonus paid every Friday by 10am quickly becomes part of the furniture and stops motivating anyone. Second, it keeps attention anchored to the task rather than to the reward, because there is no reliable moment at which to disengage and check.

There is a neurological story underneath this. Dopamine neurons fire most strongly to unexpected reward, not to reward itself. Once a reward is fully predicted, the signal shifts earlier in time and flattens. Wolfram Schultz's work at Cambridge in the 1990s established this prediction-error function clearly: the brain is tracking surprise, not pleasure. A randomised 90-second structure manufactures a steady supply of small, genuine surprises during work that is otherwise predictable and flat.

Why 90 seconds, and why the afternoon

The specific interval is not arbitrary. Research on attention and task-switching suggests that most people can hold focus on a moderately demanding task for stretches of roughly one to three minutes before a micro-drift occurs — a glance at a phone, a mental wander, a sudden urge to check email. These micro-drifts are where persistence quietly dies. A reward window shorter than the drift interval catches the person before they leave.

The afternoon effect is where the 34% figure becomes interesting. Cognitive fatigue is not linear. Studies of circadian rhythm and self-control — notably work building on Roy Baumeister's ego-depletion framework, and later refinements by researchers who questioned its size — consistently find that inhibitory control is strongest in the morning and weakest in the mid-to-late afternoon for most chronotypes. By 3pm, the cost of continuing a boring task feels disproportionately high. This is where loss aversion becomes relevant: the anticipated discomfort of another hour of dull work looms larger than the modest benefit of finishing it. Kahneman and Tversky's central insight — that losses weigh roughly twice as heavily as equivalent gains — applies to effort as much as to money. Late in the day, the perceived "loss" of rest and relief dominates.

A randomised reward changes the frame. It does not remove the fatigue, but it inserts a series of small, unpredictable gains into the near future, which competes directly with the anticipated loss. The effect compounds across the afternoon because each reward resets the immediate calculation: the next 90 seconds might contain something good.

A concrete example

Consider a 2021 field study on data-entry teams in a shared-services environment, where researchers replaced a fixed end-of-week recognition scheme with randomised micro-rewards delivered at unpredictable intervals during shifts. Output on routine entry tasks rose measurably in the final two hours of the working day — the window where output had previously collapsed — while morning output was essentially unchanged. The rewards were trivial in monetary terms: a voucher, a public nod, a coffee. The gain came from timing and unpredictability, not value. This mirrors what the 90-second finding suggests: you are not buying effort, you are interrupting the moment at which effort is abandoned.

The risk of getting this wrong

There is a real hazard here, and it is worth being blunt about it. Variable-ratio reinforcement is the same mechanism that makes certain digital products compulsive. Introducing randomised rewards into a workplace without care can create the very attention problems you are trying to solve — people refreshing, checking, anticipating, and losing the thread of actual work. The difference between a helpful structure and a corrosive one comes down to three things.

Task-relevance. The reward should be tied to continuing the task, not to checking a separate system. If the reward arrives through a notification, you have built a distraction engine. If it arrives as a brief, pleasant interruption within the work — a short break prompt, a quiet acknowledgement, a change of music — you have built a persistence aid.

Opacity of the schedule. The moment people can predict the reward, the effect decays. This means the randomness must be genuine, and the interval must not be discoverable through observation. It also means you cannot explain the schedule in detail; you can only explain the principle.

Non-exploitation. There is an ethical line between supporting sustained effort and engineering compulsion. Any structure that deliberately obscures how long a person must work, or that uses reward to extend hours beyond what is healthy, is not a productivity tool — it is a dark pattern with a payroll. The 34% figure is a measure of persistence on a defined task, not a licence to extract more.

What this means for how we design work

The broader lesson extends well beyond any single experiment. Most workplace motivation systems are built on the assumption that reward size is the lever that matters: bigger bonuses, larger recognition, more generous packages. The evidence points elsewhere. What drives continued effort on unrewarding tasks is the structure of reward over time — its unpredictability, its frequency relative to the drift interval, and its placement in the parts of the day when self-control is cheapest to lose.

Forward-looking teams are already experimenting with this in small, cautious ways: randomised short breaks, unpredictable micro-recognition during long asynchronous work blocks, and scheduling the least engaging tasks into the windows where a small, irregular uplift does the most good. The lesson from the 90-second finding is not that we should gamify everything. It is that the timing and unpredictability of small rewards can do work that large, predictable ones cannot — and that the afternoon, specifically, is where that difference is largest and most worth designing for.