ai for small business
AI that earns its keep
Three automations I keep building for small businesses (inbox triage, content repurposing, invoice chasing) and the guardrail that makes them safe.
· Robert Csaszar
Most AI advice aimed at small businesses is written for companies with a data team. You do not have a data team. You have an inbox, a backlog of half-written posts, and four unpaid invoices you keep meaning to chase. That is where the useful work is, and none of it requires a “transformation”.
Here are the three automations I end up building most often, in the order I would build them.
inbox triage
A shared inbox at a small firm carries everything: quote requests, supplier spam, complaints, the occasional actual emergency. The owner reads all of it, usually at 9pm.
The automation is modest. Every incoming message gets classified (quote request, support, billing, junk), labelled, and routed. Quote requests additionally get a drafted reply with the standard questions your team asks anyway: location, timeline, rough scope. A person reviews the draft and sends it.
The sorting matters more than the drafting. Quote requests answered the same day close at a visibly better rate than ones answered on Thursday, and the same-day reply is what the triage buys you. Setup is an afternoon with any of the current model APIs and whatever your email provider calls rules.
content repurposing
You wrote one decent piece this month: a project write-up, a seasonal notice, an answer to a question customers keep asking. That piece should become a newsletter section, two or three social posts, and an updated FAQ entry. It usually becomes nothing, because reformatting is boring.
This is the safest possible use of a language model, because the facts are already yours. The model is not asked to know anything, only to reshape what you wrote. I set up a single prompt per output channel, tuned once to the client’s voice, and the owner pastes in the source piece. Ten minutes of editing replaces the two hours that meant it never happened.
Do not invert this. Generating “content” from nothing produces the beige sludge you already scroll past, and your customers scroll past it too.
invoice chasing
Chasing money is the task owners hate most, so it gets done late or not at all. Yet it is nearly clerical: an invoice is 7, 14, or 30 days overdue, and each stage has a tone. The first nudge is friendly, the second is firm, the third mentions next steps.
The automation watches the ledger, drafts the right message at the right stage with the invoice details filled in, and queues it for a human to send. Drafting-not-sending sounds like a technicality. It is the difference between a reminder and an incident, because the system does not know that this particular client’s warehouse flooded last week. The owner does.
the guardrail
One rule covers all three: automate the paperwork, never the relationship. Anything that touches a customer gets a human glance before it leaves the building. The glance takes seconds. Skipping it is how you end up apologising for a robot.
Start with whichever of the three costs you the most sleep. Each one is an afternoon of setup and a week of tuning, and each pays for itself in the first month or it was the wrong pick.