You’ve been told to “use AI” roughly four hundred times this year. Almost nobody saying it will tell you which AI, where in your business, what it costs, or what it’ll break. So here’s the honest list — the five things AI does reliably for an ordinary small business right now, and the three things that still fail — from someone who builds these systems for trades, clinics, and restaurants every week.
One framing first: the businesses winning with AI aren’t doing anything futuristic. They’re using it to stop leaking things they already earned — enquiries that went unanswered, quotes that never got followed up, reviews that never got asked for. AI’s superpower for a small business isn’t creativity. It’s never forgetting and never being too busy.
What works today
1. Instant enquiry response. When someone fills your contact form or messages your Facebook page, an AI-written reply goes out in seconds — acknowledging the specific thing they asked, asking the one follow-up question you’d ask, telling them when a human takes over. Under the hood it’s simple: the enquiry triggers an automation, a language model drafts the reply against rules you set (your tone, your services, what it’s not allowed to promise), and it sends. Why it pays: speed-to-first-response is one of the strongest predictors of winning the job — the customer who enquired with you also enquired with two competitors, and the first useful reply usually wins. Cost: typically tens of dollars a month to run once built.
2. Missed-call catch. You’re on the tools, phone rings out, and the caller — who had a job worth real money — dials the next name on the list. The fix: a missed call automatically triggers a text within seconds (“Sorry we missed you — how can we help?”), which keeps the conversation alive until you’re free. Mechanically it’s a webhook on your phone number plus an SMS — genuinely boring technology, which is exactly why it’s reliable. Most trades can attribute the cost back to a single caught job.
3. Review generation. Nobody remembers to ask for reviews, so the unhappy customers (who need no reminder) end up writing your Google profile. An automated system asks every completed job for a review at the right moment, with a link that takes ten seconds. The AI part is timing and wording; the compounding part is that your review count and recency are exactly what both Google and AI recommendation engines weigh when someone asks “who’s good near me” — we covered how that works in Does AI recommend your business?
4. Quote and document drafting. Feed a model your last fifty quotes and it learns how you price and how you write. New job comes in, you dictate the details in two minutes, and a draft quote in your format and your voice comes back for review. Same trick works for job summaries, follow-up emails, and site reports. The key word is draft — you check it, you send it. Done properly this turns a Sunday-evening chore into a ten-minute review. (When the pricing logic is complex enough, this graduates from an AI prompt into a proper custom tool — that’s its own article.)
5. Working content. Not “10 posts a day” spam — but the real, unglamorous content work: turning one job into a case-study post, answering the same ten customer questions on your website properly, keeping your Google Business Profile alive. AI collapses the effort from “a thing you never get to” to “an hour a month with someone driving it well.” The “driving it well” clause is load-bearing: unedited AI content reads like unedited AI content, and both customers and search engines have learned the smell.
What’s still hype (for a small business, today)
1. The fully autonomous employee. “An AI agent that runs your whole front office” demos beautifully and fails on the messy 20% of real conversations — the angry customer, the ambiguous booking, the edge case that needs judgement. Every system above works precisely because it has a narrow lane and a human handoff. Anything sold as “set and forget everything” is selling the demo.
2. AI answering your phone like a human. Voice AI is improving fast and has real uses (after-hours triage, overflow). But for a small business where every call might be a $5,000 job, a robotic mishandling costs more than the missed call it replaced. A text-back system is cheaper and — today — safer. Revisit in a year.
3. The AI-subscription pile. The most common failure mode we see isn’t a bad tool — it’s three good tools on the credit card that nobody wired into the actual workflow. AI that isn’t connected to your phone number, your enquiry form, your quoting process is just a subscription. The value was never the tool; it’s the implementation.
How to think about cost
The systems in the “works” list are not enterprise projects. Each is typically a fixed, one-time build — scoped in a conversation, priced before any money moves — plus a small monthly running cost (message fees, hosting, the model’s usage). The maths to run before buying anything: what does one leaked enquiry cost me? For most trades and clinics it’s hundreds to thousands of dollars. Systems that catch two or three a month pay for themselves out of margin you were already earning and quietly losing.
Where to start
Not with a tool. With the leak. Spend one week counting three numbers: calls you missed, enquiries you answered later than an hour, completed jobs you didn’t ask for a review. Those three numbers rank your first AI project better than any vendor pitch — and if you’d rather be systematic about the whole thing, we wrote the five questions to ask before spending a dollar.
Want the honest version for your specific business? Tell us what eats your week — we’ll come back with where AI genuinely helps, and where it doesn’t.