Most "AI for small business" advice is written by people who have never run one. Here's the honest version: what actually pays back the setup time, and what's marketing hype dressed as productivity advice.
Worth your time
| Workflow | Why it works | Typical time saved |
|---|---|---|
| First-draft customer emails | Replies, follow-ups, refund explanations: you edit and send | 60-80% of typing time |
| Quote and proposal drafts | Feed in scope, get a structured proposal back | 1-2 hours per proposal |
| Review and testimonial summaries | Paste 50 reviews, get the three themes customers care about | Hours of manual reading |
| Job descriptions and SOPs | Boring documents that have to exist; AI is genuinely good at boring documents | 30-60 minutes per doc |
| Bookkeeping question lookups | "What category does this expense go in?" with your country's rules in context | Avoids a bookkeeper call |
Not worth your time (yet)
- Fully autonomous customer service bots. They hallucinate refunds and promises you didn't make. Use AI to draft replies a human approves before sending: the drafting is the valuable part, the autonomy is the risky part.
- Logo generation from a one-liner. You'll spend more time iterating through generic-looking options than it would take to hire a designer for $200 who asks you three good questions first.
- "AI-powered" CRM features that just rename existing fields. Don't pay extra for "AI" branding on a feature your CRM already had under a different name.
The rule of thumb
If a task is drafting from your inputs, AI is excellent. If a task is making decisions on your behalf with money or reputation at stake, keep a human in the loop. That's the whole framework: and it's a better filter than any vendor's feature list.
A worked example: the quote-drafting workflow
Here's what this actually looks like in practice for a small contracting business:
- Input: "Client wants a bathroom remodel: new tile, vanity, fixtures. Approx 60 sq ft. Their budget mentioned was $8-12k."
- Prompt: "Act as a contractor's estimator. Draft a proposal outline for this scope: [input]. Include sections for scope of work, timeline estimate, and a note about the budget range with a caveat that final pricing depends on material selection. Format as a client-facing document, professional but warm tone."
- Output: A structured draft you review, adjust with your actual pricing, and send: instead of starting from a blank page or an old template that doesn't quite fit.
The AI didn't quote the job: you did. It removed the blank-page problem, which is where most of the time actually goes.
Why the "not worth it" list matters just as much
Small business owners get burned not by AI failing at drafting tasks, but by trusting it with decisions it shouldn't make unsupervised. A customer service bot that promises a refund your policy doesn't allow creates a real liability. The pattern to watch for: any tool that lets AI act on your behalf without a review step before something goes out the door or money moves.
How to start if you haven't yet
Pick one workflow from the "worth your time" table: customer emails is usually the easiest entry point since you already have real examples to work from. Run it for two weeks before adding a second workflow. Small businesses that try to automate five things at once usually abandon all five; the ones that stick with it start with one and expand once it's actually saving time, not just theoretically saving time.
For a fuller starting point, the AI for Small Business Playbook walks through this same framework against your specific business.
Setting expectations with your team
If you're introducing any of these workflows to employees, the framing matters. "AI drafts it, you approve it" lands very differently than "AI is replacing this part of your job": and the first framing is also the accurate one for every workflow on the "worth your time" list. None of them remove the person from the loop; they remove the blank page.
A simple two-week trial structure
Week one: use the new workflow for every instance of the task, but don't change anything else about your process: same review, same send. Week two: track how much time you actually saved compared to your old process, honestly, including time spent editing AI drafts. If the number is meaningfully positive, keep it. If it's marginal, the workflow probably doesn't fit your specific business even if it works well for others: that's fine, not every tool fits every business.
The one question that filters most bad AI purchases
Before buying any "AI-powered" tool for your business, ask: "what decision is this making, and what happens if it's wrong?" If the answer is "it drafts something I review," you're safe. If the answer involves money moving or a promise being made without your review, slow down regardless of how impressive the demo looked.