What AI automation actually costs a small business

The most common question we get is some version of "what does this cost?" — and the most common answer in this industry is "it depends," followed by a discovery call. That is a sales tactic, not an answer. Here is the actual shape of the numbers.

Three costs, not one

Every automation project has three separate costs, and vendors who quote you one number are hiding two of them.

1. 01Build cost — one-time, to design and ship the thing.
2. 02Run cost — monthly, for hosting and AI usage. This is usually far smaller than people expect.
3. 03Change cost — what it costs to modify it in a year when your process changes. This is the one nobody asks about and the one that hurts.

Realistic build ranges

For a small or mid-sized business, in our market, for work that goes to production rather than staying a prototype:

• A single scoped workflow automation: $6,500 to $15,000. One process, one integration path, human review where it matters.
• Document extraction for one document type: $9,000 to $25,000. The range is driven almost entirely by how messy and how varied your documents are.
• An AI phone or intake agent: $4,500 to $12,000 setup. Scripting, calendar integration, and testing dominate the cost, not the model.
• A custom internal application: $18,000 to $75,000 for a first production release. Wider range because scope varies enormously.
• A two-system integration: $4,000 to $12,000, depending on how cooperative the vendors are.

Run cost is smaller than you think

This is where expectations are most wrong. Most of what we build runs for $50 to $600 a month all-in, and a large share of that is ordinary hosting rather than AI.

For a sense of scale: extracting data from a typical business document costs somewhere between two and twenty cents per page depending on complexity. A business processing 500 invoices a month is looking at perhaps $30 in AI usage. The employee currently doing it costs considerably more than that per hour.

AI phone agents are the exception — telephony has real per-minute costs and voice models are more expensive than text. Budget $200 to $800 a month at meaningful call volume.

The arithmetic that decides it

Forget transformation. The calculation is straightforward, and you can do it on a napkin before you talk to anyone.

1. 01How many hours a week does this task consume, across everyone who touches it? Be honest, and count the interruptions.
2. 02What is the fully loaded hourly cost of the people doing it? Wage plus roughly 30% for taxes, benefits, and overhead.
3. 03Multiply, then multiply by 50 weeks. That is the annual cost of the status quo.
4. 04Assume automation removes 70% of it, not 100%. There is always a remainder.
5. 05Divide the build cost by that annual saving. Under 18 months is a good project. Under 12 is an obvious one.

Worked example: two people spend six hours a week each on invoice data entry. Twelve hours weekly at $28 fully loaded is $336 a week, or roughly $16,800 a year. Remove 70% and you save about $11,760 annually. A $12,000 build pays back in a bit over twelve months and keeps paying after that.

That is a good project. If the same calculation comes out at four years, do not do it — and be suspicious of anyone who tries to talk you into it anyway.

What makes a project expensive

• Document variety. Ten formats is not ten times harder than one, but it is not the same job either.
• Uncooperative vendor software. A system with no API can add weeks.
• Undefined process. If three people do the task three different ways, the first job is deciding which way is right — and that is your decision, not ours.
• Regulatory requirements. Healthcare and financial data mean more architecture and more testing, correctly.
• Real-time requirements. Nightly batch is dramatically cheaper than instantaneous, and is usually enough.

What makes it cheaper

• Starting with one process instead of five. Sequence the work; do not buy it all at once.
• Accepting a human review step. Full automation of the last 5% often costs more than the first 95%.
• Using the software you already have. Integration beats replacement almost every time.
• Being willing to change the process slightly to match how software works well.

If you want this arithmetic run against your actual numbers, the ROI calculator on this site does it in about a minute, and the automation finder will suggest what to point it at.