Chatbots used to frustrate people, AI agents changed that

AI customer support is hot right now, and it's moving from pilots into the core of how big companies serve their customers.

We've found a lot of interesting signals. Zendesk bought Forethought, whose AI agents were handling more than a billion customer interactions monthly. And Verizon's CEO outlined plans to automate much of customer service after testing AI agents. Support platforms and large companies are preparing for more work to move to AI.

Here, we'll look at which support work companies hand to AI first, how that changes what they buy, and where it leaves the startups selling to them.

Customers expect an answer in minutes, at any hour, and support teams are asked to deliver it with fewer people.

We compare six types of support AI companies, from autonomous agents to ticket routing and quality tools. Then we rank what banks, online businesses and support teams want fixed, including backlogs, staff turnover and outdated help content.

This tells you which support problems a team will pay to fix this year, and which ones still sit at the bottom of the list.

Per-seat licenses, per-resolution fees, per-conversation pricing, add-ons to help desk platforms and outsourcing contracts with AI built in: there are many ways to charge for AI support.

We compare seat subscriptions, pricing per resolved conversation and managed support. Charging per seat is harder when AI reduces staffing. Pricing per resolution rewards useful results, but even failed conversations cost money. Outsourcers face a different challenge when they automate billable work.

You'll see which models still work when AI does most of the work, and which ones shrink with every human seat that disappears.

Our growth estimate rests on a few forces. Here, we explain each one and how they connect.

Some drivers are easy to see: models that reason better, rising labor costs, and customers who want help at any hour. Others feed each other. Every resolved ticket shows where the help center is wrong. Better knowledge means more tickets the AI can close. And once AI takes the simple requests, the human team can focus on the hard ones, which makes the case for more automation stronger.

Before building or investing, you should know what lets AI close more tickets without hurting customer satisfaction, and what makes companies bring people back.

Ada and Wonderful are among the names buyers put on their shortlist first, but they're only a slice of the AI customer support market.

Some of the most interesting startups do one narrow job. They test support agents against thousands of simulated customers before launch, keep help center articles up to date, or send each ticket to the right person in seconds. Others build agents for one industry, like insurance claims or airline rebooking.

They're hard to compare because everyone now quotes a resolution rate. One company counts a ticket as resolved when the customer stops replying, another only when the problem is fixed. And many sell through help desks or outsourcers, so their names stay hidden.

That's why we built market grids and ecosystem maps. They sort companies by the job they do, from first reply to final fix, so you can see where the gaps are.

Cresta, Observe.AI, Ada… a few AI support companies have won large contact centers as customers.

Cresta started by coaching agents during calls. Observe.AI focused on conversation quality. Ada automated customer chats. Each began with a job companies already paid for, then expanded into automation. Replacing entire support teams remains unproven.

We'll show how these companies earned a place in the support team's daily work, and how they kept it as the big platforms added their own AI.