For most organisations, a support conversation is considered finished once the customer's problem has been resolved. The ticket is closed, the chat is stored, and the team moves on to the next request.
The information contained in that conversation usually stays there too, buried in a transcript that may never be looked at again.
That has made sense when analysing customer conversations meant manually reviewing a sample of them. You could look at a few hundred interactions, identify some recurring themes and turn those findings into a report. But you could never realistically do that with every conversation.
AI changes the economics of this.
A conversational AI system can analyse thousands of interactions as they happen, looking for recurring questions, points of confusion, changes in sentiment and issues that keep appearing across different customers.
That gives support conversations a different kind of value. A customer asking why they can't find a particular product might seem insignificant on its own. If hundreds of customers are asking the same thing, it could point to a problem with navigation.
A recurring question about compatibility might reveal that product information isn't clear enough.
A steady increase in questions about cancellations could be an early indication that something has changed in the customer experience.
These are things customers are already telling businesses. The difficulty has been identifying the patterns quickly enough to do something about them.
This is also where I think the discussion around AI in customer service has been too narrowly focused on automation.
Deflection rates, response times and cost savings are important, particularly when support teams are dealing with large volumes of repetitive enquiries. But if AI is becoming involved in a significant proportion of customer interactions, the business is also creating a much richer source of customer insight than it has traditionally had access to.
McKinsey's research into AI adoption reflects part of this shift.
Businesses are using AI not only to generate content or automate tasks, but to capture and process information through conversational interfaces. Customer service and contact centre environments are among the areas where this is particularly relevant.
The scale of the opportunity is significant.
Gartner has forecast that conversational AI could reduce contact centre labour costs by $80 billion by the end of 2026.
Much of the discussion around that figure is understandably about efficiency and workforce requirements, but there is another consequence that receives less attention: as more customer interactions move through AI systems, more of the language customers use to describe their problems, preferences and frustrations becomes available for analysis.
The challenge is making that information useful.
For a CX team, knowing that 14% of conversations mentioned a particular issue isn't especially valuable on its own. The value comes when that finding reaches the product team, changes a piece of website content, prompts a change to the customer journey or reveals an issue that would otherwise have taken months to identify.
That is where conversational AI starts to look less like a support tool and more like an additional source of customer research.
There are still gaps in the technology. Forrester's 2026 review of conversational AI platforms found that analytics remains an area where many vendors fall short, with organisations often relying on separate tools to extract and report on conversation data.
The industry has spent a lot of time improving how AI handles conversations. The next question is how well it helps businesses understand those conversations.
For CX leaders, that is worth considering when evaluating AI.
Don't just ask how many enquiries the system can handle or how much support volume it can remove. Ask what the organisation will know about its customers after the system has handled 10,000 conversations that it didn't know before.
That may end up being one of the most valuable parts of the technology.
