What AI actually automates well in a small business (and what it doesn't)

Most owners have been sold AI twice already: once as magic, once as a chatbot bolted onto a website. Neither matched the work they actually do at 9pm.

Here is the line we use.

Automates cleanly today

• Structured chasing. Follow-ups on quotes and invoices, on a schedule, with an escalation path.
• Moving data between systems. Intake to CRM to invoice to books, with validation on the way through.
• Drafting from a template plus context. Quotes, job summaries, appointment confirmations, standard replies.
• Triage. Reading an inbox and sorting it into "needs you," "needs a one-liner," and "noise."
• Summarizing your own numbers. A weekly digest built from systems you already pay for.

Still needs a human

• Pricing judgment on unusual work. The machine can draft it; you approve anything outside the normal band.
• Anything with a relationship attached. Bad news, disputes, renegotiations.
• Exceptions you haven't seen before. By definition there is no rule to follow yet.
• Deciding what to build next. That's an operations question, not a model question.

The reliability part is the actual product

A demo that works once is easy. A system that runs unattended for eight months is a different discipline: it needs monitoring, retries, sane failure behavior, and someone who notices when an upstream tool changes its API.

The AI part matters less than you'd think. Knowing which fifteen minutes of the day shouldn't exist matters more.

That's why every system we build is documented in plain English, monitored, and yours to keep — and why the monthly plan exists at all. Software that nobody watches quietly stops working, and you find out from a customer.