The Interruption Tax: What “Quick Questions” Really Cost Your Experts

There is one person on every team who cannot get through a morning. Every few minutes someone leans over, opens a chat window or catches them in the corridor with the same three words: “Got a sec?” They always do, because they are the person who knows. They know why that customer gets the special discount, how the old export process really works, which supplier you never chase on a Friday, and what the number in cell G14 actually means. So the questions come, all day, and their own work slides to the evening.

This is a tax. Not a metaphor for a tax - a real, recurring charge the business pays every time it uses its most knowledgeable people as human search engines. The visible part is tiny: the thirty seconds it takes to answer. The invisible part is enormous: the expert loses an average of 23 minutes to climb back into the complex task they were pulled out of³, the colleague who asked has been idle waiting, and the same question will be asked again next week by someone else, because the answer still lives in exactly one head. Microsoft found people are now interrupted every two minutes during core hours, roughly 275 times a day². Your experts are interrupted the most, because knowing the answer is what makes you the target.

This piece names that combined charge the interruption tax, models what it costs, and shows why the usual fixes - open-door culture, wikis, help desks, generic chatbots - never remove it. Then it shows what does: a Company Brain that answers the routine questions so no one has to ring the expert, and AI employees that take over the routine follow-through, so your experts are interrupted only for the genuinely hard calls that need their judgement. The outcome is leverage, not another tool in the stack.

What the Interruption Tax Is

The interruption tax is the full cost a company pays for keeping critical knowledge inside people’s heads and retrieving it by interrupting them. It has three layers, and most leaders only ever see the first.

• The visible answer - the few seconds or minutes it takes the expert to reply. This is the only layer anyone counts, because it is the only one that looks like the whole cost.
• The broken focus - the deep work destroyed on both sides of the interruption: the expert torn out of a complex task, and the colleague who sat idle waiting for the answer before they could continue.
• The systemic drag - the slow answers that queue behind a busy expert, the bottleneck that forms because only one person knows, and the fact that the same question gets asked again and again because the answer was never captured anywhere shared.

Call it a tax because it behaves like one. It is levied on every retrieval of knowledge, nobody votes to pay it, it is never itemised, and it rises automatically as the organisation grows. And like a badly designed tax, it falls heaviest on the people you can least afford to have idle: your most senior, most expensive, most in-demand experts, whose knowledge makes them the default destination for every question.

Why people interrupt experts at all

If you audit a typical day of interruptions to a senior person, very few are about genuinely novel problems. Most fall into four categories, and every one of them is a symptom of knowledge that is not held anywhere shared.

Notice what all four have in common: the interruption is a workaround for missing shared knowledge. That is the key to the whole problem. If the knowledge were held somewhere durable and queryable, most of these questions would never reach the expert. This is why the interruption tax is really a knowledge problem wearing a productivity problem’s clothing, and why the fix is not a stricter open-door policy but where the company’s answers live.

Experts as Human Search Engines

The defining feature of the interruption tax is who pays it. It is not spread evenly across the workforce. It concentrates on the small number of people who hold the most valuable knowledge, and it grows in direct proportion to how good they are.

• Competence attracts questions - the moment a person is known to have the answer, they become the default destination for it. The better they are, the more traffic they draw, so their reward for expertise is a heavier interruption load.
• Search is faster than self-service - for the asker, tapping the expert is the path of least resistance. It is quicker to interrupt a human who knows than to search a wiki that might be wrong, so the expert becomes the company’s search engine of choice.
• Knowledge concentrates as complexity rises - the messy exceptions, the customer-specific rules, the history behind a decision live in the heads of the few people who were there. Those are exactly the questions that cannot be answered any other way today.
• The expert cannot say no - saying “look it up yourself” feels obstructive and slows a colleague down, so most experts answer, again and again, out of goodwill and at their own cost.
• Every answer deepens the dependency - each time the expert answers instead of the knowledge being captured, the company gets more dependent on that one head, not less.

The go-to-person bottleneck

When knowledge concentrates in one person, that person stops being an expert and becomes a bottleneck. Work does not flow at the speed of the task; it flows at the speed of the one human who can unblock it.

The bottleneck is the compounding form of the interruption tax. Every answer that is given but not captured makes the next interruption more likely and the dependency deeper, until the company’s throughput is capped by the availability of a handful of overloaded people. That is a fragile way to run anything, and it gets more fragile as you grow.

The Recovery Cost Nobody Budgets

The reason a quick question is so expensive has nothing to do with the answer and everything to do with what interruption does to the human brain. The cost is not the reply; it is the recovery.

• Recovery is slow - it takes an average of 23 minutes and 15 seconds to fully return to a complex task after an interruption, according to University of California research led by Gloria Mark³.
• Switching itself is the cost - the American Psychological Association reports that toggling between tasks can consume as much as 40 percent of a person’s productive time, and that only about 2 percent of people multitask without measurable performance loss⁵.
• Attention is already shattered - Mark’s later research found the average time on a single screen before switching has collapsed to about 47 seconds⁴.
• Interruptions arrive faster than recovery - if a person is interrupted every two minutes² but needs 23 minutes to fully refocus³, they never reach full cognitive recovery before the next question lands.
• People compensate with speed and stress - Mark’s work is titled “more speed and stress” for a reason: interrupted workers try to make up the lost time by working faster, which raises frustration, pressure and error rates³.

This is why a 30-second question rarely costs 30 seconds. When a colleague interrupts an expert who is deep in a spreadsheet or a specification, the answer is instant but the re-entry is not. The expert has to rebuild the mental model they had loaded, remember where they were, and re-establish the chain of reasoning. Do that a dozen times a day and the expert never gets a single unbroken block of deep work. The tax is not the sum of the answers; it is the sum of the recoveries.

Fragmentation, not just volume

Two experts fielding the same number of questions can pay wildly different taxes depending on how the interruptions are spread. Fragmentation, not raw volume, is what destroys the ability to do deep work.

Research on focus finds that around 40 percent of knowledge workers never get even 30 consecutive minutes of focused time in a day¹⁰. Deep, expert work needs far longer than that, so a fragmented day does not merely reduce output proportionally - it can push complex work below the threshold where it happens at all. Past a certain density of interruptions, the expert’s deep work simply stops, and all that is left is the answering.

Why It Falls on Your Best People

A one-off cost you can absorb. The interruption tax is dangerous because it compounds and concentrates - it grows faster than the company that carries it and it lands on the few people you most need to protect, for structural reasons that have nothing to do with anyone being disorganised.

1. Question volume grows faster than headcount - add one person to a team and you do not add one relationship, you add a new source of questions for everyone who already holds knowledge. Interruption load rises with the number of connections, not the number of people.
2. Knowledge concentrates as you grow - more people, products and systems mean more exceptions and edge cases, and those settle into the heads of the few experienced people, so the expert becomes the answer to more and more questions.
3. Every answer that is not captured deepens dependence - answering in the moment feels efficient, but it means the knowledge stays in one head, so the next identical question comes straight back to the same person.
4. Turnover resets everything - when the person who held the context leaves, the questions do not stop; they scatter across everyone else who now has to guess, rebuild and re-learn, which multiplies interruptions for months.
5. Tools add places to ask, not answers - each new chat channel and app is one more surface for a “got a sec?” to arrive on, so buying software to help often raises the interruption count instead of lowering it.

The AI trap that makes it worse

The obvious response - give everyone a chatbot - can make the interruption tax worse, not better, when the tool does not actually know your company. A generic assistant that answers from the public internet cannot answer company-specific questions, so people fall back to interrupting the expert, now with less patience.

• Generic tools answer the wrong questions - a public chatbot is good at generic knowledge and useless at your pricing exceptions, your process history and your customer-specific rules, which is exactly what people interrupt experts to get.
• A confidently wrong answer is worse than none - when a generic model guesses at a company-specific question, it produces plausible, wrong answers, so people learn not to trust it and go back to the human.
• The gap between generic and grounded is large - the Federal Reserve Bank of St. Louis found generic generative AI saves the average user only about 2.2 hours a week¹¹, while production agents grounded in company context recover a median 6.4 hours, and 10 to 12 for seniors¹. The difference is whether the tool knows your company.

Why the Usual Fixes Fail

Every company has tried to protect its experts from the flood of questions. The common remedies help at the edges, but none removes the tax, because none touches the reason the interruptions happen: the answer lives in one head, and getting it out requires interrupting that head.

Culture and etiquette fixes

• Open-door culture - well meant, but it institutionalises interruption by making the expert permanently available, so the tax is maximised by design.
• Focus hours and do-not-disturb blocks - they defer the question rather than remove it, so the interruptions pile up and arrive in a burst the moment the block ends.
• “Check the wiki first” rules - they shift blame to the asker without giving them a source they trust, so people quietly go back to asking the human.
• Office-hours slots - they batch some questions, which helps, but the urgent ones still jump the queue and the standing dependency on one person is untouched.

Wikis, intranets and knowledge bases

• They store snapshots, not reasoning - a wiki captures what someone chose to write on one day; it does not hold the live exceptions and current decisions people actually ask about.
• They go stale - the moment a process changes, the page is wrong, people get burned once, and they stop trusting it, so they ask the expert to be safe.
• They are hard to search - finding the right page is often slower than asking a human, so the expert stays the faster option and keeps getting interrupted.

Help desks and generic chatbots

• Help desks reroute, they do not remove - a ticket still lands on an expert’s desk, just with a queue in front of it, so the interruption is organised rather than eliminated.
• Generic chatbots do not know your company - they answer from the public internet, not your rules and history, so they cannot handle the company-specific questions that drive expert interruptions.
• Confidently wrong is worse than silent - a plausible wrong answer erodes trust fast, and once burned, people default back to the human they know is right.

The pattern is consistent: the usual fixes attack the interruption, but the interruption is a symptom. The disease is knowledge trapped in heads and no trustworthy, current place to get the answer without a human. Treat the symptom and the tax returns; treat the cause and it falls away.

Modelling the Cost

The interruption tax feels abstract until you put a number on it. The maths is simple and the result is uncomfortable, which is exactly why so few companies do it.

A single interrupted expert

Start with one senior expert on a fully-loaded cost of 90 euros an hour who fields a dozen routine questions a day. The visible cost is trivial; the real cost is not.

One interrupted expert, roughly 90,000 euros a year once you count all three layers - and more than twenty times the number a naive calculation of answer-time alone would show. The answering itself is a rounding error. The real money is in the destroyed focus and the idle waiting, neither of which ever appears on a budget line, which is precisely why the tax goes unmanaged.

Scaling to the company

Now generalise across a workforce. The point is not a precise figure but the order of magnitude, which is always larger than leaders expect.

Answering Without the Expert

If most interruptions exist to pull an answer out of one person’s head, then removing them requires two things: a place for the answers to live that is not a human head, and something that can act on those answers without booking the expert. That is the Company Brain and the AI employee.

The Company Brain holds the answers

• One durable store of knowledge - your decisions, rules, definitions, exceptions and the reasoning behind them live in a structured, queryable brain instead of scattered across heads, chats and stale wiki pages.
• It survives turnover - when someone leaves, the answers stay, so the flurry of interruptions that normally follows a departure never starts and the knowledge does not walk out the door.
• It answers instead of interrupting - a colleague, a new hire or an AI employee retrieves the current, correct answer directly, so the expert is never tapped for a routine look-up.
• It stays current - the brain learns from how work actually happens and from expert feedback, rather than depending on someone remembering to update a page.
• It knows what it does not know - when a question is genuinely novel, it routes to the right human instead of guessing, so the expert only ever sees the hard calls.

AI employees take over the follow-through

• They answer the routine questions - grounded in the Company Brain, an AI employee handles the recurring look-ups and how-do-I questions that used to land on the expert.
• They gather what a question needs - pulling the current status, the relevant record and the history from email, Teams, SharePoint, CRM and ERP, so the answer is complete, not a pointer to go and find it.
• They run the next steps - the action a question implies, the update, the confirmation, gets handled directly rather than handed back to a person.
• They escalate only real exceptions - when judgement is genuinely required, the AI employee brings the expert a framed decision, not a cold question, so the interruption is worth the switch.
• They act, not just advise - unlike a chatbot, an AI employee owns an outcome end to end across your real systems, with a human in the loop for genuine exceptions.

The goal is not zero questions. It is to delete the questions that only exist to move information out of one head, so the questions that need human judgement reach the expert with the context already gathered. That is leverage: the same people, freed from being human search engines, spending their hours on the work only they can do.

The Protect-the-Expert Playbook

You do not remove the interruption tax with a policy memo asking people to stop interrupting. You remove it one recurring question at a time, by making sure the answer lives somewhere other than the expert’s head. Here is a practical sequence.

1. Find your most-interrupted people - identify the two or three experts everyone routes questions to. They are the highest-value place to start and usually know exactly which questions they answer over and over.
2. Log the recurring questions - for two weeks, have those experts jot every question they get. Patterns emerge fast: a small set of look-ups and how-do-I questions makes up the bulk of the load.
3. Price the interruption - run the three-layer cost model on one overloaded expert. Nothing changes a leadership team’s mind faster than seeing one person’s interruptions cost six figures a year.
4. Capture the answers into a Company Brain - for each recurring question, put the current answer, its exceptions and the reasoning into a shared brain, drawn from the expert while they are still here.
5. Point an AI employee at the routine load - let it field the recurring questions and run the follow-through across your systems, so the answer and the action both happen without the expert.
6. Route only real exceptions to the human - configure the escalation so the expert sees a question only when it is genuinely novel, framed with the context already gathered.
7. Protect the reclaimed focus time - give experts real, defended deep-work blocks now that the routine questions are handled elsewhere, and measure whether the blocks survive.
8. Measure and expand - track interruptions removed, expert focus hours recovered and answer speed for the rest of the team. Then move to the next expert and the next department.

How Superkind Fits

Superkind builds AI employees grounded in a Company Brain, designed to learn your company rather than the internet. That combination is exactly what the interruption tax needs: a durable home for your answers and something that can run the routine follow-through without tapping the expert.

• Company Brain that survives turnover - your rules, decisions and exceptions are captured once and kept current, so the knowledge people interrupt experts to get lives somewhere it can be queried instead.
• AI employees that act, not just chat - they answer the routine question and run the next step end to end across your real systems, rather than handing a person more to do.
• Grounded, not generic - answers come from how your company actually works, so they are company-specific and correct, not plausible guesses from the public internet.
• Works across email, Teams, SharePoint, CRM and ERP - the answer to a question is assembled from the systems where the facts actually live, not from a person you have to interrupt.
• Answers first, escalates second - routine look-ups are handled without a human; only genuinely novel questions reach the expert, framed with the context already gathered.
• Knows what it does not know - when a question falls outside the brain, it routes to the right person instead of guessing, so trust stays intact.
• Live in weeks, not quarters - the first use case goes into production quickly, on top of your existing stack, with your experts giving feedback from day one.
• Human in the loop - genuine exceptions escalate to a person; the AI employee handles the routine and knows when to ask.

Decision Framework: How Heavy Is Your Interruption Tax?

Not every company needs to attack this today. Use these signals to judge how heavy your tax is and what to do about it.

Gartner expects more than 40 percent of agentic AI projects to be cancelled by the end of 2027, mostly from unclear value and hype-driven scope⁹. The way to avoid that fate here is to start from a specific, overloaded expert and a measurable outcome - interruptions removed and focus recovered - rather than from the technology.

Frequently Asked Questions

Henri Jung

Co-founder of Superkind, where he helps SMEs and enterprises deploy custom AI agents that actually fit how their teams work. Henri is passionate about closing the gap between what AI can do and the value it creates in real companies. He believes the Mittelstand has everything it needs to lead in AI - it just needs the right approach.