AI for small business is worth using for drafting, summarizing and routine replies that a person checks, and worth skipping wherever a wrong answer becomes a promise to a customer. This guide is for owners of plumbing, HVAC, electrical, roofing and similar service businesses in the Twin Cities East Metro, the St. Croix Valley and western Wisconsin. It answers one question: should this task go to AI at all?
The tool comparisons, prompt libraries and automation lists already live in our guides to AI tools for small business, ChatGPT routines and the AI automation playbook. This one sits above them. It sorts tasks into use, skip and automate, names the risks, and gives an order for adopting it, with each outside claim linked to the page we read in November 2026.
AI for small business in the Twin Cities: where adoption stands
The Census Bureau asks about this directly. Its Business Trends and Outlook Survey asked whether a business used AI in any of its business functions in the last two weeks. For the collection period October 19 to November 1, 2026, the survey's published estimate was 27.7% of nonfarm employer businesses in Minnesota, with a standard error of 2.48, and 24.0% in Wisconsin, with a standard error of 2.64. For the Minneapolis-St. Paul-Bloomington, MN-WI metropolitan area, which includes western Wisconsin counties, the estimate was 33.2%, with a standard error of 2.87. We read those figures from the survey's public API at census.gov/hfp/btos/api for the 10/19/2026 to 11/1/2026 period.
Three cautions apply. The estimates cover nonfarm employer businesses of every kind and size, so they do not describe service trades specifically. The Census Bureau's About the Data page says the question was reworded in November 2025 and the instrument was updated again beginning with the October 19, 2026 collection, so compare only figures collected under the same wording. And that questionnaire now names the technology "Super Intelligence or SI (formerly referred to as Artificial Intelligence or AI)," which is why you may see "SI" on the Census pages. We say AI throughout.
The takeaway is modest. A sizable share of businesses in the metro report some AI use, and more say they did not, so you are not behind for starting small.
What to use: drafting, summarizing and routine questions
The jobs where AI helps a service business share one feature: a person can check the output in seconds against something real.
- Drafting replies. A first draft of a customer email, a follow-up after an estimate, or a response to a review. You read it, fix it and send it.
- Summarizing. A long email thread, a voicemail, or a technician's rough job notes turned into a clean summary. You compare names, addresses and amounts against the source.
- Answering routine questions. Hours, service area, how to book and what to prepare for a visit, answered from facts you wrote, with a handoff to a person for anything else.
Our own practice follows the same pattern. On our brand and copy page, we say we "use AI the way we use any tool in the stack, for drafting and research speed, and never as the final word." That sentence is our standard for client work, and it is a reasonable standard for yours.
What to skip: prices, licensed advice and final commitments
These calls are ours, and the reasoning is plain. A wrong answer in these areas costs you money or your license, and a tool carries neither.
- Prices and estimates with no person involved. A chatbot that quotes a water heater replacement price from a general guess creates a number the customer may expect you to honor.
- Technical or licensed advice. Whether a gas line is safe or an electrical panel needs replacing is a call for the person who holds the license.
- Final commitments. Arrival windows, refunds, warranty promises and anything else that binds the company should come from a person.
- Anything the owner has not checked. If nobody read it, it is not ready to send or publish.
Here is an illustrative example, with an invented business. Hollyhock Hollow Plumbing adds a chat widget that answers "how much to replace a water heater?" with a flat number. The number ignores the home's venting, the tank size and the access, so the technician arrives to a customer who expects the chat price. A better setup has the widget collect the address, the tank size and a photo, then hand the thread to an estimator with the details attached.
A decision table: use, skip or automate
This table sorts common service business tasks. "Automate" here means the system acts on its own with a message you wrote and approved, or with a person ready to take over.
Scroll to see all columns
| Task | Verdict | Why |
|---|---|---|
| Draft a reply to an email or review | Use | A person reads and sends it, so errors get caught |
| Summarize a call, voicemail or job notes | Use | Quick to check against the original |
| Answer hours, service area and booking questions | Automate with handoff | The facts come from you, and anything else goes to a person |
| Text back after a missed call | Automate, after checking the texting rules | A fixed message you wrote can do the job, but business texting has carrier registration and opt-out rules, so confirm yours with your attorney before you turn it on |
| Quote a price or estimate | Skip without a person | A wrong number becomes a promise |
| Give technical or licensed advice | Skip | The license and the liability belong to you |
| Promise schedules, refunds or warranties | Skip | These bind the company |
| Write reviews or testimonials that pretend to come from a customer | Skip | The FTC's review rule covers AI-generated fake reviews |
| Publish unedited blog posts or service pages | Skip | Google's guidance asks for fact-checking before publishing |
The missed-call text and the booking and review workflows are the part we build. Our CRM and automation service is described on our site as GoHighLevel built around how you work, covering speed to lead, missed-call text back, review generation, pipelines, booking and getting paid. We build on GoHighLevel, so we have an interest in that part of the advice.
The risks: wrong answers, customer data and disclosure
Wrong answers
Google's guidance on generative AI content, last updated October 1, 2026 according to the page, says generative models predict a likely sequence of words and do not retrieve facts, so outputs may contain inaccuracies. It calls it critical to manually fact-check and review AI-generated content before publishing, including titles, meta descriptions and image alt text. The same page says using AI to generate many pages without adding value for users may violate Google's spam policy on scaled content abuse.
Customer data
The plan you use decides what happens to what you paste in. OpenAI's help page on model improvement says that for individual plans (Free, Plus, Pro) it may use your content to train its models and you can opt out, while by default it does not use inputs or outputs from ChatGPT Business, ChatGPT Enterprise, ChatGPT Edu or its API for that. Anthropic's privacy page, dated August 18, 2026, says that by default it will not use inputs or outputs from its commercial products to train its models, with exceptions for feedback you submit or data you choose to share. Both pages are about their own plans, so read the page for the one you pay for. Our AI tools guide, linked at the top, compares the terms plan by plan.
Minnesota's consumer data privacy law has its own scope and a small business exclusion, which the same guide summarizes. This is general information and not legal advice, so ask your attorney how it applies to you.
Disclosure and fake reviews
The FTC's rule on consumer reviews, 16 CFR part 465, took effect October 21, 2024, and the Commission's Federal Register notice says the final rule covers reviews that are AI-generated. Have AI draft a reply to a real review if you like, and never have it invent a customer. Confirm any question about the rule with your attorney.
The order to adopt it
- Pick one task that costs you an hour or more a week and has an output you can check, such as review replies or follow-up emails.
- Choose the plan before the tool. Use a business plan for anything that touches customer names, addresses or job details, and read its data page first.
- Write three rules. What may be pasted in, who checks the output, and what the tool may never send on its own.
- Run it by hand for a month. Count the time saved and the mistakes you caught.
- Automate only the fixed parts. A missed-call text or a booking link is a message you wrote once, and it does not need a model to improvise.
- Expand when the numbers show a gain. Add the next task only after the first has paid back its setup time.
If you want help deciding which of your own tasks belong in each column, contact us.