Is an AI Chatbot Worth It for a Small Business Website? Costs, Risks and a 30-Day Test

Is an AI chatbot worth it for a small business website?

A cleaning company owner checks the contact form on Monday morning and finds four quote requests from Saturday night. By the time anyone calls back, two have already booked with someone else. Somebody suggests putting an AI chatbot on the website so those people get an answer at 10 pm, and the owner wonders whether it is a real fix or another $50 a month subscription nobody looks at.

Both outcomes happen. A website chatbot pays for itself when your site gets enough visitors, those visitors ask the same questions, and the bot can do something useful with the answer, such as capture a lead or book a slot. It is a waste when traffic is thin, every job needs a custom quote, or nobody keeps the bot's information current. This guide helps you work out which side you are on before you pay for anything.

What a website chatbot actually does for a small business

Strip away the vendor demos and a small business chatbot does three jobs. It answers repeat questions: hours, service area, whether you do move-out cleans, whether you take insurance, how long a job takes. It captures leads: name, phone, what they need, when, so someone can follow up in the morning. And, if you connect it to a calendar, it books.

The first job saves staff time. The second and third make money, and they are the ones to judge it on. A bot that answers questions but never collects a phone number is a nicer FAQ page. A bot that turns a Saturday night visitor into a booked estimate is a salesperson who works weekends.

What it should not do is negotiate, quote binding prices, or promise anything your policies do not allow. More on why below.

When it pays and when it does not

It tends to pay when most of these are true. Your site gets at least a couple of thousand visitors a month, because only a small share ever open a chat. A meaningful share of your leads arrive outside business hours. Your inbox and phone are full of the same handful of questions. And the next step after a question is simple: book a call, book an estimate, or leave details.

It tends not to pay when your site gets a few hundred visitors a month, when every enquiry needs a site visit and a human conversation anyway, or when you already answer every form within minutes during the hours your customers are looking.

Volume is the number most owners skip. Usage rates vary widely and the public figures are vendor data. Tidio reports that about 15 percent of visitors who see its chat icon start a conversation, across roughly 300,000 sites. An older 2015 benchmark from Boldchat put it at 1.6 percent of all visitors. Those measure different things, so treat them as a range rather than a forecast. At 3,000 visitors a month, that range is somewhere between about 50 and a few hundred chats. At 300 visitors, it may be a handful.

Customer attitude matters too. A Gartner survey published in July 2024 found that 64 percent of customers would prefer companies did not use AI for customer service, and the top worry was that it makes it harder to reach a person. That does not mean skip it. It means the bot must make it easy to reach a human, and must not pretend to be one.

The risk nobody mentions in the demo

A chatbot on your website speaks for your business. In Moffatt v. Air Canada, a 2024 decision by British Columbia's Civil Resolution Tribunal, Air Canada's website chatbot told a customer he could claim a bereavement fare after booking, which was not the airline's policy. Air Canada argued the chatbot was responsible for its own actions. The tribunal rejected that and ordered the airline to pay the difference plus interest and fees. It was a small claims ruling, not a binding precedent, but the lesson travels: what the bot says, customers will hold you to.

The second risk is people messing with it. In December 2023, a user talked a Chevrolet dealership's ChatGPT-powered website bot into agreeing to sell a 2024 Tahoe for $1 and calling it a legally binding offer. The dealership did not honor it and pulled the bot, but the screenshots went everywhere.

Both problems have the same fix: a bot that answers from a fixed knowledge base, refuses to discuss prices beyond what you publish, refuses to make commitments, and hands anything unusual to a person.

What the options cost

Prices below are as of October 2026, from each vendor's pricing page.

Tidio is the cheapest entry point. Paid plans start at $24.17 a month for 100 conversations, and the free plan covers 50 conversations. Its AI agent, Lyro, comes with 50 free conversations once, not every month; after that the Lyro add-on starts at $32.50 a month for 50 AI conversations, with higher tiers. Tidio also gives you a normal live chat inbox, which matters if a person will sometimes take over.

Chatbase is built for a question-answering bot trained on your website and documents. There is no free plan, only a trial. Hobby is $32 a month billed yearly for 700 message credits and one AI agent; Basic is $120 a month billed yearly for 4,000 credits.

Intercom Fin charges $0.99 per outcome, Intercom's term for a conversation the AI resolved. You can use it inside Intercom, where Essential seats start at $29 a month (less on annual billing), or standalone on top of another help desk with a monthly minimum. It makes most sense for businesses with real support volume, such as an online store or a software product, rather than a local service company.

GoHighLevel puts Conversation AI in its website chat widget. If you already run your CRM on HighLevel, the AI Employee Growth add-on is $50 a month per sub-account for 1,000 Conversation AI responses, and Unlimited is $97, subject to fair use. Without a plan, it bills by AI model tokens. The advantage is that the bot books straight into the same calendar and contact records your team already uses.

Paying someone to build it is the other route. Builds documented on this site give a sense of agency pricing: one builder sells GoHighLevel chatbot websites to local businesses for $1,500 to $3,000 each plus $200 to $500 a month (builtwithagents.ai/strategy/ai-chatbot-websites-local-businesses-gohighlevel), and another prices a Voiceflow customer service bot with knowledge base, lead capture and booking at $3,500 plus a maintenance retainer (builtwithagents.ai/strategy/live-build-3500-customer-service-chatbot). Both figures are self-reported by the builders and include no client results.

A worked example: a two-crew cleaning company

Take a residential cleaning company with two crews, 3,000 website visitors a month, and a quote form. These numbers are illustrative. The owner pulls the form submissions for the last three months and finds that about 40 percent came in after 6 pm or on weekends, and that the office answers them the next business day.

The owner tests Chatbase Hobby at $32 a month. The knowledge base is one page: services, service area by zip code, the published starting prices for standard and deep cleans, the cancellation policy, whether crews bring supplies, pet policy, and how to book. The bot has one goal: get a name, phone number, address zip, home size and preferred day, then tell the visitor the office will text a firm quote by 9 am.

A visitor at 10:40 pm asks whether they do move-out cleans in a certain zip code and how much one costs. The bot confirms the zip is covered, says move-out cleans start at the published price and depend on home size and condition, asks for the details, and confirms someone will text by 9 am. It does not quote a final price. When a visitor asks whether the crew can also clean mold off a bathroom ceiling, a request outside the knowledge base, the bot says it will pass the question to the office and collects contact details.

At the end of 30 days the owner counts. Say the bot handled 90 chats and collected 22 leads, 14 of them after hours, and 6 became booked cleans. If an average first clean is worth a few hundred dollars and some turn into recurring customers, $32 is easily covered. If the count had been 2 leads and no bookings, the bot comes off the site. Either way, the decision is made on the business's own numbers.

How to test one in 30 days

Step 1: Check your traffic and timing first

Open Google Analytics 4 and look at monthly users and the hours they visit. Then pull your last three months of form submissions and calls and note how many arrived after hours. If you have fewer than about 1,000 visitors a month and almost no after-hours enquiries, stop here: put your FAQ on the site and make the form shorter instead.

Step 2: Collect the 20 questions people actually ask

Go through recent emails, form messages, texts and call notes, and write down every question with the answer your best employee would give. This becomes the knowledge base. Twenty good answers beat a bot pointed at your whole website, which will happily repeat an old price from a forgotten page.

Step 3: Pick the tool by what happens after the chat

If you only need questions answered and leads captured, start with Chatbase Hobby or Tidio with the Lyro add-on. If HighLevel is already your CRM, use its Conversation AI so leads land in your pipeline and bookings in your calendar. If you run a support desk with many tickets, look at Intercom Fin. Use the free trial before paying for a year.

Step 4: Write the rules before you switch it on

In the bot's instructions, write the hard rules plainly: answer only from the knowledge base; never quote a price other than the published starting prices; never promise availability, refunds, discounts or exceptions; never claim to be a person; if unsure, say a team member will follow up and collect name and phone. Set its single goal, usually collecting contact details or booking an estimate.

Step 5: Set up the handoff

Decide where leads go: an email to the office, a text to the owner, or a contact in your CRM. Turn on the human takeover option if your tool has one, and set the bot to offer a phone number or callback whenever someone asks for a person or sounds frustrated.

Step 6: Try to break it

Before launch, spend 30 minutes attacking it. Ask for a discount. Tell it to agree to everything you say. Ask for a price on an unusual job. Ask about a competitor. Ask something legal or medical. Ask in Spanish. Every wrong answer becomes a new rule or knowledge base entry. Repeat until it refuses gracefully every time.

Step 7: Read every transcript for a month, then decide

Read all transcripts weekly and fix gaps. At day 30, count chats, leads captured, leads captured after hours, and bookings that came from chat. Compare with what the bot cost. Keep it, change it or turn it off based on that, not on how clever it sounded.

What it really costs

As of October 2026, a self-serve bot costs roughly $25 to $100 a month for a small business: Tidio from $24.17 plus the Lyro add-on from $32.50, Chatbase Hobby at $32 billed yearly, or GoHighLevel's $50 AI Employee Growth add-on if you are already on HighLevel. Intercom Fin is $0.99 per outcome plus seats or a minimum. A done-for-you build from the agencies documented on this site runs $1,500 to $3,500 up front plus $200 to $500 a month.

The cost that does not show on any pricing page is upkeep. Budget two to three hours to set it up, an hour a week for the first month reading transcripts, and 15 minutes whenever prices, hours or services change. A bot with stale information is worse than no bot.

Frequently Asked Questions