A website AI chatbot costs anything from nothing (Shopify Inbox's built-in AI agent) to about $25 to $150 a month for small-business tools, or roughly $0.50 to $1 per resolved conversation on per-outcome plans, plus 5 to 15 hours of setup. It pays off when it saves more staff time, or wins more orders and enquiries, than that each month.
For most small businesses the subscription is not the big number. Setup time, keeping the bot's knowledge current and the cost of the occasional wrong answer usually outweigh it. The payback question comes down to volume: how many repetitive questions arrive each week, and how many enquiries go unanswered because nobody's there to reply.
What you actually pay for
| Cost | Type | Typical range for a small business |
|---|---|---|
| Platform subscription | Monthly | $0 to about $150 |
| Per-use AI charges (per outcome, conversation or credit) | Monthly, varies with volume | $0 to a few hundred dollars |
| Removing the vendor's branding | Monthly, on some tools | $0 to about $100 |
| Writing the knowledge base: FAQs, policies, product facts | One-off internal time | 5 to 15 hours |
| Connecting order lookup, email or CRM | One-off | 1 to 10 hours, or paid help |
| Reading transcripts and fixing gaps | Ongoing internal time | 30 to 90 minutes a week |
| Wrong answers: refunds, goodwill, lost customers | Hidden | Low if well tested; can be the biggest cost if not |
The knowledge base is where the setup hours go. A bot can only be as accurate as the pages and documents it reads, and most websites don't state delivery days, cut-off times or refund rules clearly enough for a machine to quote. Training an AI chatbot on your FAQs, policies and prices covers that work.
Price snapshot of common options, September 2026
Prices change often; confirm on each vendor's pricing page before you decide.
| Option | How it's priced | Worth knowing |
|---|---|---|
| Shopify Inbox AI agent | Free | Basic plan or above with new customer accounts; can use web search as a secondary source, so test it on your own policies |
| Tidio | Starter from about $24 a month billed annually; Lyro AI add-on from about $32.50 a month for 50 AI conversations | Free plan with 50 billable conversations for testing |
| Chatbase | Hobby $40 a month (700 message credits); Standard $150 (4,000 credits); about 20% off annually | Extra credits $40 per 1,000; removing branding is $99 a month |
| Intercom with Fin | Essential from $29 a seat a month billed monthly, plus Fin from $0.99 per outcome | An outcome includes a customer not asking for more help, so read the definition |
| HubSpot Customer Agent | 50 HubSpot credits (about $0.50) per resolved conversation | Professional and Enterprise tiers only, not Starter |
| Zendesk | Suite Team $55 an agent a month billed annually; AI agents included, charged on automated resolutions | Better suited to businesses already running a help desk |
Cheap, mid and high: three monthly totals
Illustrative totals, counting internal time at $20 an hour so it isn't treated as free.
- Cheap, about $40 a month all in: a free built-in agent, 30 minutes a week reading transcripts ($40 of time). Suits a shop already on a platform with an agent included.
- Mid, about $120 to $180 a month: a $40 to $65 subscription with AI conversations included, plus about an hour a week of upkeep ($80 to $90), plus occasional top-ups. Suits a business with a few hundred website questions a month.
- High, $400 a month or more: a help desk seat or two, per-outcome AI at a few hundred resolutions a month, and weekly upkeep. Suits a busy online shop where the alternative is hiring customer-service hours.
Add the one-off setup on top: 8 hours of your time at $20 is $160; paid help to connect order lookup might be a few hundred dollars more.
The payback sum
Monthly value = (questions answered fully by the bot × minutes each would have taken × staff cost per minute) + (extra orders or enquiries captured × their profit) − subscription and usage − upkeep time.
Payback in months = one-off setup cost ÷ monthly value. If monthly value is negative, it never pays back.
Two numbers are worth measuring before you buy anything. Log every website and email question for one normal week, marking which are repetitive (hours, delivery, stock, booking). And note how many enquiries arrived out of hours and waited until morning.
Keep a one-week question log first
The log doesn't need software. A shared spreadsheet with five columns is enough: time, channel, the question in the customer's words, repetitive or not, and minutes spent. Here's an illustrative extract from the farm shop's log:
| Time | Channel | Question | Repetitive? | Minutes |
|---|---|---|---|---|
| Mon 07:40 | Can I skip this week's box, away till Sunday | Yes | 4 | |
| Mon 21:15 | Website chat | Do you deliver to the next village over | Yes | 3 |
| Tue 10:02 | Box had a split bag of flour, can you replace it | No | 9 | |
| Tue 22:48 | Website chat | What's in the large box this week | Yes | 3 |
| Wed 13:30 | When do turkey orders open | Yes | 4 | |
| Thu 18:05 | Can we book the café for a 60th birthday | No | 12 | |
| Fri 20:12 | Website chat | Is the shop open Sunday | Yes | 2 |
Two patterns jump out of even this small extract: the repetitive questions are quick but constant, and a lot of them arrive after 8pm. Both matter for the payback sum.
If the log is long, an AI assistant will sort it faster than you can:
Below is a week of customer questions from my [business type].
Group them into topics. For each topic give: number of questions,
whether a chatbot could answer it fully from our website and policies
(yes / partly / no), and why. List "no" topics separately; these must
go to a person. Don't invent facts about my business.
[paste log]Illustrative output: "Delivery area and days: 14 questions, yes. Pausing or skipping a box: 11, partly: the bot can explain how, but only the account page can make the change. Box contents this week: 9, yes, if the contents page is updated each Monday. Turkey orders: 6, yes. Damaged or missing items: 5, no: needs a refund decision. Café bookings: 3, no: needs availability." One fix was needed: the AI had marked "opening hours" as yes, but the shop's website listed winter and summer hours in two places that disagreed. The chatbot would have repeated the contradiction. The pages were corrected first.
A farm shop's numbers
Take an illustrative farm shop with an online veg box service. The one-week log showed 80 questions through the website chat and email. Forty-four (55%) were repetitive: delivery days, box contents, how to pause a subscription, Christmas turkey ordering, opening hours. Each took a staff member about four minutes.
- Staff time on repetitive questions: 44 × 4 minutes ≈ 2.9 hours a week.
- Share the bot answered fully in testing: about 70%, so roughly 2 hours a week saved, or 8.7 hours a month.
- Value of that time at $16 an hour: about $139 a month.
- Upkeep: 30 minutes a week reading transcripts, about 2.2 hours a month, $35.
- Setup: 8 hours writing and checking delivery, subscription and turkey pages, $128.
| Built-in free agent | Paid tool at $65 a month | |
|---|---|---|
| Monthly time value saved | $139 | $139 |
| Subscription and usage | $0 | $65 |
| Upkeep time | $35 | $35 |
| Monthly value | $104 | $39 |
| Payback on $128 setup | About 5 weeks | About 3.3 months |
On time savings alone, both pay back, but the free option is far stronger. The shop also noticed something the time sum misses: six people in the test month asked about veg boxes between 9pm and midnight, and three signed up that night after the bot answered. If each subscriber is worth about $90 of profit over their time with the shop, that's potentially more than the time saved, though a month is too short to be sure. It keeps counting.
A catering company: payback through leads, not time
An illustrative catering company gets only about 25 website questions a week, so time savings are small. What matters is that wedding and party enquiries arrive in the evening, when couples are planning together, and a slow reply loses them to whoever answers first.
Its bot does one job: answer the basics (minimum guest numbers, dietary options, how far ahead to book) and collect the date, guest count, venue and budget, passing a complete enquiry to the owner's inbox. The illustrative sum: four more complete enquiries a month than the old contact form produced, one in eight turning into a booking, and around $1,800 of margin on a typical booking. That's half a booking a month, roughly $900 of margin, against a $40-a-month tool. Even if the real figure is a quarter of that, it pays. The only way to know is to tag chatbot-sourced enquiries and follow them to booked or lost; adding an AI chatbot that captures leads on your website shows the setup.
What the AI itself costs, if you build your own
It helps to see how little of a chatbot bill is the AI model. On OpenAI's API, gpt-5.6-luna costs $0.20 per million input tokens and $1.20 per million output tokens (a million tokens is roughly 750,000 English words). A four-exchange website conversation that sends about 2,500 tokens of your knowledge base and chat history each time uses around 10,000 input tokens and 1,200 output tokens:
- Input: 10,000 × $0.20 ÷ 1,000,000 = $0.002
- Output: 1,200 × $1.20 ÷ 1,000,000 ≈ $0.0014
- Total: about a third of a cent per conversation, or roughly $3.40 per 1,000 conversations
So when a product charges $0.50 to $1 per resolution, you're paying for the interface, knowledge base, handover, reporting and support, not the model. Building those yourself is possible, and API use is billed separately from any ChatGPT or Claude subscription. But for a small business the build and upkeep usually cost more than they save. Whether to build or buy a customer service chatbot goes through that decision.
Hidden costs that sink the payback
- A wrong answer that costs a customer. An illustrative butcher's bot told a customer that a Christmas order could still be placed on 23 December; the cut-off was the 18th. One unhappy regular and a refunded deposit wiped out a month's savings.
- Knowledge that goes stale. Seasonal changes, new prices, holiday closures. Every change is a page to update and a question to retest.
- Billing that counts drop-offs. On per-outcome plans, conversations where the customer simply leaves may be billed. Check the definition against your transcripts.
- Customers who can't reach a person. If the handover is clumsy, people give up. Watch for chats that end in "I'll just ring you", and ask whether they did.
- Seat creep. Help-desk-style tools price per staff seat; adding part-timers to read chats adds cost.
A food truck that shouldn't buy one
An illustrative food truck gets perhaps five questions a week through its website, and forty through Instagram direct messages: where are you today, are you at the Saturday market, do you do private hire. A website chatbot would sit almost unused. The cheaper answers cost nothing: a pinned Instagram highlight with the week's locations, saved replies for the three most common questions, and a private-hire page with a proper enquiry form. If the DMs grow past what the owner can answer between services, an Instagram auto-reply tool is the thing to cost, not a website bot. The general rule: put the bot where the questions already are.
When a chatbot won't pay off
- You get fewer than about 20 repetitive questions a week, and a free built-in agent isn't available.
- Most questions are unique or need judgement: custom quotes, complaints, technical advice.
- Your delivery, refund and booking information isn't written down clearly anywhere yet. Write the pages first; they help customers even without a bot.
- Your customers overwhelmingly phone or message on social media rather than use your website.
If you're heading towards buying, the questions to ask a chatbot vendor before you sign up will pin down the real price. After launch, check the numbers at 60 days: repetitive questions answered fully, handovers, leads captured and staff time compared with your one-week log. Measuring whether your AI chatbot is actually working sets out those measures.
The farm shop's 60-day check
At day 60, the farm shop compared the bot against its original one-week log. Illustrative results:
- Conversations: 612 over eight weeks, about 77 a week, slightly up on the log because customers used chat for things they'd never have emailed about.
- Answered fully: 64%, a little under the 70% seen in testing. Most of the gap was questions about this week's box contents on weeks when the page wasn't updated by Monday.
- Handed to staff: 21%, all reached a person within opening hours.
- Staff time on website and email questions: down from about 2.9 hours a week to 1.2.
- Wrong answers found in transcripts: 3, all about the café's winter closing time, fixed with one page edit.
- Veg box sign-ups after a chat: 7, of which 4 were after 8pm.
Put through the payback sum, 1.7 hours a week saved rather than the 2 estimated gives a time value of about $118 a month instead of $139; after $35 of upkeep, the free agent still returns about $83 a month before counting any sign-ups. The verdict was to keep it, and the one change was operational rather than technical: updating the box contents page became a named person's Monday-morning job. Most chatbot improvements after launch look like that. The software rarely needs changing; the information it reads does.
Chatbot cost questions
Is a free chatbot good enough for a small business?
Often, yes, if it's the AI agent built into a platform you already use and it passes your own test questions. Free tools usually limit volume, branding or connections rather than answer quality. Test it on your twenty most common questions and your five most awkward ones before deciding you need to pay for something better.
Why do per-resolution prices cost so much more than the AI itself?
Because you're paying for the software around the model: the interface, the knowledge base, handover to staff, reporting, security and support. The model's own processing cost per conversation is usually a fraction of a cent on cheaper models. Building that surrounding software yourself is possible, but the build and upkeep usually cost more than a small business would save.
How many questions a week does a chatbot need to be worthwhile?
As a rough guide, fewer than about 20 repetitive questions a week rarely justifies a paid tool on time savings alone, though a free built-in agent may still be worth switching on. Above that, the maths usually works if the bot can answer at least half of them fully. Lead capture out of hours can change the picture for high-value services.
Do I have to keep paying if the chatbot is quiet in winter?
On per-seat or flat monthly plans, yes, unless you cancel or downgrade. On per-outcome or credit plans, a quiet month costs less automatically. If your business is strongly seasonal, compare a year of costs on each pricing model using your own monthly question counts, not an average month.
Further reads
- Best AI Chatbots for Small Business Websites — A comparison of chatbots suited to small business websites.
- 9 AI Chatbot Mistakes That Lose Small Businesses Customers — The chatbot mistakes that quietly cost you customers.
- AI Chatbot vs Live Chat vs Help Desk: What a Small Team Needs — Whether you need a bot, live chat or a help desk.
- How to Stop an AI Chatbot Giving Customers Wrong Answers — Keep the bot's answers accurate once it's live.
- How Much Does a WhatsApp Chatbot Cost? Fees, Tools and Setup — What the same job costs on WhatsApp instead of your website.
- How Much Does a Custom AI Assistant Cost a Small Business? — What a custom-built AI assistant costs by comparison.
- Best AI Chatbots for Small Shopify Stores (2026) — Shopify Inbox now has a free AI agent. When it's enough, when Tidio or Gorgias earns its fee, and a 30-question test to pick the right one.
- How Much Does It Cost to Add AI to a Shopify Store? — Adding AI to Shopify can cost $0 extra, or $35 to $500+ a month with apps. Itemised prices, three store budgets and the costs listings don't show.
- Can a Hotel Chatbot Take Bookings or Only Answer Questions? — The three things a hotel chatbot can do with a booking enquiry, what each needs from your booking engine, and which suits a small property.
- AI Client Intake for Law Firms: Qualify Enquiries Out of Hours — Set up overnight AI intake that gathers facts, screens fit and urgency, books consultations and hands a clean summary to a person each morning.
- AI Receptionist or Chatbot: Which Suits a Cleaning Company? — How to choose between an AI receptionist and a chatbot for a cleaning company, with an enquiry tally, prices and the intake script both need.
- Answering Pupil and Parent Enquiries With AI at a Driving School — Build an answer sheet from your last 50 enquiries, put AI on web chat or WhatsApp, and hand test dates, complaints and progress questions to a person.
- AI Chatbots for Nursery Enquiries and Visit Bookings — What a nursery chatbot should answer, what it must hand to the manager, how show-round booking works, and the two mistakes that cost parents' trust.
- Can AI Give Customers a Ballpark Price for Decorating Work? — How a decorator can let AI give honest ballpark prices from a room-by-room grid, and which jobs it should refuse to price.
- Tidio vs Intercom: Which AI Chat Tool Suits a Small Business? — Tidio includes ten seats and sells AI in conversation blocks; Intercom bills per seat and per resolved outcome. The sums below show when each wins.
- Should a Custom GPT Answer Your Customers? Limits and Safer Options — Customers must sign in to use a GPT, you can't read the chats, and there's no handover. Here are the chatbot options built for customers, with costs.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: vendor pricing pages checked Sep 2026 (Intercom, Tidio, Chatbase, Zendesk); HubSpot credits and Customer Agent pricing; Shopify's documentation on the Shopify Inbox AI agent; OpenAI API pricing page.