What does "cheapest" actually mean for website chat?
Cheap splits into three questions that rarely share an answer: which tool has the lowest advertised price, which costs the least across a full year once you count your own team's time, and which carries the lowest risk of a bill you did not plan for. A free widget that needs someone technical to patch it and a staff member to watch it at night is not cheap, it has just moved the cost from your card statement to your payroll. A hosted plan that looks small at signup is not cheap either if it charges for each person who logs in and each question the AI answers, because the better it performs the more it costs.
A more useful test takes about ten minutes with a notepad. Write down how many people will need access, including the part-timers and the weekend cover. Write down which channels you actually want covered, not the ones you might want eventually. Write down the handful of questions that make up most of your inbound volume. Then price every candidate against that picture rather than against its headline tier. Most of the cost difference between vendors shows up in those three lines, not in the feature list.
Where do the free options stop being free?
There are three honest zero-cost starting points, and all three are priced in labor rather than in money.
Open source is the first. Chatwoot ships as self-hostable software with no license fee, which makes it genuinely attractive if you already run servers. You supply the hosting, the upgrades, the backups, and somebody who can read logs when the message queue stalls on a busy afternoon. For a business with in-house technical capacity, that is a reasonable trade. For a business without it, the first outage costs more attention than a year of subscription fees.
Free and near-free hosted tiers are the second. Several vendors, Tidio among them, publish an entry plan at no cost or close to it. These are real products, not bait, but they are usually capped on conversations or AI replies, and the automations that remove the most work tend to sit on paid tiers. The risk is not the cap itself, it is where the cap lands: free allowances run out in your busiest week, which is exactly the week you least want the widget to go quiet.
Building your own is the third. A model API plus a chat widget is not hard to prototype. The cost arrives afterward, in retrieval that keeps answers tied to your real content, in handoff to a human, in conversation logging, in spam handling, and in the ongoing maintenance of all of it. Unless you have engineering time you are not otherwise using, this is usually the most expensive option wearing the cheapest label.
Which pricing model costs you the least as you grow?
Four models dominate the market, and the shape of each one matters more than its starting number.
Per-seat pricing charges by the person with a login. It is predictable, and it quietly discourages the thing you want most, which is giving everybody who touches customers a window into the conversation. Teams on per-seat plans end up sharing a login, which breaks their own reporting.
Per-resolution or per-conversation pricing charges by the work the AI does. The logic is tidy, but the incentive is inverted from your point of view: a bot that handles more questions on its own produces a larger invoice, so the better your setup works, the more you pay for it. Volume spikes, including the seasonal ones you cannot control, land directly on the bill.
Per-contact pricing charges by how many people are in your database, which tends to climb whether or not those contacts ever message you again.
Flat pricing caps the exposure. You pay one number and the usage question stops being a budgeting question. Corebee sits at that end: $99 per month with unlimited seats and unlimited AI conversations, no per-seat fee and no per-resolution fee, which you can read in full on the pricing page. The point is not that flat is the only sane model, it is that flat is the only model where your busiest month and your quietest month cost the same, and that predictability is worth real money to a small team.
What do you need ready before you install anything?
An AI chat agent is only as cheap as the content it can answer from, because every question it cannot answer becomes a human reply you pay for anyway.
Start with the answers you already send by hand. Shipping times, opening hours, pricing, returns, booking, service areas, what you do and do not cover. Most businesses find these already written down somewhere, scattered across email templates, an FAQ page, and the heads of two long-serving staff. Pulling them into one place is the highest-value hour in the whole project, and it pays off regardless of which vendor you pick. A tool that reads your public website and a structured knowledge base gets most of the way there without anyone writing new documentation.
Then decide the escalation rule before you turn anything on. Who takes over when a customer asks for a person, and how fast. Written down in a sentence, this prevents the failure mode that makes chat widgets expensive, which is a bot that argues with somebody who already wants a human.
Finally, list the channels. Web chat alone is the cheapest install, but if half your questions already arrive on WhatsApp or by phone, a web-only widget leaves that work untouched and you pay for it twice.
How do you install one without a developer?
For almost every hosted tool, installation is a single script tag placed before the closing body tag of your site template. On WordPress, Shopify, Squarespace, Webflow, or Wix, that goes in a settings field or a small plugin, with no code editing at all. If you already run Google Tag Manager, the tag can go there instead, which means you never touch the site files.
The part that takes real time is not the tag, it is the content and the routing. Setup for Corebee runs about ten minutes and needs no developer: connect the site so it can read your pages, confirm the escalation rule, and open the shared inbox where human replies land. Budget the rest of the first day for reading the early transcripts, which is where you find the three questions you forgot to answer.
One practical note: install on a staging site or during a quiet hour if your site is heavily customized. Not because widgets commonly break layouts, but because fixing a z-index conflict at nine in the morning is a worse experience than finding it at nine at night.
Which hidden costs catch people out?
Channel add-ons are the most common. Web chat is included; voice, WhatsApp, and SMS frequently are not, and they are often sold as separate line items or reserved for higher tiers. If phone calls and messaging apps are part of your reality, price them in at the start rather than discovering them in quarter two. Platforms that include WhatsApp, voice, Slack, Facebook, Instagram, SMS, and Telegram in the base plan, as Corebee does, change the comparison substantially against a web-only quote.
AI as an upsell is the second. Several established help desks price the inbox at one level and the AI capability at another, sometimes with its own usage meter on top. Comparing a base help desk price against an AI-inclusive price is not a like-for-like comparison.
Integration gating is the third. CRM sync and calendar booking are often tied to mid or upper tiers. If the bot cannot write a lead into your CRM or put a meeting on the calendar, somebody on your team is retyping it, and that is the real cost.
Annual commitments and migration cost round out the list. Discounted annual billing is fine when you are confident; it is expensive when you are still testing. And moving your macros, saved replies, and conversation history between vendors is tedious work nobody budgets for, which is a good argument for testing cheaply before committing.
When is paying more the cheaper choice?
Sometimes the larger platform is the right call, and it is worth saying so plainly. Zendesk and Freshdesk offer deep ticketing workflows, granular permissions, and reporting that larger support organizations genuinely need. Intercom has strong product-tour and in-app messaging tooling for software companies. Gorgias is built tightly around ecommerce order data and suits Shopify merchants who live inside order lookups. Help Scout is a clean shared inbox for teams who want email handled well without a heavy agent console. Chatwoot gives technical teams full control of their own stack.
If your bottleneck is complex routing across many queues, strict audit requirements, or a workflow that your whole operation already runs on, the cheapest tool will not fix it and switching may cost more than it saves. If your bottleneck is that the same questions arrive every day on several channels and a small team answers them by hand, the flat-rate end of the market tends to win on total cost. A side-by-side read of Zendesk alternatives is a fair way to see where the line sits for your situation.