The honest maths of business automation

Most articles about AI return on investment quote percentages from vendor case studies. Those numbers are close to useless, because they describe somebody else's business with somebody else's cost base.

Here is a framework you can apply to your own.

The three costs

Build. A well-scoped automation for a single process typically lands somewhere between a few thousand and the mid tens of thousands, depending on how many systems it has to touch. The integration work, not the AI, is almost always the largest line.

Run. Model usage, hosting and any third-party services. For most SME automations this is surprisingly small, often tens of pounds a month rather than hundreds. Voice is the exception, because telephony and speech processing carry real per-minute costs.

Maintain. This is the one nobody budgets for. Your suppliers change their forms. Your prices change. A system you integrated with releases a new version. Budget something like fifteen to twenty per cent of build cost annually, and you will not be caught out.

The return side

Be honest about the two different kinds of saving.

Hard saving is money that stops leaving the business. Overtime you no longer pay. A temp you no longer need at month end. A subscription you can cancel. These are real and they show up in the accounts.

Soft saving is time returned to people who then do something else. This is usually the larger number and it only becomes real if the freed-up time goes somewhere valuable. Four hours a week returned to an owner who uses it to win work is enormously valuable. Four hours returned and absorbed into general slack is worth nothing.

Be precise about which kind you are claiming. Boards and bank managers can tell the difference.

A worked example

A business missing roughly twelve calls a week outside working hours. Historically, around one in six of those callers would have become a customer. Average first job value four hundred pounds.

Twelve missed calls, two converted, eight hundred pounds a week, call it forty thousand pounds a year of enquiries currently walking away.

A voice agent to capture them: perhaps twelve thousand to build, two hundred a month to run, two thousand a year to maintain.

Even if the agent only converts half as well as a human would, the payback period is a matter of months. That is the sort of maths that is worth doing before you commit, and it is the sort that survives scrutiny afterwards.

The costs people forget

Change management. People need training, and productivity dips for a fortnight while everyone adjusts. Budget the time.

Data cleaning. Almost every project discovers that the data is worse than anybody believed. This is normal and it takes longer than you think.

The exception cases. Automating eighty per cent of a process is usually straightforward. The last twenty per cent is where the cost lives, and frequently the right answer is to leave it to a human rather than chase perfection.

The question to ask

Not "what is the ROI of AI", which is unanswerable. Ask "what is this specific process costing us today, and what would it cost to change it".

That question has a number attached, and the number is either compelling or it is not.