The Work That Happens Away From a Desk

Your field team records every job twice. Once on site, however they can, on a paper docket or in a phone note or in a message to the supervisor. Then again that evening or the following morning, at a desk, into the system that actually counts.

The second time is where the detail goes missing. Not deliberately. It goes missing because eleven hours have passed, because the docket got wet, because the photo is in a message thread and the note is in a different one, and because the person doing the retyping is doing it at the end of a long day and wants to go home.

What gets lost between the site and the system

The extra hour nobody logged. The material that was swapped on the day. The thing the client asked for standing in the doorway, which everyone remembers agreeing to and nobody wrote down. The photo that would have settled the argument three months later.

Each one is small. Together they are the difference between a job that made money and a job you think made money. When the margin on a job comes out wrong at the end of the quarter, this is usually where it went, and it is invisible because nothing was ever recorded incorrectly. It was simply never recorded at all.

Why the desk step survived

Because the alternative was worse. Getting a field team to enter structured data on site means asking somebody wearing gloves, in the rain, standing on a roof, to fill in a form designed by somebody sitting down. It gets abandoned in the first fortnight and everybody goes back to the paper docket, which at least works.

So the desk step is not laziness or bad process. It is a rational response to software that was never built for the conditions the work happens in. The general purpose field apps solve this by having a field for everything, which is exactly the thing that makes them unusable.

The compliance load underneath it

This sits on top of an administrative burden that is already heavy. The Canadian Federation of Independent Business, in its Red Tape Report published in January 2025, found that businesses spent 768 million hours on regulatory compliance in 2024, and that the average small business owner lost the equivalent of 32 business days to it. The smallest businesses pay over five times more per employee than the largest ones.

Every one of those hours is somebody at a desk, transcribing something that already happened, into a format somebody else requires. The field docket is the same problem wearing work boots.

What AI changed about this

The form is no longer the only way in. A person on site can now say what happened, out loud, in the order it occurred to them, and get structured data at the other end. Not a transcript sitting in a folder, but the hours against the right job, the material against the right line, the variation flagged as a variation.

Photographs stopped being dead weight too. A picture of a serial plate, a meter, a delivery note or a damaged panel can now be read and filed against the job rather than sitting in a camera roll waiting for somebody to caption it. That was the single biggest reason photos never made it into the record.

And the checking moved forward. A missing reading, a job closed with no photo, an hour count that does not match the visit length: these can be caught while the person is still on site and can fix it in thirty seconds, rather than a fortnight later when nobody can remember.

The argument three months later

The other cost of the desk step only appears when something is disputed. A client says the extra day was never agreed. A supplier says the delivery was short. An insurer wants to know the condition of the plant room on the day you arrived.

At that point the record is a docket in a box, a message thread on somebody's personal phone, and a memory. The person who was actually there has left, or cannot remember, and you settle for less than you should have because proving it costs more than the difference.

Evidence captured at the moment it existed removes that entirely, and it is worth more than the time saved on the retyping. Most businesses only find this out once, expensively.

What AI will not do for you

It will not decide what matters on a job. Which fields are worth capturing, what counts as a variation, what has to be signed before the team leaves: those are your rules and they have to be written down by somebody who knows the work. A system built without them just produces tidier noise.

It will not survive a bad connection either, and most of this work happens in basements, plant rooms and half built sites. Anything built for field teams has to work with no signal and reconcile later. That is engineering, not intelligence.

What to build instead

Start with one job type, not all of them. The team does the same eight things most weeks and one of those eight is where the money leaks. Build for that one and let it prove itself before anything else is added.

Design it for one hand and bad light. If it cannot be completed standing up in under two minutes, it will be abandoned, and an abandoned app is worse than the paper it replaced because now there are two records and neither is complete.

Capture the evidence at the moment it exists. A photo taken on site and filed automatically is worth more than a perfect form completed at nine that evening from memory.

Then delete the desk step deliberately. If the evening retyping is still possible, it will keep happening, and you will keep having two versions of the truth.

We build these as applications shaped around one team's actual day rather than configurations of a general purpose field tool, because the reason the general purpose ones get abandoned is that they were shaped around everybody's. See Custom Application.

Sources

Canadian Federation of Independent Business, Canada's Red Tape Report, 7th edition, January 2025. cfib-fcei.ca