In ContactBabel's research, 67 percent of U.S. contact centers expect AI to handle most routine contacts within the next two to three years, with human agents concentrating on complex and emotional cases. The industry is unusually united on this: only 1 percent expect near-total automation, and only 2 percent expect AI to make little difference.
ContactBabel's new study, "AI in U.S. Contact Centers: The Reality," surveyed 175 U.S. contact centers and 1,000 U.S. consumers, and the consumer findings support that plan on one condition: Routine has to mean what customers think it means. Their definition is clear and consistent but a good deal narrower than the one around which many automation roadmaps are built.
Sixty-six percent of U.S. consumers name at least one situation in which they would actively prefer AI to a person, which is a genuine preference rather than grudging tolerance. A quick answer to a simple question tops the list, at 41 percent, followed by avoiding a wait on hold, at 32 percent, getting help outside normal hours, at 27 percent, and simply not having to talk to anyone, at 25 percent.
The appetite collapses once a contact carries history. Only 14 percent of consumers would choose AI when following up on an existing case, where the customer has already explained the problem at least once, might have been promised something, and is usually less patient than the first time around.
Urgent is fine, complicated isn't.
Consumers were also asked which channel they would use in three scenarios designed to isolate emotion, urgency, and complexity. The phone wins all three, but the margins tell the story. For an urgent interaction, website self-service runs the phone close at 32 percent against 36 percent. For an emotional one, self-service falls to 11 percent, while for a complex one the phone and an in-person visit take 62 percent between them against 12 percent for self-service.
What makes urgency work for automation is that an urgent request is usually a simple one: the customer needs a single piece of information and needs it now. Customers will happily take a bot that delivers speed and simplicity together, while emotion and complexity send them to a person by margins that no plausible improvement in bot capability is likely to close soon.
Age sharpens the picture. Preference for the phone in the emotional scenario rises from 35 percent of 18 to 34 year-olds to 52 percent of those over-65, and concern about being unable to reach a human agent rises from 52 percent to 80 percent across the same range. An organization with an older customer base is working to a much higher bar than the averages suggest.
The exit matters as much as the bot.
Customers' worry about reaching a person was earned long before AI arrived, through decades of touchtone menus and hard-to-find phone numbers. Any organization deploying a bot inherits that skepticism whatever its own record. The handover is where it gets confirmed or overturned.
On contact centers' own assessment, only 32 percent move customers from a bot to a human agent seamlessly with full context. Seventeen percent say most transfers go smoothly but context is sometimes lost, 15 percent say transfers often lose context or history, and 31 percent say it depends on the channel. Five percent offer no route from their bots to a person at all.
That is organizations grading their own work, so the customer's view is unlikely to be kinder. The depends-on-the-channel group is the encouraging one: Those operations already have a working handover somewhere in their own operation, and passing the conversation history and customer identity with a transfer is an integration task rather than a question of AI capability.
Automating on the customer's terms requires the following:
- Define routine strictly. A contact qualifies for automation when it can be completed in a single exchange, has no case history behind it, and carries little consequence if it needs correcting. That still covers a great deal of volume.
- Treat an open case as a reason to route to a person. Recognize a follow-up at the start of the interaction and offer an agent straight away rather than asking a bot to pick up a story it never heard.
- Offer the way out early. A route to a person should be visible from the first turn. Customers who feel trapped will zero out, pick irrelevant menu options to reach someone, or take their business elsewhere.
- Fix handover one channel at a time. Start from whichever channel already works and make sure every transfer carries the conversation so far, so nobody has to explain the problem twice.
The 67 percent expectation is entirely compatible with what customers want, provided the description of routine means a single-step request with no history behind it. If it drifts toward meaning anything the organization would simply rather not staff, the plan runs straight into the people it is meant to serve.
Customers have drawn the line clearly: Automate the simple and the urgent, leave the difficult and the emotional to a person, and never make that person hard to reach. Most of them will choose AI willingly on those terms.
Steve Morrell is principal analyst at ContactBabel. He can be reached at smorrell@contactbabel.com.