What AI automation actually costs a small business.

fig. 22.1 · what the market charges

The going rates, roughly

fig. 22.3 · what moves any quote

Four things that change the price

• +How many systems have to talk: connecting two tools is a week; connecting six tools with permissions and edge cases is a project.
• +The state of your data: clean customer records automate fast. A decade of inconsistent spreadsheets needs cleanup first, and that's real work.
• +What happens when it's wrong: a draft email that a human reviews is cheap. Anything touching money, safety or compliance needs guardrails, testing and rollback paths.
• +Who maintains it: one-off builds are cheaper up front; a retainer keeps someone accountable when your tools change under you (they will).

fig. 22.3½ · what the money buys, concretely

Real builds, real effort

Dollar figures without context are noise, so here's effort instead: real things I've shipped and roughly how long each took inside the retainer.

• +Self-serve booking for a trade company: about two weeks including the calendar rules, reminders, and the reschedule-yourself flow. It runs daily and hasn't needed me since.
• +A review pipeline: about a week for the follow-up system riding on jobs the crew already finishes. It took one Kansas City company from 34 reviews in five years to 80 in months.
• +An instant estimating tool: the biggest thing on this list: a scan-the-yard-to-priced-plan tool that cut estimates from hours to about 30 minutes. Multiple weeks, shipped in stages so it was useful the whole way.
• +Marketplace listing automation: one product record in, six live listings out across three Shopify stores, eBay, Amazon and Walmart, for a catalog 100,000 products deep. Weeks, not days, and worth every one.

The math that matters isn't the invoice, it's the hours. If an automation gives an owner back six hours a week, that's over 300 hours a year, most of a working quarter. At $1,500 a month, the question isn't whether the retainer is expensive. It's what your hours are worth, and whether a $50-a-day system that returns them is a good trade. For most owner-run businesses it embarrassingly is.

fig. 22.3¾ · take this to any vendor

Five questions to ask anyone quoting you

Including me. If a vendor squirms on any of these, the price doesn't matter.

• ?Who owns the code and the accounts when we're done? the only acceptable answer is you. Anything else is rent dressed up as a project.
• ?Who does the actual building? meet that person before you sign. If the person selling can't build and the person building was never on the call, you've found the markup.
• ?What happens when it breaks at 7pm? listen for monitoring, a rollback plan, and a named human. "Submit a ticket" is not a plan.
• ?Can I see a client's real numbers, with the client's name? case studies without names are creative writing. You should be able to call somebody.
• ?What would you NOT automate in my business? anyone who answers "nothing" is selling hours, not outcomes.

fig. 22.4 · the part nobody selling automation says

When you shouldn't automate

• –The task changes every single time it's done. Automation loves patterns; judgment calls stay human.
• –You do it less than once a month. The build costs more than the chore.
• –The real problem is the process, not the labor. Automating a broken process just makes the mess faster.
• –You'd be automating your only contact with customers. Some busywork is secretly your relationship.

If your situation is on that list, the audit will say so, and it will still be free. Telling a business not to hire me is the cheapest marketing I do.