For four of these six jobs, you should hire a person.

The Luddites are used as an insult now, which is convenient for anyone selling machinery. What they actually did was more precise than the insult allows: they broke particularframes — the ones producing cheap goods with unskilled labour — and left others alone.² Their objection was not that machines existed. It was about what was going to happen to the price of their work. On that narrow point they were correct.

Keep the useful half of that and throw away the rest. The useful half is that the sum only works out in a particular place, at a particular price of labour, and moving it somewhere else changes the answer.

Automation does not replace jobs. It replaces tasks, and whether a task is worth replacing depends on what that task costs where you are standing.¹ Every glossy payback figure you have been shown was computed in a place where the person doing that task is paid several times what they are paid here. That is not a small adjustment. It flips the answer.

The formula, which fits on one line

You need four numbers, and you already have three of them.

• S— what the person costs you a month, fully loaded. Not the salary. Salary plus everything.
• f— the fraction of that role the automation actually takes. Be honest and be pessimistic. It is never all of it.
• R— what the automation costs to run each month. Licences, messages, hosting, the person who fixes it when it breaks.
• B— what it costs to build, once.

Monthly saving is f × S − R. Payback in months is B ÷ (f × S − R). That is the whole model, and it is enough to kill most proposals in about ninety seconds.

The same build, two countries

Numbers here are mine and made up. Replace them.

Say a back-office person costs twenty-five thousand a month, fully loaded. Say the automation genuinely takes sixty per cent of what they do, which is generous. Say it costs two thousand a month to run and two and a half lakh to build.

Saving is fifteen thousand minus two thousand, so thirteen thousand a month. Two and a half lakh divided by thirteen thousand is about nineteen months.

Now do it where the same role costs the equivalent of three lakh a month. Sixty per cent of that is one lakh eighty. Take off the same two thousand running cost and the saving is one lakh seventy-eight thousand a month. A two and a half lakh build pays back in under two months.

Same tool. Same code. Same vendor deck. Nineteen months against two. Every case study you have been sent was written in the second column.

A nineteen-month payback is not a bad investment. It is a bet that nothing about your business changes for nineteen months, which is a much braver bet than it sounds.

Six jobs, run through the formula

These are the six a small business is most often sold. The verdicts below are reasoned from the formula on my assumed numbers, not measured across businesses. Yours may differ, and the section after this tells you when.

1. Keying invoices and statements. High volume, fully specifiable, mistakes are expensive and come back. Clears itonce volume is real — and note it clears on the cost of errors more than on the hours.
2. Appointment reminders and confirmations. Enormous volume, trivially specifiable, and it does not save labour so much as recover revenue that was walking out. Clears it comfortably, and it is the one nobody gets excited about.
3. Stock counting. The counting is not the hard part. Deciding what a discrepancy means is, and that is a person. Hire the person.
4. First-reply customer support. Half your questions are not really questions, they are somebody who is annoyed. Answering the easy half fast makes the hard half angrier, because they now know a fast reply was available. Hire the person.
5. Payroll and attendance.Low volume, once a month, highly exception-driven, and every exception is somebody’s pay. Hire the person. Buy them a decent tool.
6. Sales follow-up. The follow-up is not the value. The judgement about who to follow up and what to say is, and it is the part that cannot be specified. Hire the person.

Four out of six. And I want to be clear that this costs us money to write, because four of those six are things we are perfectly capable of building and would be paid to build.

What actually decides it, and it is not the hours

Notice what the two survivors have in common, because it is not “the boring ones.”

Neither of them wins on labour cost. Reminders win because a no-show is lost revenue and the automation recovers it at a volume no person could cover. Invoice keying wins because a wrong invoice costs far more than the minutes it took to type it. At Indian wages, hours are cheap and mistakes are not, so the automation that pays is the one that prevents an expensive event, not the one that saves time.

That is the whole local adjustment, in one sentence, and it is why imported payback maths misleads here. Where labour is expensive, saving hours is the win. Where labour is not, it usually is not.

When the four flip

Nothing above is a law. Three things move it, and if any of them is true for you, redo the sum:

• Volume ten times higher. f × S is per person, but volume decides how many people. At enough scale the fourth and sixth change sides.
• The mistake is very expensive. If one wrong stock figure costs a week of production, counting stops being about counting.
• You cannot hire. If the role is genuinely unfillable where you are, the comparison is not automation against a person. It is automation against the work not happening, and that is a different sum with a different answer.

Run the four numbers before your next conversation with anybody selling this, including us. If the payback is past about twelve months, the honest answer is usually a person and a better tool for them to use.