Live Chat Companies for Small Businesses: What to Compare Before Buying

For a small business, pick the live chat company by what the AI answers cost and how it hands over to a person, not by the feature list. Seat prices run from about $19 a user a month to flat workspace plans from $45, but AI answers are billed separately, and the handover rules decide whether customers stay happy.

Every vendor on a shortlist will have a widget, a shared inbox and canned replies. The things that actually separate them are three: what one seat costs, how AI answers are counted and billed, and what happens at the moment the bot gives up. Compare those, with your own chat volume, and the choice usually makes itself.

Follow me on Instagram@sagnikteaches

Plans and prices as of November 2026

All figures are USD list prices from each vendor's pricing page, and the vendors change them often, so confirm before you buy. The table gives the entry prices that matter to a team of one to five.

Connect on LinkedInSagnik Bhattacharya

One premise has changed. LiveChat's own pricing page now sends new trials to a product called Text, from the same team, which bundles live chat, a shared inbox, ticketing and an AI agent on one subscription. Existing LiveChat customers keep their plan and price. If you saw an older comparison quoting LiveChat plans, those numbers will not match what a new customer is offered.

Subscribe on YouTube@sagnikteaches

Why price per seat is the wrong first column

A seat price is easy to read and easy to be misled by. A per-user plan at $19 looks far cheaper than a flat $95 workspace, until you notice the first includes ten AI answers and the second includes about 450 automated conversations. If most of your value comes from the AI deflecting questions, the seat price is the smaller number.

Flat workspace pricing, as at Crisp, favours teams with several people watching chat part-time: the receptionist, the owner, a technician on a quiet day. Per-user pricing favours one person who handles everything. Count the people who will actually log in. Text states that every user who answers chats or tickets needs a seat, so a casual helper counts.

Compare AI answers by their definition, not their price

Four vendors, four units. Text bills a "resolution" when the AI directly solved a question, and multiple questions in one chat count once. Tidio counts a Lyro "conversation" as any interaction with at least one AI reply, however many replies follow, even if the customer returns later. Crisp sells credits that translate roughly into automated conversations. Intercom's Fin bills per outcome, and its pricing counts workflows and some hand-offs, not only the clean wins.

That matters because the same chat can be priced differently. Suppose a customer asks about delivery times, gets an answer and then asks for a person. Tidio would count one Lyro conversation. Text would count a resolution only if the AI solved it. Fin could count a hand-off outcome. You want the vendor whose definition matches what you consider a success, so that you pay for answers that saved you work.

Ask each vendor three questions in writing: what exactly triggers a charge, what happens when the monthly allowance runs out (stop, throttle or bill overage), and whether you can set a spending cap. A cap that stops the AI and routes everyone to a person is a safer default than an uncapped overage.

A worked monthly bill: a security installer's website

Take a security installer with a website that gets 400 chats a month. Two staff share the inbox. Say the AI could answer about half, 200 chats, on opening hours, quote requests and fitting times. Illustrative bills, from the list prices above:

• Text Essential, two seats: 2 x $19 = $38. It includes 10 AI resolutions, so 190 more are needed: four packs of 50 at $49.50 = $198. Total $236, and it buys 200 resolutions.
• Text Growth, two seats: 2 x $79 = $158. It includes 200 resolutions, so there is nothing extra. Total $158, but you pay for Growth features you may not need.
• Crisp at $95: a flat workspace for up to 10 seats, with about 450 automated conversations included. Total $95. Check that the AI features you need are in that tier.
• Intercom Essential seats plus Fin: 2 x $29 = $58, plus 200 outcomes at $0.99 = $198. Total $256, or $236 while Intercom's $19 new-customer seat price applies.

Tidio is harder to total from its public page, because its Lyro tiers beyond 50 conversations are chosen in a slider. Add the Customer Service plan from $24.17 to the Lyro price shown for 200 conversations to get your figure.

The lesson is not that one vendor wins. At 200 AI answers the spread is $95 to $256, a factor of 2.7, and it flips at other volumes. At 40 AI answers, Text Essential is around $38 + $49.50 = $87.50, Intercom's list seats plus Fin about $58 + $39.60 = $97.60, and Crisp's $45 Mini plan, with about 90 automated conversations, undercuts both. Do the sum at your volume, and again at double it.

The handover test is where small teams get burned

Handover is the moment the AI passes the chat to a person. Done well, the customer barely notices. Done badly, they repeat themselves, or they sit in a queue nobody is watching. Before you buy, run these six tests during the trial, which is 14 days on Text and 7 on Tidio, with Crisp offering 14 days on paid plans.

1. Ask for a person in plain words. "Can I speak to someone?" must reach a human, not another bot answer.
2. Fail the bot twice. Ask something it cannot know. Does it hand over after two misses, or keep guessing?
3. Chat outside opening hours. Does it say when someone will reply and capture an email address, or does it promise a reply that will not come?
4. Check the transcript. Does the agent who takes over see the whole conversation, so the customer does not repeat their question?
5. Check who gets the alert. If two people share the inbox, does one get a notification, and is there an owner?
6. Try a complaint. Type something angry. The right behaviour is an immediate human flag, not a cheerful answer.

Here is an illustrative record from a bicycle repair shop's test, which shows what to look for:

Test 3 - after hours (Saturday 19:40)
Customer: Is my wheel truing finished?
Bot: I can't see job status. A team member will reply when we open at 09:00 Monday.
     Can I take your email so we can reach you?
Result: PASS - honest about timing, asks for contact details.

Test 5 - complaint
Customer: This is the second time my brakes have been done wrong.
Bot: I'm sorry to hear that. Here are some tips on brake maintenance...
Result: FAIL - answered a complaint instead of handing over.

The second result would be a reason to change a setting or a vendor. Most tools let you add a handover rule or an instruction for words such as "complaint", "refund" or "second time", and re-running the test tells you whether it works. The test plan for a customer chatbot goes further if you want a longer script.

Teaching the AI what your business actually says

A chat AI is only as good as the content you point it at. Before the trial, gather the pages and documents you would want a new employee to read: opening hours, delivery or call-out areas, prices you are willing to publish, returns, warranty and the booking steps. Most vendors can read your website directly, and several accept uploaded files or a list of question and answer pairs.

Write the answers you do not want the AI to give as carefully as those you do. A typical rule set for a solar installer would say: never quote a price for a full system, always offer a site survey, never promise a date for installation, never comment on grants or tax. Put these in the instructions, then try to break them in the trial. If the AI quotes a price you never published, treat that as a failed test and ask the vendor how to stop it.

After launch, read a sample of transcripts weekly for the first month, and find out whether the tool shows you the questions it could not answer. That list tells you which page to add or fix, and is often worth more than the AI's answers. The tutorial on triggers, saved replies and AI answers covers the setup side.

Integrations: only count the ones you will use

Vendors list dozens of integrations, and most small teams use two. Make a list of the systems that need to know about a chat: your email, the calendar you book jobs in, the shop platform, the CRM or spreadsheet that holds leads, and perhaps WhatsApp or SMS. Then ask for each one: native, via a marketplace app, or via an automation tool such as Zapier, which costs extra. Text, for example, says its plans include a live Shopify catalogue, SMS through Twilio and workflows. Crisp lists inbox connections to WhatsApp Business, Instagram and SMS on its higher plans. These details differ by tier, so check the one you would buy.

For a business that books jobs, the integration that matters most is the one that turns a chat into a lead record without retyping. If that needs a paid connector, add its price to the vendor's total.

Where vendors hide the extra cost

Four items catch small teams out. Branding: Tidio sells a no-branding add-on at $16.67 a month, so the widget carries its name until you pay. Extra seats: Crisp charges $10 for each seat beyond its plan's included number on its top tier. API and usage limits: Text includes 10,000 API calls a month on Essential and sells extra packs. Support level: priority support often sits in a higher plan. Add whichever apply to your shortlist before you compare totals.

Which chats the AI should take, with a sorting exercise

Before you price anything, sort a sample of real enquiries. Take the last 50 emails, calls or chats and put each in one of three piles: the AI can answer from published information, the AI can gather details and hand over, and a person must handle it from the first word. The first pile decides how many AI answers you need and therefore which pricing model suits you.

An illustrative sort for an HVAC contractor, from 50 enquiries: 17 asked about opening hours, call-out areas or whether the firm services a particular brand (pile one); 21 wanted a quote or a visit, where the useful job is capturing the address, system type and urgency (pile two); 12 were existing customers chasing a job or a complaint (pile three). So the AI can fully answer about a third, and can usefully front-load about 40% more. If that contractor got 600 chats a month, the first pile would be about 200 chats, the same AI volume as the security installer's bills above.

Notice that the second pile, where the AI collects details and passes them on, is worth more than the first, and it is where the unit definitions matter. A vendor that bills a hand-off as an outcome will charge for it. A vendor that bills only for solved questions will not, but may offer less help on that task. Decide which pile you care about most before you read a price list.

A second sum: out-of-hours coverage

Many small firms buy chat to catch leads after 5 pm. Suppose a solar installer gets 30% of its website enquiries outside working hours, and 120 chats a month come in at those times. If the AI captures a name, phone number and roof type from 40% of them and one in ten of those leads becomes a survey, that is 4.8 surveys a month. If a survey is worth $60 of margin in expected value, that is about $290 a month. Against a plan costing $95 to $240, the AI pays back. Against a plan where every one of the 48 captured leads is billed as an outcome at $0.99, it costs $48 for those leads alone, which is still comfortable.

These are illustrative numbers, and the conversion rate is a guess. The point of writing the sum down is to replace "chat feels useful" with a figure you can check after two months, using the leads captured by the tool against the jobs booked.

What a good trial looks like

A trial that only tests the happy path tells you nothing. Plan it before you start the clock, because 7 and 14 days go fast.

1. Day 1. Install on one page that gets steady traffic, usually pricing or contact. Load the AI with your real opening hours, areas and returns text.
2. Days 2 to 4. Have two colleagues try to break it with the questions from your three piles. Log each wrong answer.
3. Days 5 to 7. Switch on for real visitors with a clear human option. Read every transcript.
4. Last day. Count chats, AI-answered chats, hand-overs and leads. Price your volume on the vendor's own calculator, not on the entry price.

Run two vendors at once if you can, on different pages, so you compare on the same traffic. Two free trials cost nothing but an afternoon of setup.

Contract terms to read in the small print

Annual billing saves money but locks you in. Text says annual plans run for the year and that monthly plans cancel at any time. On a first purchase, pay monthly for the first quarter even at a higher rate: Text's Essential is $25 monthly against $19 annual, a $6 difference per seat that buys you the freedom to leave once the tests above show a problem. Convert to annual once the tool has survived 90 days. Also confirm who owns the chat transcripts and whether you can export them, because they are the best training material you will have if you ever change tools.

A decision rule that fits on a card

• One person handles chat, AI answers under 50 a month: a per-seat plan with a small AI allowance, such as Text Essential, or Tidio's free plan while you learn.
• Two to five part-timers, 100 to 450 AI answers: a flat workspace plan such as Crisp's middle tiers, provided its AI features suit you.
• Heavy AI volume with a clear definition of success: per-outcome pricing such as Intercom's Fin, with a spending cap set.
• Not sure: start on a free tier, collect a month of real transcripts, then decide with numbers. Tidio's free plan comes with its 50 one-off Lyro conversations; Crisp's free tier gives two seats but no AI credits, so it tests the chat and handover side only.

The checks that rank highest are the handover tests, then the AI unit definition, then the seat price. If you would rather map your chat volume with someone first, that is part of my AI implementation consultation. For a printable version of this decision, the live chat buying checklist is a companion tutorial. Whatever you choose, put a calendar reminder at 90 days to compare the invoice with the sums above, because chat volume and vendor pricing both move.

Live chat buying questions

Do I need a live chat company if I only get a few chats a week?

Probably not at first. Under about ten chats a week, a contact form or WhatsApp link does the job and costs nothing. Add live chat when people abandon pages with unanswered questions, or when you notice the same five questions arriving by email every day.

What is an AI resolution and why does the definition matter?

It is the unit some vendors bill for when the AI answers a question without a person stepping in. Definitions differ: one counts a chat once however many questions it handles, another also bills hand-offs. Read the definition before comparing prices, because the same chat can cost nothing at one vendor and a dollar at another.

Can the AI answer from my own website content?

Most vendors let you point the AI at your website pages or help documents so it answers from them. Check how often it re-reads them, what it says when it does not know, and whether you can see the questions it failed on.

How do I stop a customer getting stuck with the bot?

Set a rule that any request for a person, any complaint and any two failed answers hand over to a human, and show them a clear button for it. Then test it yourself, including outside opening hours, where the handover often fails.

Further reads

• How to Build the FAQ Your AI Chatbot Needs Before Launch — Write the FAQ the chat AI will answer from.
• AI Chatbot vs Live Chat vs Help Desk: What a Small Team Needs — When a help desk fits better than live chat.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: the pricing pages and pricing FAQs of Text (on livechat.com), Tidio, Crisp and Intercom, read on 5 November 2026.