83 AI customer service tools. Nobody paid to be here.
Find what each one costs after the trial ends, which meter it bills you on, what it cannot do, which ones were quietly acquired, and which cannot be bought without a sales call. Every price here was read on the vendor’s own page and carries the date we read it.
Start with the Job, Not the Brand
A Shopify returns agent and an enterprise contact centre suite market themselves in almost identical language. They are not competing for the same buyer, and picking from the wrong shelf costs more than picking the wrong brand.
AI Agents & Chatbots
Autonomous agents that answer customer questions end to end and escalate what they cannot handle.
AdaHelp Desk & Ticketing
Shared inboxes and ticketing systems with AI drafting, triage and summarisation layered on top.
E-commerce Support
Support built around orders, returns and refunds, wired directly into Shopify.
Voice & Phone AI
Agents that answer the phone, not just chat. Priced per minute rather than per resolution.
Agent Assist & Copilots
AI that sits beside a human agent suggesting replies, rather than replacing them.
Balto
Observe.AIQA & Conversation Analytics
Automatic scoring of support conversations, including grading the AI agents themselves.
Enterprise & CCaaS
Full contact-centre suites. Sales-led, six-figure contracts, no public pricing.
Or Start with What Is Going Wrong
Nobody wakes up wanting an AI agent. They wake up with a queue full of where-is-my-order, a bill that doubled the month they switched it on, or a vendor who will not say what counts as a resolution.
What We Found
Cost modelled at real ticket volumes, vendor claims checked against source, and the acquisitions that comparison sites are still four months behind on.
The security page and the contract disagree — and at one vendor the contract is the generous one.
We read eleven vendors' security pages against their own data processing agreements, annexes and product terms. The promises turn out to live in different documents from the obligations, and not always in the vendor's favour.
One vendor prints how long a deployment really takes. It is sixty days, for the simplest product here.
We counted the documented setup steps at seven platforms and put them next to the time each vendor claims. Two publish no documentation at all. And at three of them, the testing you do before launch is billable.
Chatfuel ships four autonomy levels and starts every bot on the one where the AI writes nothing.
Every vendor here will tell you their agent takes action, and almost none will say who approves it or where the record lands. One small vendor publishes a four-level autonomy slider the rest of the category should be measured against.
Your support agent can do outbound now. Read what it costs before you switch it on.
Support platforms are quietly becoming outbound tools — WhatsApp campaigns, proactive messages, lead qualification. The capability is real. The pricing is where it gets strange: at one vendor the identical mechanical act costs ten times more when you call it sales.
Intercom raised its own resolution rate from 33% to 50% without resolving a single extra ticket.
Every vendor publishes a resolution rate and no two of them measure the same thing. The silence window that turns a conversation into a resolution runs from three minutes to seventy-two hours, abandonment counts as success at some vendors and failure at others, and one platform bills for outcomes its own headline metric excludes.
Buyers add "reddit" to escape vendor content. We counted what is waiting for them.
Adding reddit to a software search is now standard practice, because buyers do not trust vendor listicles. We counted every multi-vendor ranking posted to the main AI customer support subreddit over three months. Fifteen posts, four accounts, and one product first in thirteen of them.
Meta started billing its own WhatsApp agent by the token on 1 August. Almost every explainer of it is wrong.
Meta Business Agent is the first AI in this directory billed per token rather than per resolution or per seat. The rate is published and checkable, the vendor explainers describing it get the mechanism wrong, and there is a spend ceiling that does not exist unless you set it.
Zendesk makes you choose messaging or email when you create the agent, and you cannot change it later.
Running a separate bot per channel multiplies prompts, knowledge updates and failure modes, so every vendor now promises one agent everywhere. We read the documentation. Two platforms make you build two agents, one publishes a hard limit on where its memory travels, and the biggest omnichannel claim covers exactly three channels.
Tidio remembers you for three hours. Fin remembers you for 180 days. Most publish nothing at all.
A returning customer explaining their problem again is the fastest way to lose them, and every vendor says its agent remembers context. We read the documentation of ten platforms to find out what that actually means, and the published retention windows range from three hours to a hundred and eighty days.