Automatic timesheets shouldn't require surveillance

It's four o'clock on Friday and you are trying to remember Tuesday.

Not in general — specifically. Was the Acme call ninety minutes or two hours? Did the contract review happen Tuesday afternoon or Wednesday morning? There was something between lunch and the standup, and it was billable, and it is gone.

So you do what everyone in professional services does. You open your calendar, your sent mail, and your browser history, and you reverse-engineer a week you already lived. Then you round down, because you would rather under-bill than defend a number you're unsure of.

That rounding-down is the part nobody puts a figure on. Reconstructed timesheets don't fail randomly — they fail in one direction. Hours you can't quite place become hours you don't quite charge for. Across a year, in a firm that sells time, that is not a rounding error.

The industry's answer, and what it costs

The category's response to this has been automation, and for the most part automation has meant one thing: install an agent on the machine that records every application, window title, document, and URL, continuously, all day. At the end of the week it hands you a draft timesheet assembled from that record.

It works. That's the uncomfortable part — it genuinely does reduce the Friday afternoon problem.

But look at the trade. To stop reconstructing your week, you agree to have your working life recorded. And the recording doesn't distinguish between the parts you'd happily show anyone and the parts you wouldn't.

It captures more than work. The tracker doesn't know that the tab you opened at lunch was your bank, or a job listing, or a symptom you were worried enough to look up. It captures everything and trusts a redaction layer to sort it out afterwards.

It quietly implicates your clients. Much professional work happens under confidentiality agreements. A tool that logs document titles and URLs is now holding fragments of that material, and in most engagements nobody has told the client it exists. That is a conversation you'd rather not have retroactively.

It changes the team, whether or not anyone looks. The existence of a record is enough. People begin working for the log — closing the reading tab, keeping the "right" document open. You wanted a timesheet and you built a stage.

And often it answers the wrong question. The question isn't what was on someone's screen at 2:14pm. It's which client should be billed for that hour. Screen contents are a noisy, lossy proxy for that. A person reading a contract and a person reading the news look remarkably similar to a window-title logger.

The data you need is already written down

Here's what the surveillance approach overlooks: for most client-facing work, the answer already exists in structured form, and you created it deliberately.

Your calendar.

"Acme Corp — Q3 roadmap review — 90 minutes — four attendees, two of them acme.com" carries more billing-relevant signal than six hours of window titles. It has a client, a duration, a subject, and participants. It was written on purpose, by a human, in advance.

This differs from activity capture in kind, not merely in degree:

• It's bounded. A calendar has edges. It covers meetings, and nothing else. Continuous capture has no natural stopping point, which is exactly why scope creep is its default failure.
• It's already consented. Everyone in that meeting agreed to be in it. No new surface of information about anyone was created.
• It's already structured. No inference required to know who the client was — the invitation says.
• It's inspectable. You can look at your calendar and see precisely what the system sees. Nobody can meaningfully audit what a screen recorder collected last Thursday.

Where this approach genuinely falls short

If we only made the case for our own side, you'd be right to discount the rest.

Calendar-derived time tracking misses solo work. Three hours in a document, a long stretch of code, an afternoon of thinking — none of it appears on a calendar, and none of it gets captured. Activity tracking catches all of that, and for people whose billable day is mostly heads-down individual work, it will produce a fuller picture. That's a real advantage and it should be stated plainly.

So this isn't a strictly-better argument. It's a different trade: less coverage in exchange for far less intrusion. For consultants and agencies — people whose work is largely meetings, calls, reviews, and workshops — the coverage gap is small and the intrusion saved is large. For a solo developer working eight uninterrupted hours in an editor, the trade may well go the other way.

Knowing which one you are is more useful than being sold either.

The line that matters more than the data source

There's a second question underneath all of this, and it survives regardless of where the data comes from: should the system log the entry, or propose it?

Even with perfect data, we'd argue for proposing. Three reasons.

A time entry isn't a measurement. It's a claim you make to a client, on an invoice, with your name on it. Somebody should be willing to stand behind each one. Automatic logging removes the moment where a person takes that responsibility.

Confirmation costs almost nothing. Accepting a correct suggestion takes a second or two. Against the cost of a wrong entry reaching a client's finance team, that is not a meaningful tax.

And systems that log automatically train you to stop reading them. This is the one people underestimate. Anything correct often enough becomes invisible, and the errors then ship silently — which is worse than the Friday reconstruction, because at least on Friday you were paying attention.

The principle is old and it's a good one: the system proposes, the person disposes. Anything else means a professional judgment about what a client owes has been quietly handed to a heuristic.

The axis we've been arguing on is the wrong one

The debate has been framed as automatic versus manual, as though the only question is how much typing you can eliminate.

The better question is what you have to give up to get it. Some automation costs you nothing but a confirmation click. Some costs you a permanent record of everything you did at work, plus whatever your client would think about that arrangement if they knew.

Those are not the same product, and they shouldn't be sold under the same word.

You shouldn't have to be watched to stop reconstructing your week.

---

Disclosure: We build Timsio, a time-tracking and invoicing tool that takes the calendar-derived approach described here. That's the bias to read this with. The argument stands or falls on its own merits — and if activity tracking genuinely fits your work better, use it.