AI for business: separating the hype from what works
Short answer Today's AI drafts, sorts, summarizes and answers routine questions well. It cannot guarantee a correct answer, take legal responsibility, or replace licensed judgment. In a Federal Reserve survey, only 31% of small-business owners reported higher sales from AI. Give it repeat tasks, keep a person checking, and measure hours saved before you pay more.
Want help applying this to your business? Talk it through on a free 30-minute call.
Book a callWhat can AI actually automate in a business today?
Work that repeats, follows a pattern, and that a person can check in a minute or two. The U.S. Small Business Administration's own list fits that shape: scheduling and reminders, sorting email, summarizing meetings, drafting content, answering common questions with a chatbot, routing calls and writing review replies. It also says to start small and to have another person review everything a free AI tool produces (SBA).
| Task | What AI does well | What stays with you |
|---|---|---|
| Email and inbox | Sorts messages, drafts replies | Anything about money, complaints or legal matters |
| Meetings and calls | Transcribes and summarizes | Deciding what to do next |
| Writing: posts, quotes, job ads | First drafts in seconds | Facts, prices, promises, final send |
| Customer questions | Answers hours, service area, booking steps | Refunds, exceptions, anything unusual |
| Bookkeeping and reports | Sorts transactions, explains numbers in plain words | Approving entries, filing, tax decisions |
| Ads and marketing | Writes variations, adjusts bids inside your limits | Budget, offer and which claims are true |
"AI agents" go a step further: they take a goal and decide their own steps inside your accounts. That is real, but it raises the stakes of a mistake, so the same rule holds: you set limits and approve risky actions. See agentic AI explained.
What can't AI do yet?
Four things, and vendors rarely say them out loud.
1. Be right every time. AI states wrong answers as confidently as right ones. The company behind ChatGPT says so itself:
"Hallucinations remain a fundamental challenge for all large language models, but we are working hard to further reduce them." (OpenAI, September 2025)
A hallucination is a made-up statement that sounds true. That is why anything a customer will rely on needs a human check.
2. Take the blame. If your AI tells a customer something wrong, you own it. In Canada, a British Columbia tribunal held Air Canada responsible for wrong refund information its website chatbot gave a customer and rejected the airline's argument that the chatbot was responsible for its own words. It ordered Air Canada to pay $812.02 in damages, interest and fees (Moffatt v. Air Canada, 2024). In the U.S., the FTC put it in one line:
"The FTC's enforcement actions make clear that there is no AI exemption from the laws on the books." (FTC, September 2024)
3. Replace licensed judgment. The FTC acted against DoNotPay, which sold "the world's first robot lawyer" but, the FTC alleged, had no testing to back that up and employed no attorneys. The settlement called for $193,000 (FTC, 2024). Legal, medical, tax and safety calls stay with a licensed person.
4. Know what you didn't tell it. AI doesn't know today's price list, this week's schedule or the promise you made a client unless it's connected to that data. Feed it only what you're allowed to: the SBA advises against putting sensitive or proprietary information into AI tools (SBA). More on where it falls short in AI marketing limits.
Does AI pay off for a small business?
Mostly in time saved, not yet in sales. The Federal Reserve's 2025 small business survey, analyzed by the New York Fed:
- 46% of firms with at least one employee used AI tools.
- "Only 31 percent of survey respondents reported increased sales from their firm's use of AI."
- 77% reported no change in labor costs from AI.
(New York Fed, October 2026)
Your own sense of "it's saving me time" can also be wrong. In an early-2025 study by the research group METR, 16 experienced software developers expected AI to make them 24% faster. They were actually 19% slower, and afterwards still believed AI had sped them up by 20% (METR, July 2025). METR has since said that result is dated, as the tools changed. The lesson holds: time the task before and after.
On jobs, the picture is change more than collapse so far. The U.S. Bureau of Labor Statistics projects computer and IT jobs to grow faster than average from 2025 to 2035, with about 280,000 openings a year (BLS). For a small business, the practical question is narrower: which hours of which person's week can AI take.
How do you spot AI hype?
Look for a promise of results without the work, or "AI" with no detail. Regulators have already punished each of these.
| Red flag | What happened |
|---|---|
| Guaranteed income or "passive" profits from AI | The FTC sued several AI business-opportunity schemes; one promised "five-figure monthly income by the second year" and, the FTC says, took at least $25 million from consumers (FTC, 2024) |
| "Uses AI" with no specifics | The SEC fined two investment advisers $400,000 in total for false claims about using AI, which its chair called "AI washing" (SEC, March 2024) |
| AI that replaces a professional | DoNotPay's "robot lawyer" (above) |
| "Set it and forget it" | No vendor can take your legal responsibility (Air Canada, above) |
Before you buy, ask the vendor:
- Which tasks does it do on its own, and which need my approval?
- What data does it read, where is it stored, and is it used to train models?
- How do I see what it did, and undo it?
- What does setup take, in my hours, before it saves any?
- What results can you show from a business my size, and how were they measured?
A vendor that can't answer plainly is selling the word "AI", not a working tool.
How to test AI without falling for the hype
- Pick one task that eats at least 2 hours a week and repeats.
- Write down the baseline: hours a week, and errors or missed leads a month.
- Run a free or cheap version for 30 days with a person checking every output. See human oversight in AI marketing for how to set that up.
- Keep it only if a number moved. Then, and only then, pay for more.
Example
A cleaning company (US dollars)
Uses a free AI assistant to draft quotes and reply to review requests. Timed before and after: 5 hours a week down to 2. Leads are unchanged. It keeps the free tool and skips a $400-a-month "AI growth" package that promised to double bookings.
An accounting firm in Ontario (Canadian dollars)
Adds a website chatbot for after-hours questions at $100 a month. It limits the bot to hours, services and booking, with a line sending tax questions to a person, since the firm stays responsible for anything the bot says.
A contractor
A vendor pitches "fully autonomous AI that runs your business." Asked the five questions, it can't say what data the tool reads or how to undo its actions. The contractor passes.
FAQ
Is AI just hype?
No. It reliably saves time on drafting, sorting and summarizing. The hype is in promises of sales, income or hands-off running that no tool can guarantee.
Will AI replace my staff?
Not so far for most small firms: 77% in the Fed survey saw no change in labor costs from AI. It takes over tasks, not whole jobs.
Can I trust what AI tells me?
Check anything that matters. OpenAI says hallucinations remain a challenge for all language models, and you're responsible for what your AI tells customers.
What's the difference between generative AI and an AI agent?
Generative AI creates text or images when asked. An agent takes a goal and acts in your tools on its own, which saves more work but needs tighter limits.
Different for you?
- You want to know which marketing uses pay off: AI marketing for small business.
- You're weighing software that acts on its own: agentic AI explained.
- You handle patient or health data: is ChatGPT HIPAA compliant?
- You're missing calls and leads: Conversion.
- You want to show up in Google and AI search: SEO & GEO.