The real cost of context-switching

A number we didn’t believe at first

The first time we plotted switch frequency for a 40-person support organisation in Pune, we assumed the pipeline was broken. The chart said agents were changing application context every 74 seconds during peak hours. We re-ran the aggregation, checked the agent logs, and pulled a second team as a control. The number held.

Support is an extreme case, but knowledge workers in general were not far behind. Across teams on Momentum, the median gap between meaningful app switches sits at one minute and 52 seconds. Engineers do a little better. Marketers, oddly, do a little worse, mostly because their work lives across six browser tabs at once.

None of this shows up in a calendar. The day looks like four meetings and five hours of open time. The open time is where the tax gets collected.

What a switch actually costs

The academic literature puts full recovery from a hard interruption at over twenty minutes. Our data suggests most workplace switches are softer than that, but they are not free. When we compare output signals before and after a switch, typing cadence and task progression take roughly two to four minutes to return to their prior level.

Two minutes sounds trivial. Multiply it by thirty avoidable switches a day and you get an hour of degraded work per person, every day. For a 100-person organisation, that is roughly twelve full-time salaries spent on re-finding your place.

The cost is also uneven. Deep, stateful work like debugging or financial modelling pays the highest re-entry tax. Shallow work barely notices. Which means switch-heavy cultures quietly punish exactly the people doing the hardest thinking.

Self-reports make this worse, not better. When we survey teams alongside the telemetry, people estimate they lose maybe twenty minutes a day to interruptions. The measured figure runs three times higher. Nobody is lying; the brain simply does not keep receipts for attention. That gap between felt cost and real cost is why the problem persists for years untreated.

The chat tool is not the villain

It is tempting to blame Slack or Teams, and plenty of think-pieces do. Our classification data tells a more annoying story. Chat accounts for about a third of switches. Another third is self-inflicted: email checks nobody asked for, a quick look at a dashboard, a browser tab opened out of habit.

The final third is structural. Tickets that require three systems to resolve. Approval flows that bounce between tools. One logistics customer found that closing a single purchase order touched eleven applications. No wellness webinar fixes that.

So before you declare a no-notifications policy, look at where the switches actually originate. In our experience the split surprises almost every leadership team that sees it.

The Pune experiment

Back to that support org. Their leaders did not send a memo about focus. They changed two concrete things: tickets were batched by system instead of arriving in strict time order, and agents got a protected 90-minute block each morning for complex cases, with the queue covered on rotation.

Within three weeks, median switch interval moved from 74 seconds to just over three minutes during protected blocks. Average handle time on complex tickets fell 18 per cent. Customer satisfaction did not move at all, which was the point — nothing was sacrificed to get the gain.

The agents noticed before the dashboard did. One told her manager the mornings finally felt like ‘doing the job instead of dodging it’.

Worth stating plainly: this cost the company nothing. No new tooling, no headcount, no consultant. The only input was seeing the switching data at team level and being willing to rearrange work around what it showed. Most of the wins we observe look like this, embarrassingly cheap once the pattern is visible.

Measuring without creeping people out

A reasonable objection: measuring switches sounds like surveillance. We design against that deliberately. Momentum reports switching behaviour as team-level aggregates by default, and every individual can see exactly what the agent recorded about their own day before anyone else does.

This matters for data quality, not just ethics. People who trust the measurement do not game it. Teams that rolled out with full transparency showed stable behaviour from week one. Teams that tried quiet deployments, against our advice, showed a suspicious dip in switching that recovered once people stopped performing.

Honest numbers are the only numbers worth acting on.

What doesn’t work

We have watched dozens of interventions fail. Blanket ‘focus Friday’ policies collapse within a month because the work that caused the switching still exists; it just piles up until Monday. Notification bans fail for the same reason, plus they punish people whose role genuinely requires responsiveness.

Individual productivity scores tied to switching also backfire. People learn to leave one window open and look busy. The behaviour you wanted to reduce goes underground instead of away.

The pattern behind every failure is the same: treating switching as a personal discipline problem rather than a workflow design problem.

Where to start on Monday

Pull one week of aggregate switching data for a single team and sort by origin: chat, self-directed, structural. Pick the largest structural source and fix the workflow behind it. That is usually worth more than any behavioural campaign.

Then protect one recurring block, 90 minutes, for the work with the highest re-entry cost. Cover the interrupts on rotation so responsiveness does not drop. Measure the same numbers three weeks later.

Across the organisations we work with, that modest sequence recovers between 45 minutes and an hour of effective work per person per day. The tax never goes to zero. But you stop paying it on things that were never worth buying.

Writes for The Signal about focus and the future of measurable, humane work — drawing on anonymised patterns from the teams and focus hours analysed on Momentum.