In June 2025, Microsoft published a special report using anonymous Microsoft 365 signals from employees around the world. The number that stood out: the typical employee is interrupted every 2 minutes — about 275 times a day — by meetings, emails and instant messages.
But striking as it is, that number hides a deeper problem. Most of those interruptions are not real emergencies. It's a colleague asking for the third time where the vacation policy lives, a manager checking on the same order that was already answered last week, a customer wanting to know again who handles their account and what their complaint history looks like. Questions someone at the company already answered. Answers that already exist somewhere — in an email, in a shared document, in the memory of a colleague who knows. Interruptions that, with a different way of organizing work, simply would not happen.
The cost isn't only the time lost in that moment. According to researcher Gloria Mark, of the University of California, Irvine, after an interruption a person takes on average 25 minutes to fully refocus on the original task. Since the interruptions in Microsoft's report happen every 2 minutes, much of the workday never gets close to full concentration — it's one fragment after another, with no time to return to what mattered before the next interruption.
What gets lost to 275 interruptions a day
According to the McKinsey Global Institute, the typical knowledge worker spends 28% of the workweek managing email — a little over 11 hours — and another 19% searching for internal information or trying to track down a colleague who can help with an answer. Together, those two activities eat up nearly half the week before any actually valuable work gets done.
Add the 275 daily interruptions to that and what emerges is a picture of productive chaos. In the same Microsoft report, nearly half of employees (48%) — and more than half of leaders (52%) — said their work feels chaotic and fragmented. They sense, intuitively, that they can't catch their breath between one thing and the next.
The phenomenon is global and accelerating. Microsoft's data comes from more than 31,000 professionals across 31 countries. Teams messages are growing 6% a year, with spikes of 20% in regions like Central and Eastern Europe. Meetings after 8 pm are up 16% year over year. 30% of all meetings now span multiple time zones — meaning people meet outside their normal working hours just to sync up with a team on another continent.
It's the infinite workday: work that never ends because it never manages to start. Every form of "urgent communication" is designed around what works for the sender, not what works for the receiver. And when everything is urgent, nothing is.
At many Brazilian and Latin American companies, interruption has picked up one more channel: the messaging group on the phone. The question about a supplier's deadline arrives mixed in with the meeting reminder and a photo from a client, with no organized history and no way to search by topic. The following week, the same question comes back — and whoever knows the answer types it all over again.
Why banning meetings and messages doesn't work
The first reaction is always to restrict: block meetings past a certain hour, ban after-hours email, impose "offline time" and turn off notifications. Those policies sound good in a memo. In reality they rarely hold: when there's no alternative for getting the work done on time, people find a way around them.
When a company doesn't offer an official place to answer a frequent or routine question — when there's no system that captures the answer once and reuses it — the person has to choose between delivering late or breaking the policy. Delivering is what pays the bills. Most people choose to deliver. A policy with no alternative doesn't reduce the behavior; it only reduces its visibility.
A common example: at month-end close, finance gets the same question from several managers — has the supplier's invoice been posted yet? Every answer means stopping, opening the system, searching and typing a reply. Banning the question doesn't fix it. What fixes it is having the answer available to whoever asks, with the right permission, without depending on someone interrupting their own work.
The real cost shows up when nobody can measure what's happening. The daily interruptions turn into fragments of shadow productivity: private conversations, parallel searches in Slack or email that the company can't optimize because it can't see them. And the pattern repeats every month.
What has to be in place
To cut down on interruptions, you don't need to ban communication. You need to eliminate the communication that shouldn't exist in the first place — the questions whose answer is already sitting somewhere.
Approved knowledge with visible sources. Documents with an owner, a current version, a review date. When AI answers a question about an internal policy, it cites where it found the information. Whoever reads it knows whether it's the valid memo or something outdated. That's where the question dies.
Agents on repetitive routines. Instead of a colleague dropping what they're doing to answer what the status of order number X is, an agent checks the system, returns the answer, updates the history. The colleague is freed up. The routine is documented for the next identical question.
Access role by role. Not every agent needs to reach every tool. If finance only needs two functions inside an integration, those two get released — not the whole tool because "that was the only way."
Approval matching the risk. An agent answering an FAQ doesn't need review. An agent changing critical data asks for confirmation before acting, inside the conversation itself. Critical documents can require two rounds of review before becoming official knowledge, with whoever writes them kept separate from whoever approves them.
Audit trail. Creating an agent, approving a document, running a sensitive action: all of it logged. In an audit or an incident, that's the difference between reconstructing what happened in minutes and not being able to at all.
Reuse with permissions. When an agent works well, it can be made available to other teams — with reach defined by role, permission and area. Someone on the team builds it; the whole company moves forward.
That is how Skyller was designed: approved knowledge with sources, agents for the routines, approval and an audit trail as the default.
The payoff nobody expects
When AI use happens on a personal account or through side channels, whatever works well stays locked to whoever discovered it. The analyst who spent weeks building the perfect script for answering a tax dispute is the only person at the company who has it. When she leaves or moves to another area, the routine leaves with her. The company paid for the learning curve and kept nothing from it.
In a governed environment, the path is different. Agents, scripts and flows can be made available to other people and groups, with roles and permissions defined. Work that worked once doesn't die with the person who built it. On top of that, AI credits shared across the team solve the problem of underused licenses on one side and people hitting their limit on the other — usage becomes visible and the budget can actually be balanced.
But the bigger payoff is this: when an employee stops answering "what is policy X?", they get to focus on answering "why should policy X change?" When AI handles the routine questions — and an agent handles the recurring task — the person is freed up for the decisions that actually need them. That's the difference between a team that responds and a team that governs.
Three questions to bring to your next meeting
Before writing another policy about "focus time," it's worth answering these three with IT and operations leadership:
- How many times a week is the same question answered manually in your area? If the answer is "we don't know," the waste is invisible. If it's "dozens of times," there's potential for agents.
- When an agent answers something about an internal rule, where did that content come from and who approved it to be used that way? With no answer, any outdated file can become the official truth.
- What did the company learn this month about who's asking what? If the answer is "nothing," the pattern of interruptions stays invisible — and invisible doesn't get optimized.
