How to Run a Time Study on Yourself

Why this matters

Every owner already has a theory about where their week goes, and the theory is wrong in a predictable direction: it overstates the work you find meaningful and understates the work that arrives in ninety-second pieces. Ask an owner how much of the week goes to answering other people's questions and you will hear "an hour or two." Measured, it is routinely five to ten times that, because nobody remembers the twenty-third interruption of a Tuesday.

The problem is that a badly run time study produces numbers that feel authoritative and are not. You then reorganize your week around them and get nothing back. This article is about getting data you can actually act on. A sibling article covers what to do with the result once you have it, so spend your effort here on the measurement.

Step 1: Pick the method that survives your worst day

Three methods work. Pick by how chaotic your week is, not by how thorough you feel.

The default worth committing to: a continuous log at 30-minute resolution for two weeks, plus a separate interruption tally, with a reconstruction pass at the end of each week as a cross-check. Thirty minutes rather than fifteen because a fifteen-minute log has to be updated so often that the logging itself becomes the interruption you are trying to measure.

If you know you will not sustain that, run interval sampling for three weeks instead. A sustained mediocre method beats an abandoned good one, and the most common outcome of an ambitious time study is four solid days and then nothing.

Step 2: Fix the codes before you start, and make them mutually exclusive

The single biggest destroyer of a self-run time study is inventing categories while you log. By Thursday you have nineteen of them, half overlap, and you cannot add anything up.

Write six to eight codes before day one. Every code must answer one question: what kind of work was this. Not what it was about. "The Henderson job" is a topic, not a code, and it will absorb four different kinds of work.

A workable starting set for an owner still running calls:

• Field - on the tools, doing the work
• Drive - moving between sites
• Routing - answering someone else's question so they can proceed, including approvals
• Selling - estimating, quoting, walking a job with a customer
• Admin - invoicing, payroll, ordering, paperwork
• Customer - conversations not attached to a specific job in progress
• Owner work - planning, numbers, hiring, anything about the business rather than in it

Test the set before you use it by walking through yesterday and coding it. If any half hour genuinely fits two codes, your codes are broken and you fix them now, not in week two. The usual collision is Routing and Selling: a call that is a customer question about an outstanding quote could be either. Pick a rule and write it down - if you are the only person who could have answered it, it is Selling, otherwise it is Routing - and apply it every time.

Add one flag, not a second code. On each logged block, mark it only-me, could-train, or should-not-be-mine. Three values, marked in the moment, when you still remember whether anyone else could have handled it. Reconstructing that flag later is guesswork and it always flatters you.

Step 3: Count interruptions separately from time

Time and interruptions are two different measurements and a log that mixes them loses both. A thirty-minute block containing nine interruptions and a thirty-minute block containing none look identical in the time column, and they are not remotely the same half hour.

Run a tally alongside the log. One mark every time someone pulls your attention to something that was not what you were doing: a call, a text, a walk-up, a question shouted across the shop. Do not write what it was, just the mark, or you will stop doing it by Wednesday. Once a day, at a fixed time, note roughly how many of the day's marks were questions somebody could have answered from a written rule. That single ratio is often the most useful number the whole study produces.

Step 4: Sample two weeks that are not the same kind of week

One week is not a sample, it is an anecdote. Small shops have enormous week-to-week variance, and whichever week you pick will be either unusually busy or unusually quiet, because there is no such thing as a normal week in a service business.

Two weeks is the working minimum, and they should differ. If your trade is seasonal, do not run both weeks inside a peak. If your first week is unusually quiet, say so in your notes and run a third.

Do not schedule the study for a week you have already arranged to be clean. An owner who clears the decks first is measuring a week that does not exist.

Step 5: Reconcile the hours before you interpret anything

This is the step that separates a study you can act on from a pile of numbers.

Add your logged hours. Compare them against elapsed working hours, meaning from when you started to when you stopped, minus real breaks. They will not match, and the gap is data.

Under about 10% unaccounted, treat the study as sound and move on. Over about 10%, do not adjust the numbers to close the gap. Instead, look at when the unaccounted time sits. Missing hours cluster, and they cluster in exactly the periods that are too fragmented to log, which means whatever category dominates those periods is being undercounted in your results. In practice the missing hours are almost always Routing, because a two-minute interruption is the one thing nobody stops to write down.

If the gap runs over roughly 25%, the study is not usable as proportions. Rerun it with interval sampling, which does not depend on you remembering to log during chaos.

Step 6: Discount the first two days

You will run cleaner while you watch yourself. Everybody does. Batches of Routing get deflected, admin gets done in one sitting, the phone gets ignored once or twice. The effect is real and it fades by about day three.

Treat day one as calibration and exclude it from your totals, or keep it and note it. Do not attempt a numerical correction, since you cannot know the size of the effect. The important consequence is directional: your measured Routing is a floor, not an estimate. If the study says you spend nine hours a week answering other people's questions, the true figure is that or higher.

A worked study: two weeks, one owner

Six techs, an office manager, owner still running about two calls most days. Continuous log at 30 minutes, interruption tally, two weeks in different parts of a month.

Reconciliation first. Elapsed working time across the two weeks: 112 hours. Logged: 101 hours. Unaccounted: 11 hours, about 9.8% of the 112 elapsed. Inside tolerance, so the proportions are usable, and the missing 11 hours sat mostly in mid-afternoon, which is when the phone runs hottest.

The 101 logged hours across two weeks:

The delegability flag across the same 101 hours: only-me 24, could-train 46, should-not-be-mine 31.

The interruption tally: 213 marks over 10 working days, about 21 a day. On the daily note, roughly 147 of the 213, about 69%, were questions somebody else could have answered from a written rule.

Reading it. The headline is not Field at 33.7% of logged hours. Every owner already knows they are on the tools too much, and that number is not actionable this quarter because the shop cannot replace those hands by Friday.

The actionable finding is Owner work at 3 hours across two weeks, about 3.0% of the 101 logged hours, roughly 1.5 hours in a typical week - against Routing at 9.5 hours in a typical week. The person responsible for where the business goes is spending about six times as many hours a week being a lookup service as steering. And 69% of the interruptions driving that Routing number are answerable from a written rule that does not exist yet.

Notice what the flag adds. Only-me came to 24 of 101 logged hours, so roughly a quarter of measured working time genuinely required this specific person. That is the number that tells you whether a management layer is the answer or a written rulebook is. At 24 hours in 101 over two weeks, a rulebook and a raised authority ceiling will move more than a hire will, because the constraint is not hands, it is that the answers only exist in one head.

When a study cannot be trusted

Four conditions that invalidate a result, each with its own tell:

• The reconciliation gap ran over about 25%. Proportions are unusable. The categories that survive logging are overstated relative to the ones that do not.
• The codes changed mid-study. If week two used a category week one did not, you have two studies, not one. Recode week one against the final set, or throw week one out.
• Both weeks were the same kind of week. Two peak weeks tell you about peak. Say so, and do not generalize the ratios to the year.
• You cleared the calendar first. You measured an ideal week. Rerun without preparation and expect the Routing share to rise by a lot.

One more, harder to see: a study you ran because you already knew what you wanted it to prove. If you started this to justify a hire, check whether your codes are drawn so that the hire's work lands in one big obvious bucket. Redraw them, recode two days, and see if the answer holds.

References

• See related: Owner Time Audit: Where the Week Goes
• See related: Where an Owner's Hours Actually Go
• See related: The Context-Switching Cost in a Small Shop
• Trade-standard practice for work sampling and self-measurement in small businesses