Deep Work: How to Build a Focus Routine

Why does deep work need a routine instead of discipline?

Because the thing that breaks focus is not a lack of willpower. It is the cognitive cost of switching, and switching is triggered by environment far more than by character.

The clearest evidence comes from Sophie Leroy’s research on attention residue. In her 2009 experiments, people who switched from an unfinished task to a new one kept part of their attention on the first task, and their performance on the second suffered. The effect was strongest when the first task was left incomplete and when time pressure was high — which describes almost every knowledge worker’s Tuesday. You do not leave the previous task behind. You carry a fraction of it into the next one.

Gloria Mark’s field research at UC Irvine points the same direction from a different angle. Her workplace studies found that returning to an interrupted task takes several minutes of reorientation; the widely cited figure of about 23 minutes is her estimate for resuming interrupted work. Her later work, summarised in *Attention Span* (2023), measured average time on a single screen at around 47 seconds, down from about two and a half minutes when her studies began in 2004. That number describes screens, not thinking, but it describes the environment your routine has to survive.

Put those findings together and the design implication is obvious. You cannot out-discipline an environment that interrupts you every few minutes. You can only change the environment for a fixed window and make entry into that window cheap.

How much deep work can you actually do in a day?

For most people, three to four hours of genuinely demanding cognitive work is a realistic ceiling. Experts in focused domains often top out at four to five hours of deliberate practice before performance degrades — a limit documented in Anders Ericsson’s research on expert performance and popularised, sometimes too loosely, as the basis of the “10,000 hour” idea.

The practical version: do not plan eight hours of deep work and then conclude you are broken when you manage two. Plan for the ceiling and treat anything above it as a bonus.

Worked example: what interruptions actually cost

Suppose you block four hours on Wednesday. You receive six interruptions during those hours — two chat threads, two “quick questions”, one call, one notification you answer yourself. Use Mark’s 23-minute reorientation estimate as a rough cost model:

• 6 interruptions × 23 minutes = 138 minutes of recovery time
• 4 hours blocked = 240 minutes
• Remaining effective focus = 240 − 138 = 102 minutes, or about 1.7 hours

Even if you halve the recovery estimate to 12 minutes, you are left with 168 minutes. The point is not the precision of the constant. The point is that six small interruptions consume somewhere between 30% and 60% of a four-hour block, and none of them appear on your calendar as “cost”.

What are the four parts of a focus routine?

A working routine has four components, and each one fails in a predictable way when it is missing. Build them in order.

1. A start cue that takes under two minutes

The cue is a physical or digital action that always precedes the same kind of work: the same desk setup, the same playlist, the same document opened from the same bookmark, the same line written at the top of the page (“draft, section 3, 90 minutes”). It does two jobs: it removes the decision about whether to start, and it removes the five minutes of setup that usually become fifteen minutes of browsing. If your start ritual involves tidying the desk, closing eleven tabs, and finding the right notebook, it is not a cue, it is procrastination with a procedure.

2. A block with a written scope

The scope is one sentence, written before the block, naming the artifact you will produce. “Work on the migration plan” is not a scope. “Write the rollback section of the migration plan, three paragraphs, using Monday’s call notes” is. An unscoped block forces you to make planning decisions while you are trying to concentrate, and planning decisions are exactly what your brain will use to escape.

3. One interruption rule, decided in advance

Pick one rule and apply it without negotiation: phone in another room, notifications off, chat status set to focus, and a single capture note for anything that arrives. The rule is not “ignore everything”. It is “everything goes to the capture note and gets handled at 12:30”. A rule with an exception is not a rule; it is a suggestion that loses to the first person who says “quick question”.

4. A shutdown line

The last two minutes of the block are for writing one line: where you stopped, what the next action is, and what is still unresolved. This is the cheapest intervention in the system. Leroy’s finding is about unfinished tasks, and the shutdown line converts an unfinished task into one with a defined next step, which reduces how much of it follows you into the next hour.

How long should a deep work block be, and how do you ramp up?

Start at 50 minutes if you have not done this before, and move to 75 or 90 once 50 feels short. The 90-minute figure is a reasonable default: long enough to reach engagement, short enough to fit between commitments.

Ramp deliberately rather than optimistically:

Notice that week 4 schedules 7.5 hours and expects about four. That gap is not failure; it is the difference between calendar time and cognitive time. Track completed deep hours — blocks where you produced the artifact you named — and ignore scheduled hours entirely. A week with four honest deep hours beats a week with ten blocked hours and no output.

What do you do with the shallow work that still has to happen?

Shallow work does not disappear because you scheduled deep work. If you do not give it a home, it will take yours.

Three moves cover most of it. First, batch communication into two or three fixed windows instead of continuous availability; 15 minutes at 12:30 and again at 16:30 handles most of what arrives. Second, publish office hours: a 30-minute slot when anyone can bring questions, which converts random pings into a queue you control. Third, make status visible without a conversation — a task board, a weekly note, a dashboard — so that “what is the status?” stops being a reason to interrupt you.

Worked example: the cost of asking instead of looking

Take a 20-person team where each person sends and receives about 60 status-related messages a day. Assume 40% exist only because the answer is not visible anywhere: 24 messages per person per day, 480 across the team. At 90 seconds each to read, interpret, and reply — including the context switch — that is 720 minutes, or 12 person-hours per day.

At a blended cost of $55 per hour, that is roughly $660 per working day, about $158,000 in a 240-day year. The exact figures matter less than the shape: duplicated status traffic is one of the largest unmeasured costs in knowledge work, and it eats the hours you tried to protect.

What does a four-week build plan look like in practice?

Use the table above as the skeleton, and add one measurement and one removal each week.

• Week 1 — measure before you change anything. Log every interruption and every block over 30 minutes for five days. Most people find their real deep time is under 90 minutes a day, and that self-interruption outweighs other people.
• Week 2 — add the cue and the scope. Keep the calendar identical and change only how you start and what you write down. It is the highest-return week because it costs nothing politically.
• Week 3 — negotiate one protected morning. Ask for a recurring 90-minute block before 11:00, three days a week, and offer a trade: you will attend the 16:00 standup and answer within two hours during your communication windows.
• Week 4 — remove one recurring meeting or convert it to async. A 30-minute weekly meeting with eight attendees costs four person-hours a week, about 190 hours a year.

Track two numbers weekly: completed deep hours and how often you broke your own rule.

Why do most focus routines collapse in week three?

Because they were designed for an ideal week. Four failure patterns account for most of the collapse.

The block is too long, too early. A four-hour block on day one produces four hours of guilt and one hour of work. Start at 50 minutes.

The scope is a topic, not an artifact. Topics invite reading. Artifacts require decisions.

The routine has no answer for legitimate urgency. If a production incident can happen at any time, “do not disturb” is a fantasy. Define the path explicitly: one channel, one person, one threshold.

Nothing is measured. A routine you cannot count is a routine you cannot defend when someone asks for that 10:00 slot back. Two numbers — deep hours and rule breaks — are enough.

> Practical tip: keep a “distraction log” for one week. Write down what interrupted you and what you were about to do. You will usually find that 60–70% of interruptions arrive in the first ten minutes of a block, which is exactly when your brain is still looking for an excuse to leave.

Does a focus routine work for teams, not just individuals?

It does, but only if the team agrees on what silence means. An individual routine dies in a team that treats availability as a proxy for commitment. Three team-level agreements make individual routines survivable: a shared definition of “focus hours” (no internal meetings before 11:00, for example), a single place where status lives so nobody has to ask, and a norm that a message can wait two hours unless it is flagged urgent with a stated reason.

This is where tooling matters more than most productivity advice admits. When tasks, owners, dependencies, and progress reports live in one workspace, “where are we?” messages drop without anyone being told to communicate less. That is the problem Doitify is built for: tasks and sub-tasks, owners and due dates, dependencies, workload views, and progress reports in one system, plus an AI assistant that turns a stated goal into a plan. Doitify is our product, so treat that as a disclosure rather than a review — and note that a shared workspace only helps if the team keeps it current. A stale board generates more questions than no board at all.

When is deep work the wrong answer?

For some roles, protecting long blocks is not just impractical, it is harmful. Support engineers, incident commanders, and front-desk staff are paid to be interruptible; their value comes from responsiveness, and a focus routine would degrade it. Junior team members often need more feedback, not less, and a manager who disappears for four hours is trading their own output for someone else’s blockage.

The honest framing: deep work suits individual contributors and leaders whose output is documents, decisions, designs, analysis, or code. Everyone else is better served by reducing the cost of context switching than by eliminating interruption — rotating on-call, batching escalations, or staffing a shared inbox. Less romantic than a 90-minute block, and more useful.

There is also a trade-off inside knowledge work. A protected morning means someone else waits. If you protect four hours and the team’s approval queue stalls for a day, you have not created value, you have moved it. Keep the routine only if the work you produce in those hours is visibly better than what you produced in fragments. Check that quarterly, not annually.

Conclusion

Deep work is not a personality type, and a focus routine is not a productivity aesthetic. It is small infrastructure: a cue that removes the decision, a scope that removes the planning, a rule that removes the negotiation, and a shutdown line that removes the residue. Build those four things and you will get more from four protected hours than most people get from a full day of availability.

Start smaller than feels impressive: three 50-minute blocks, one written scope each, one interruption rule, one measurement. Then defend the result with data rather than conviction. The routine earns its place only if the work it produces is better than what the fragmented day was producing, and that is a claim you can test in four weeks.

If this post on focus routine was helpful, you might also enjoy Workplace Context Switching Statistics: The 2026 Data Guide and Time Blocking vs Time Tracking: What’s the Difference and Which Should You Use?.