Should My Business Use an AI Chatbot?

Should your business use an AI chatbot? Yes, if you get more than roughly twenty repetitive questions a week and you already have written answers to them. No, if your inquiry volume is low, your questions require judgment, or nobody has ever written the answers down. A chatbot is a delivery mechanism for information you already have. If you do not have the information, all you have built is a faster way to disappoint people.

This guide gives you the volume test, real costs, a setup sequence that avoids the classic embarrassments, a worked example for a Lake Geneva restaurant, and the disclosure question everybody asks last and should ask first.

What Does an AI Chatbot Actually Do Well?

It answers the same twenty questions perfectly at eleven at night, captures the contact information of someone who would otherwise have closed the tab, and routes the rest to a human without making the customer repeat themselves. An AI chatbot here means a chat window on your website or messaging channel that uses a language model to answer from your own documents, as opposed to the old menu-tree bots that made you pick from four options and then hung up on you.

Where it genuinely earns its place:

• Hours, location, parking, policies. The questions that are not interesting but are asked constantly, and that cost your staff real minutes.
• After-hours capture. A message at 9:40 p.m. that would have gone unanswered until Tuesday now gets an answer and a booking link.
• Pre-qualifying. Asking the three questions your estimator always asks anyway, so the call starts halfway through.
• Wayfinding on a big site. Pointing someone to the right page faster than your menu does.
• Multi-language basics. Answering a simple question in the customer's language without you hiring for it.

Where it does not earn its place: pricing that depends on a site visit, anything emotional or a complaint, medical or legal questions, and any conversation where being wrong costs you real money. Those need a person, and the bot's job is to get them to that person quickly.

For context on the wider shift, McKinsey's State of AI global survey, fielded across 105 nations in mid-2025, found 88 percent of organizations regularly use AI in at least one business function, and Intuit QuickBooks' April 2025 survey of more than 2,200 US businesses with up to 100 employees found 68 percent use AI regularly. Customer-facing chat is one of the most visible versions of that, which is exactly why it is worth doing carefully rather than quickly.

Should My Business Have One? Run This Test First

Count two things for one week: how many inbound questions you get, and how many of them have the same answer every time. If the second number is above twenty a week, a chatbot will pay for itself. If it is under ten, put your effort into a better contact page and a real FAQ instead.

Four Signs Yes

1. Your staff answers the same handful of questions all day, and you can list them from memory.
2. You get meaningful traffic outside business hours, which for seasonal Lake Geneva businesses often means evenings and Sunday nights.
3. You already have written answers somewhere: an FAQ page, a house manual, a policy document, a menu, a service list.
4. Missed inquiries actually cost you. A booking, a table, a quote request.

Four Signs No, or Not Yet

1. Your answer to most questions is "it depends, let me look." That is judgment, and judgment does not compress into a widget.
2. You have no documented answers. Write the FAQ first; you will get most of the benefit before you buy anything.
3. Your inquiry volume is a handful a week. A prominent phone number and a fast reply will beat a bot.
4. Your work is regulated or high-stakes enough that a wrong answer creates liability. See our post on putting customer data into AI tools before you point any bot at customer records.

What Does an AI Chatbot Cost?

There are three price models, and they behave very differently at volume. Approximate pricing as of March 2026, check current pricing.

• Priced per resolution. Intercom's Fin charges $0.99 per outcome, meaning per resolved conversation, on top of its per-seat support platform pricing. This model is honest and easy to forecast, and it is the most expensive per conversation at high volume. Two hundred resolutions a month is about $198 plus seats.
• Priced per conversation bucket. Most website chat widget vendors sell an AI add-on with a monthly cap on AI conversations, priced in tiers. Check each vendor's current pricing directly, because this segment changes prices often and the caps matter more than the headline number.
• Built on a general AI subscription or API. A custom assistant loaded with your documents, either inside a business plan (roughly $20 to $30 per user per month for ChatGPT Business or Claude Team) or built against the API and wired to your site. Running cost on a low-traffic site is small; the money goes into the build and the maintenance.

Add the plumbing. If the bot needs to create a lead in your CRM, send a text, or notify a person, that runs through an automation platform, and paid tiers there start around $12 to $20 a month. Our comparison of Zapier, Make, and n8n covers how those bills scale.

The cost people forget is the human one. Somebody has to read the transcripts every week for the first two months, fix the wrong answers, and add the missing ones. Budget an hour a week for that and it works. Budget zero and it drifts.

How Do I Set One Up Without Embarrassing Myself?

Build it narrow, test it against real transcripts, and give it one job it can do perfectly rather than five it can do badly. The sequence matters more than the tool.

1. Write the knowledge base. Your twenty questions with real answers, plus hours, address, parking, policies, service list, and anything seasonal. Keep it in one document you can update, not scattered across your site.
2. Define the fence. Write, in the bot's instructions, what it must never do: quote a price it was not given, promise a delivery date, discuss another customer, or answer a medical, legal, or financial question. Tell it what to say instead.
3. Build the escalation path first, not last. Decide exactly what happens when the bot cannot answer. Where does the message go, who owns it, and what is the promised response time? A bot with a dead end is worse than no bot.
4. Set the tone with examples. Paste three replies you actually wrote and tell it to match that voice. Descriptions of tone work far less well than samples.
5. Test with real history. Take thirty genuine messages from the last two months, strip the names, and run them through. Score each answer as correct, incomplete, or wrong. Do not launch above a couple of wrongs.
6. Add the honest limits. Say up front that it is an assistant, that it can make mistakes, and how to reach a person. That sentence prevents most complaints.
7. Launch on one page, not the whole site. Put it on the page where the questions actually come from, watch a week of transcripts, then expand.

The National Institute of Standards and Technology's AI Risk Management Framework, released in January 2023, sorts this kind of work into govern, map, measure, and manage. In small-business terms: one person owns it, you know what it touches, you read the transcripts, and you have a plan for when it goes wrong.

What Does This Look Like for a Lake Geneva Business?

Picture a supper club just outside Lake Geneva with a bar, a dining room, and a Friday fish fry that fills the parking lot. In season they field a steady stream of messages through the website and social inbox, and the same questions dominate: are you open Mondays, do you take reservations for six, is there a gluten-free option, where do we park, do you have a dock.

What they built:

• A knowledge base of 24 questions, written by the manager in about ninety minutes, including the awkward ones (no, the patio is not covered; yes, the fish fry sells out by seven).
• A chat widget on the website and the same assistant answering first-touch messages in the social inbox, with an explicit line at the start: it is an assistant, it can get things wrong, and here is the phone number.
• A hard fence: it never confirms a reservation. It answers the question, then hands off to the reservation system or the phone. Confirming a table it cannot see is the fastest way to ruin a Saturday.
• An escalation rule: anything containing the words allergy, complaint, refund, large party, or private event goes straight to the manager's phone with the transcript attached.
• A weekly review: the manager reads every transcript on Tuesday morning, adds new answers, and fixes wrong ones. That took an hour a week at first and about fifteen minutes by month three.

The results worth watching were not "conversations handled." They were how many after-hours messages got a useful answer, how many turned into a reservation, and how many escalations arrived with enough context that the manager did not have to ask the customer to repeat themselves. Those are the numbers to put on the scorecard described in our post on measuring the ROI of AI and automation.

What went wrong in month one: the bot cheerfully told someone the kitchen closed at nine on a night it closed at eight thirty for a private event, because the knowledge base said nine and nobody had updated it. That is the real failure mode. Not a rogue AI, just a stale document. The fix was adding "check the bot's hours" to the same checklist that already covered updating the sign.

Do I Have to Tell People They Are Talking to a Bot?

Label it. There is no single federal rule that makes every small business disclose a chatbot, but the Federal Trade Commission's authority over deceptive practices applies to how you present it, and some states have their own bot disclosure requirements. More practically, disclosure lowers expectations, which lowers complaints.

The FTC announced an enforcement sweep called Operation AI Comply on September 25, 2024, targeting both deceptive claims about AI products and deceptive conduct carried out using AI, including a service that generated fake reviews and a company marketing an "AI Lawyer." The lesson for a small business with a chat widget is narrow and clear: do not claim the bot does more than it does, do not let it make promises you will not honor, and do not present machine-written content as a human's.

A disclosure that works reads like this: "Hi, I am the assistant for [business]. I can answer questions about hours, menu, and parking. I get things wrong sometimes, so for anything important, call us at [number]." That is one sentence of honesty, one sentence of scope, and an exit. It performs better than pretending to be a person, because customers can tell anyway.

What Goes Wrong With AI Chatbots?

The failures are predictable, which means they are preventable.

• Confident wrong answers. The bot states a price, a policy, or an hour that is not true. Prevention: a narrow knowledge base, explicit instructions to say it does not know, and a weekly transcript review.
• Stale information. Seasonal hours, a closed date, a discontinued service. Prevention: put the bot's knowledge base on the same update checklist as your Google Business Profile and your door sign.
• Dead-end escalation. The customer asks for a human and nothing happens. Prevention: test the handoff yourself, from a phone, on a weekend.
• The trap loop. The bot keeps asking clarifying questions instead of giving up. Prevention: cap it at two clarifying questions, then hand off.
• Scope creep. Somebody points it at the customer database so it can look up orders. Now it is holding personal information, and that is a different risk conversation entirely.
• Nobody reads the transcripts. The most common failure of all, and the only one that guarantees the others persist.

Do This This Week

1. Count one week of inbound questions and mark which ones have the same answer every time. That count is your decision.
2. Write the top twenty questions and their real answers into one document. Fix anything on your website that the exercise reveals is wrong.
3. Decide the fence: three things the bot must never do, and what it should say instead.
4. Design the escalation path before you build anything, including who owns it and the promised response time.
5. Pull thirty real messages from the last two months, remove names, and use them as your test set.
6. Write your disclosure line: what it is, what it can help with, and how to reach a person.
7. Launch on one page only, and put a recurring 30-minute transcript review on the calendar for the next eight weeks.
8. Add the bot's knowledge base to whatever checklist you already use when hours or services change.

Where to Go From Here

The decision is not really about AI. It is about whether you have twenty repeatable questions and written answers to them. If you do, a chatbot is one of the fastest, cheapest wins available to a small business, especially one with evening and weekend traffic. If you do not, writing the answers is the actual project, and you can decide about the widget afterward.

If you would like help deciding where a chat assistant fits alongside everything else in your operation, our AI automation service starts with the processes rather than the tools, and the AI Opportunity Audit ranks the options by payback. If people cannot find your business yet, the free Local Visibility Audit on our homepage is the better first move. A chatbot answering questions nobody is asking is an expensive quiet room.

Sources and Further Reading

1. Fin AI Agent Pricing. Intercom, March 2026.
2. Pricing. OpenAI, March 2026.
3. Plans and Pricing. Anthropic, March 2026.
4. Plans and Pricing. Zapier, March 2026.
5. FTC Announces Crackdown on Deceptive AI Claims and Schemes. Federal Trade Commission, September 2024.
6. AI Risk Management Framework. National Institute of Standards and Technology, January 2023.
7. The State of AI: Global Survey 2025. McKinsey and Company, 2025.
8. Survey Reveals Small Businesses Are Using AI to Boost Productivity. Intuit QuickBooks, June 2025.