Is it worth automating? Work it out on the back of an envelope

You don't need a business case. There's no committee, no procurement review, no finance director to convince — it's your money, and the question is simply whether this pays for itself or whether it's another monthly bill you'll resent by March.

That question has an answer, and you can get most of the way to it on the back of an envelope with numbers you already have. Here's the sum, the costs that never appear on anyone's quote, and the three ways this arithmetic quietly flatters itself.

Before anything else: count it for two weeks

Every argument you will ever have about whether the thing was worth it comes down to a number that either got written down before you started or didn't.

Pick the job. Count it. How many after-hours calls come in, how many you got back to, how many turned into work. How many delivery slips a week, how long one takes to key in, who does it. How many exceptions you found the same day against how many you found when a customer rang.

Two weeks, on paper if that's what you've got. Do it before anything changes, because the “before” number stops existing the moment the process changes and you cannot reconstruct it afterwards. And count for long enough to see a normal and a bad week, because a project that starts in a quiet fortnight and reports the return of normal volume as a benefit isn't lying — it just has no way of knowing better.

The counting is also the cheapest way to talk yourself out of a bad idea, which is worth as much as talking yourself into a good one.

The sum

Three things go on the benefit side, in descending order of how much I'd trust them.

Hours, at what an hour actually costs you. Not the wage. The wage plus payroll costs plus the share of the overhead that person carries — the number your accountant would use, which is meaningfully bigger than the number on the payslip. Multiply by the hours the job takes each week, then by the weeks you actually work.

There's a catch on this line and it's the one that sinks most of these sums, so it gets a section of its own below.

Work you're currently losing. This is the one people leave out, and in trades it is often larger than the hours. If calls after five go to voicemail and you ring back in the morning, some of those callers have already booked somebody else. You know roughly how many calls come in out of hours from your two weeks of counting. You know what an average job is worth to you and roughly what share of enquiries you close. Multiply the three and you have an estimate of what the voicemail greeting costs you a year.

That estimate is soft — you can't know how many of those callers would have booked you — so treat it as a range rather than a number, and write down the assumption next to it.

Mistakes that cost money. The two pallets nobody noticed until the customer rang. The return visit because the part on the van was wrong. The submittal that went out with last month's revision. Put your own cost against one of those, times how often it happens.

Risk sits below all three: fines you didn't get, claims that didn't happen. Worth mentioning to yourself, never worth building the case on, because you're counting things that didn't occur.

The catch on the hours line

Say your two weeks of counting shows the keying takes a morning a week. That morning is a real thing, and it becomes money only if one of three things is true.

You don't make the hire you were about to make. You reduce hours somebody was being paid for. Or that morning goes on work that brings money in — quoting the jobs sitting in the inbox, chasing the invoices nobody chased, getting to site.

If none of those is true, the morning gets absorbed. The work is genuinely more pleasant, your office manager is genuinely less fed up, and the P&L is exactly where it was. That's not nothing — people leave over paperwork — but don't put it in the money column and then be confused when the money doesn't show up.

Say out loud which of the three you're claiming. In a small business the honest answer is usually the third, and it's the most defensible, as long as you can name the work that fills the gap.

What goes on the cost side, including what nobody quotes you

The bit you'll be quoted for is the software and the build. Three things sit underneath it and together they usually outweigh the line item.

Connecting it to what you already run. Reading a delivery slip is the easy half. Getting the numbers into the system you already use, in the shape it wants, without breaking whatever else writes to it — that's the half with the hours in it. If a quote covers the reading and not the writing, it covers half a job.

The changeover. A few weeks of running both ways at once, somebody learning it, a period where things take longer than they used to. If your sum assumes full benefit from week one, it's wrong by however long that adjustment takes.

Keeping it alive. This is the one I'd argue about with anyone selling you a one-off build. Your supplier changes their delivery note layout. The phone provider changes something. A new customer's paperwork looks like nothing the system has seen. None of it announces itself; it shows up as the thing getting quietly worse. Budget a standing monthly for somebody watching it. A quote that shows the cost dropping to nothing after the build is describing a system nobody is looking at, and I price my own work as build plus monthly for exactly this reason.

Then add your own time, which is real even though nobody invoices you for it.

Three ways this sum flatters itself

It takes the easy ones and leaves you the hard ones. Automation rarely removes a job; it removes the straightforward instances. If the tidy slips get handled and the photographed, water-stained, half-legible ones still come to a person, the average time on what's left goes up. Model the work you're left with, not just the work you removed. This one surprises people badly, because the headline hours look great and the person doing the residue is having a worse week than before.

Some of the improvement would have happened anyway. You also hired someone, or the season turned, or the supplier finally started sending PDFs instead of faxes. The new system gets credit for all of it because it's the visible change.

Counting the same win twice. Faster handling and fewer hours are frequently added together when one caused the other. So are “errors avoided” and “rework avoided” when they're the same error.

Numbers that move without anything improving

Be suspicious of any measure you can improve by changing your own behaviour, because eventually it will be improved that way and nobody will have meant to cheat.

Calls handled without a person counts the caller who gave up in frustration the same as the one who got what they needed. Pair it with whether those callers rang back, or it tells you nothing.

Time per document falls when the system works, and also when the hard documents got quietly routed to somebody else.

Usage counts tell you people opened it, not that it helped. If you've told everyone they have to use it, usage will look excellent.

For whichever number you plan to watch, name the second number that would catch the first one lying, and look at both.

The figure that actually decides it

Not the return percentage. The break-even.

Take your biggest assumption — how many of those after-hours callers you'd genuinely have won, how much of the keying really goes away — and ask how far it can fall before the thing stops paying for itself. If it only works when every assumption lands at the good end simultaneously, you've learned something valuable for the price of an afternoon.

Then compare it against what else that money does. A monthly is a monthly whether it goes to software or to hire. The relevant comparison isn't “AI versus nothing”, it's this against a part-time person in the office, or a second van, or an apprentice. Sometimes the honest answer is the apprentice.

Decide now what would make you stop

Write down the condition under which you'd switch it off. “If I'm still checking every summary by hand in three months, this isn't working.” “If it hasn't caught a short delivery by the end of the quarter, stop paying for it.”

A thing that can only ever be declared a success is a thing you'll be paying for in five years out of embarrassment. Deciding the stopping condition while you're calm costs nothing, and it's the single most useful sentence in any of this.

Afterwards, report against what you originally said

Keep the envelope. When the numbers come in, put them next to what you predicted and leave the prediction unedited.

Variance in your favour is as informative as variance against you: a benefit that comes in far above what you projected usually means an assumption was wrong, not that the thing overperformed, and knowing which makes the next decision better. Quietly restating the prediction as the results arrive turns the whole exercise into a story you tell yourself.

Do the sum

I built a calculator that runs this arithmetic for you: the ROI calculator. It asks what an hour costs you, how many hours a week go on repeat work, and which job you'd hand over first, then gives you a payback period in months alongside the three-year figure. It'll email you the breakdown to keep.

If it comes out marginal, that's a real answer and you've spent ten minutes. If it comes out clearly worth doing, book a call and bring the numbers you already have. Tell me which job goes wrong most often and I'll tell you whether I'd take it on — including when the answer is that a phone message and a fortnight of counting would get you most of the way there without me.