What it actually costs to automate a small business

Most automation pricing conversations go badly because both sides are arguing about the wrong number. The buyer is comparing the quote against a monthly software subscription. The builder is comparing it against a day rate. Neither of those is the relevant figure.

The relevant figure is what the work costs you now, and it is almost always larger than people expect, because nobody has ever added it up.

Count what it costs today

Take one process. Write down how long it takes each time, and how often it happens. Multiply. Then multiply again by the loaded hourly cost of whoever does it, which is their wage plus the roughly thirty percent that employment actually costs on top.

A task that takes forty minutes and happens daily, done by someone costing thirty five dollars an hour loaded, is about six thousand dollars a year. Not a large task. Not a task anyone complains about. Six thousand dollars.

Now add the parts that do not appear on a timesheet:

• The error rate. Manual re-entry produces mistakes. Each one costs somebody an afternoon of untangling, and occasionally costs a customer.
• The context switch. The interruption is more expensive than the task. A five minute job that lands eight times a day is not forty minutes, it is most of an afternoon of broken concentration.
• The bus factor. If one person does it and knows the exceptions, that knowledge is an unfunded liability sitting in one head.
• The work that never happens. The follow-ups nobody sends because there is no time. This is usually the biggest number and the hardest to defend, so I leave it out of any business case I present.

What a build actually costs

Roughly, and honestly, for a small business:

• A single workflow, well built and supervised: a low four figure sum. Two to three weeks, most of it the supervised run rather than the writing.
• A connected system of several workflows: mid four to low five figures. Four to eight weeks.
• Running cost afterwards: usually tens of dollars a month. Model calls, a small server, maybe a scheduler.
• Supervision, if you want it: a few hundred a month, and genuinely optional.

Against a six thousand dollar a year process, a two thousand dollar build pays back in four months and then keeps paying every year after. That is the whole business case, and it does not need a spreadsheet to be convincing.

The costs nobody quotes you

Three of these are real and routinely omitted.

The first month is not free. Prompts need tuning to your voice, edge cases surface that nobody described, and the exceptions your team handles automatically turn out to be undocumented rules. Budget attention for that month.

Automation rots. APIs change, tokens expire, platforms move endpoints. A system nobody maintains degrades over about a year. This is either a retainer or an afternoon of your own time every couple of months, but it is not zero.

Somebody has to own it. Not maintain it. Own it, in the sense of being the person who notices the alert. If nobody owns it, the monitoring is decoration.

When not to automate

I turn down work on these three grounds more often than people expect from someone selling automation.

The process is about to change. Automating a workflow that is being replaced next quarter is money set on fire. Wait.

It runs less than monthly. Something done four times a year rarely justifies a build, and it will have drifted between runs anyway.

The judgement is the job. If the work is deciding rather than executing, automation can gather the inputs and lay them out, but it should not decide. Pricing exceptions, hiring, anything with a legal edge. Automate the preparation and leave the call to a person.

Where to start if you are not sure

Pick the single process that annoys your team the most, and count it properly for one week. Actual minutes, actual frequency.

That number is either obviously worth fixing or obviously not, and either answer is worth the week it took to get. Most people are surprised, and most are surprised in the same direction.