Tracking time across four clients in one day
A single-client day is easy to track. Start in the morning, stop in the evening, subtract lunch.
A four-client day is where every method falls over, and it is also the day where getting it wrong is most expensive, because the errors do not cancel out. They land on whichever client you happened to be thinking about when you reconstructed the afternoon.
Why timers specifically fail here
The timer model assumes a switch is an event you notice. On a fragmented day it is not.
You are in client A’s codebase. Client B messages about something urgent. You answer, which takes four minutes of typing and eleven minutes of thinking about their problem. Then you go back to A. At no point did you consciously begin or end a work session, and the timer is still on A.
Multiply by nine switches in a day. Some of them you catch. The ones you catch are the long ones, because a forty-minute detour eventually makes you notice. The ones you miss are the short ones, and short ones are most of them.
The result is not random noise. It is a systematic transfer of small amounts of time from the client who interrupted you to the client you were already billing. Both invoices are wrong, in opposite directions, every day.
The three approaches that people try
Timers, disciplined. Start and stop on every switch, no exceptions. It works, for about three weeks. It fails not because people are lazy but because it asks you to interrupt your own concentration at precisely the moments when concentration is hardest to hold.
Blocks, planned in advance. Assign mornings to A and afternoons to B, and bill the blocks. Clean, and it works if your clients cooperate. Most do not, because urgency does not respect your calendar.
Reconstruction at the end of the day. Sit down at six and write out where the day went. Honest, and much better than at the end of the week. Still relies on remembering nine switches, several of which lasted four minutes.
None of these is stupid. They are all reasonable responses to a real problem. They just all depend on the same fragile thing.
What works instead
Separate the two jobs that a timer was doing at once.
A timer is trying to do capture and classification in the same action: recording that time passed, and recording whose it was. Capture has to happen live, because time that was not captured is gone. Classification does not.
So capture everything automatically, and classify afterwards.
The day gets recorded as it happens: which application, which window, which document, how much activity, and when each stretch began and ended. Nobody presses anything. At the end of the day, or the end of the week, you assign the stretches to clients, with the evidence in front of you.
The four-minute interruption from client B is in the record, because it was captured with everything else. You may still decide it was too small to bill. But it is a decision you are making, rather than an omission you never saw.
Assigning is faster than you expect
The objection is usually that reviewing a whole day sounds like more work than pressing a button nine times.
In practice it is not, for two reasons.
Most of a day resolves in a handful of gestures. Six hours in one repository is one assignment, not six. The fragmented parts are the parts that need attention, and they are also the parts that a timer was never going to catch.
And the assignments repeat. Once a particular project directory, application or client domain has been assigned to a client, the same evidence maps to the same client next time. What starts as a review job becomes a confirmation job within about a fortnight.
Two things worth setting up on day one
Make billable a property of the client, not a decision per entry. Set the rate and whether the work is chargeable once, on the project. Then internal work, admin and idle stretches stay off the invoice without you deciding anything nine times a day.
Keep the idle time visible. On a fragmented day the gaps matter. Forty minutes with a client’s document open while you were in a call about a different client is not forty minutes of their work. A record that separates active from idle lets you make that call honestly, which is worth more than the forty minutes.
The part clients actually care about
There is a version of this that is about accuracy and a version that is about defensibility, and they are not the same.
Accuracy is billing the right number. Defensibility is being able to answer, three weeks later, what the six hours on the 14th were, without a note of apology in your voice.
On a multi-client day, defensibility is the harder one and it is the one that keeps clients. A timeline showing which application, which document, and how much of that stretch had activity in it is an answer. A number you remember typing is not.
If you want to see the shape of that, how automatic time tracking works covers what actually gets recorded and what does not. If you are weighing tools, the comparison pages put the automatic and manual options side by side, with the dates the figures were checked.
The short version
Do not try to be more disciplined. The discipline was never the problem, and a method that requires perfect attention on your worst day is not a method.
Capture the whole day without deciding anything, then decide once, with the record in front of you. On a four-client day that is the difference between an invoice you can defend and an invoice you hope nobody asks about.