Studio
AI Consulting

What that one job is costing you, in your own numbers.

Not what it costs businesses like yours. What it costs you. Put four numbers in, see the arithmetic, and get an honest answer about whether it is big enough to be worth anyone fixing.

Chapter 02 of 08 · 8 min read

Why it has to be your numbers

There is a whole genre of statistic that exists to make you feel behind. Small businesses lose this many hours. Bad data costs the economy that many trillions. The numbers are large and they are about somebody else, and reading them produces a feeling rather than a decision.

We know why they do not work, because it has been measured. The single largest reason small businesses have not adopted AI is that they see no use for it in their business— 77% of non-adopters, in a U.S. Chamber of Commerce survey of 3,870 firms fielded in June 2025, reported by the JPMorganChase Institute on 14 April 2026. At firms under five people it is 82%. A statistic about businesses in general is answering a question those owners are not asking.

So here is the opposite. Four numbers you already know, and the arithmetic done in front of you.

Put your own numbers in

One repetitive job in your business. Not all of them, just the one you already have in mind. Every starting number below is deliberately low, so if yours are higher the answer goes up. Change any of them.

That job costs you, a year$10,977
Which is, over 18 months$16,466 and 828 hours
Show the arithmetic
  1. An hour of that person costs $20 — $3,500 a month divided by 176 hours, which is 22 working days at 8 hours.
  2. 2 × 6hours × 46 weeks = 552 hours a year. We use 46 weeks, not 52, to allow for leave and public holidays.
  3. 552 hours × $20 = $10,977 a year.
  4. Wages only. It does not count the mistakes that job produces, the things nobody got round to, or what those hours would have earned somewhere else. Those are real and we have not measured them, so they are not in the number.

What would have to change

A first fix starts at $9,000. Against $10,977 a year, roughly 82% of that time would have to stop for it to pay for itself inside twelve months, or 41% to pay for itself over two years.

Twelve months is our choice of hurdle, not a law, and it is a hard one — most businesses would accept longer for something they keep using. Both figures are the bar, and neither is a promise. They also assume the freed hours actually come off your wage bill or go to work that earns; often they get absorbed somewhere else instead, which is a real outcome and not one this arithmetic can see. We have not published a figure for the share of a job like yours that typically goes away, because we have not measured one. What the free call is for is deciding, together, whether 82% looks reachable in your business. If it does not, we will say so.

This runs entirely in your browser. Nothing you typed was sent to us or to anyone else, and nothing was stored. There is no request in this page that carries it.

What this deliberately leaves out

The number above is wages and only wages. That is a choice, and it makes the answer smaller than the truth. Three things it does not count:

  • The mistakes. A job done by hand at speed produces errors, and errors cost money in ways that never appear as a wage. We have not measured the error rate of your process and neither has anybody else, so it is not in the arithmetic.
  • The things that did not happen. The hours are not free time being wasted, they are hours that would otherwise be spent on something. What that something is worth is specific to you and we cannot compute it.
  • Your own time.If some of those hours are yours rather than an employee’s, the wage figure understates it badly, because your hour is the one that decides what the business does next.

We could put multipliers on all three and the number would get much bigger and much less true. A calculator that inflates its own answer is a sales instrument, and you would be right to distrust the rest of the page after finding one.

Every assumption in that calculator is visible and editable, and the starting numbers are deliberately low. If you think it is overstating your case, turn something down. It will still be your number.

The borrowed numbers, and why we are not using them

For completeness, here is the sort of figure we could have opened with, and what is wrong with it.

UK small-business owners report spending an average of 11 hours a week on administrative or finance tasks, in the SME Business Barometer run by American Express with Small Business Saturday UK across 1,000 owners, reported on 20 July 2026. That is a real survey with a real sample. It is also commissioned by a company selling to small businesses, its fieldwork dates and margin of error are not published, and it is about the United Kingdom.

The more dramatic versions are worse. The widely quoted figure that bad data costs the United States three trillion dollars a year traces to a vendor marketing infographic with no published methodology that anyone has been able to produce. The “manual data entry has a one percent error rate” line has no single source at all; the closest real evidence is a meta-analysis of 84 studies finding 0.29% for single data entry in clinical research, which is roughly a third of the figure everyone repeats.

We are showing you our working on other people’s numbers for the same reason we show it on yours. If we are willing to quote a statistic without checking it, you have no way to know which of our other claims we checked.

What to do with the answer

If the number came out small, believe it. Most small businesses using AI are spending very little: 63% of them spend between one and forty dollars a month, measured from real payments across 4.6 million businesses by the JPMorganChase Institute and published on 14 April 2026. That is not a starter tier people graduate from. For a great many businesses it is the correct and final answer, and a project would be worse than the subscription.

If the number came out large, the useful next question is not “what AI should I buy”. It is whether this particular job is the kind that a machine can take a real bite out of, or one of the ones where it cannot. Those are genuinely different, and the difference is not obvious from outside. That is the next chapter but one.

Either way, you now have a figure you can defend, in a conversation with a vendor, with your accountant, or with yourself at eleven at night. That was the point of the exercise, and you did not have to give anyone your email address for it.

Where these numbers came from

Every figure on this page, with what it is and where it is from. If a number is illustrative rather than measured, it says so here and it says so in the text.

  1. 77%Of small businesses that have not adopted AI, the share who say they see no use for it in their business. This is why a general number about other businesses does not work, and your own does.U.S. Chamber of Commerce survey of 3,870 U.S. small businesses under 250 employees, fielded June 2025, reported by the JPMorganChase Institute, 14 April 2026.From a named study
  2. 63%Of small businesses that use AI, the share spending between one and forty dollars a month on it. Measured from real payments, not from what owners said.JPMorganChase Institute, published 14 April 2026, from de-identified transaction data across 4.6 million businesses.From a named study
  3. 11 hours a weekAverage time UK small-business owners report spending on administrative or finance tasks. Quoted here only as a sense of scale, and it is a commissioned survey.SME Business Barometer, American Express with Small Business Saturday UK, 1,000 UK business owners, reported 20 July 2026. Fieldwork dates and margin of error are not published.From a named study
  4. 46 weeksThe working year the calculator uses, rather than 52, to allow for leave and public holidays. Chosen by us to keep the answer conservative.Our own choice, stated so you can disagree with it. 176 full-time hours a month, used to convert a salary into an hourly cost, is 22 working days at 8 hours.Illustrative
  5. 0.29%Measured error rate for single manual data entry, against the round one percent that circulates without a source. Roughly a third of the figure everyone repeats.Garza, Williams, Ounpraseuth and others, Error Rates of Data Processing Methods in Clinical Research: A Systematic Review and Meta-Analysis, 84 manuscripts in the final meta-analysis, version 2, 21 December 2023. It measures clinical research, not your back office.From a named study
  6. twelve monthsThe payback window the calculator measures the bar against. It is a deliberately hard hurdle and it is ours, not a standard. The two-year figure is shown beside it because most businesses would accept longer for a system they keep using.Our own choice, stated on the page next to the number it produces rather than buried in the arithmetic.Illustrative