Seven shifts reshaping phone self-service, and what they mean for organizations that serve customers by phone
The phone is still the channel customers use the most, especially when trying to solve a complicated issue. However, getting help by phone has become more difficult. A third of callers who start in an automated menu now press out of it. Average wait times to reach a real agent have more than doubled since 2012. Nearly half of customers who try using self-service say the system didn’t understand what they were trying to accomplish.
Aside from caller behavior, there are three other major changes coming to phone lines. Vendors are retiring support for several widely used on-premise IVR and contact center platforms between 2027 and 2028. California has passed a law that gives callers to large businesses the right to ask for a person, and bars passing off a chatbot as a human. And analysts who spent years predicting big AI savings are now telling buyers to be more careful.
This paper walks through seven shifts in two parts: what callers are doing and what’s changing in the industry. It ends with four numbers to pull from your own reports and the dates worth putting on your calendar.
More callers are pressing out of the menu
ContactBabel has surveyed US contact center leaders every year for well over a decade, and one of its longest-running questions is how many callers who start in automated self-service leave the menu to reach a real person. In 2013, the average was 13 percent. By 2019 it was 23 percent. In 2023, it reached 32 percent, more than double the 2013 figure.1
Contact center managers in the same survey named the reasons. The top one, cited by 76 percent, was that the menu doesn’t offer what the caller needs. Next came too many security questions (65 percent) and too many options to choose from (59 percent). About half (53 percent) said callers don’t trust the system or want reassurance from a person.1
Why callers leave the menu
The menu doesn’t offer what they need76%
Too many security questions65%
Too many options59%
Share of US contact center managers citing each reason. Source: ContactBabel, 2024.
Pressing out isn’t always a bad sign. ContactBabel points out that a clearly marked way out, offered early, is part of a good experience. But when a third of callers bail, and the most common reason is that their question isn’t on the list, the menu has stopped matching the calls coming in. Customers don’t call with preset menu options in mind. They call with a problem best described in their own words.
Waits have more than doubled
In 2012, the average speed to answer in ContactBabel’s data was 31 seconds. In 2023 it was 79 seconds, after peaking at 101 seconds in 2021. Over the same stretch, the share of callers who hung up before anyone answered has stayed between roughly 5 and 7 percent, landing at 7.1 percent in 2023.1
Those figures only count the wait for an agent. ContactBabel argues that time spent in the automated menu belongs in the same number, “as the customers themselves surely do so.”1 A caller who spends two minutes in menus and then 79 seconds on hold has waited more than three minutes, whatever the speed-to-answer report says.
For an organization with many locations, this compounds. Each site answers its phones a little differently, the busy ones back up first, and the caller has no idea any of that is happening. They just know it took a long time to solve the issue they called about.
Callers will use AI, as long as there’s a way out
Traditional self-service tools still leave a lot of problems unsolved. In a survey of 5,728 customers, Gartner found that only 14 percent of customer service issues were fully resolved in self-service. Even for issues customers themselves called “very simple,” the figure was 36 percent. Forty-five percent of customers who started in self-service said the company didn’t understand what they were trying to do.2
That’s the gap conversational AI is built to close, and customers are noticing. In Gartner’s newer survey of 3,566 customers, run in early 2026, half said their interactions are easier when companies use generative AI. The same survey found that 87 percent consider it essential to have an option to reach a person. Gartner’s advice was blunt: “Service leaders should not use GenAI as a mandatory first step for every issue.”3
Those two findings fit together. Callers will happily let a system handle a quick question when they know a person is there if they need one. The cost of getting it wrong is real: in a follow-up release, Gartner reported that only 27 percent of customers would try a chatbot again after a bad experience.4 The organizations that do well with AI make the exit obvious and the hand-off quick.
Repeat calls are the cost nobody budgets for
SQM Group, which benchmarks hundreds of North American call centers every year, puts numbers on what happens when a problem isn’t solved the first time. Customer satisfaction drops 15 percent with every callback. It takes 1.4 calls on average to resolve an inquiry, and 2.5 calls for the 30 percent of customers whose issue isn’t settled on the first try. Repeat calls account for 23 percent of the average call center’s operating budget.5
A live call costs around $6 to $7 to handle, and an automated one well under a dollar.1 A repeat call actually pulls the average cost per call down, which makes the report look better while the caller and the budget are worse off. ContactBabel calls this out directly: it lowers cost per call “although being negative to both business and customer.”1 Internal reporting flatters the picture too. SQM found that call centers measuring first-call resolution with their own internal methods report results 10 to 20 percent higher than what customers say when asked directly.5
Legacy platforms are reaching end of support
A cluster of end-of-support dates lands between 2027 and 2028. Many self-hosted phone systems depend on at least one of these platforms.
- Nuance speech components. By August 2024, Microsoft had stopped selling Nuance’s on-premise speech products to new customers. Under Microsoft’s updated schedule, as published by Genesys, the Nuance Dialog Engine for IVR reaches end of support on June 30, 2027, and the speech recognition (Recognizer) and text-to-speech (Vocalizer) components follow on June 30, 2028.6 Exact dates vary by product and by how you bought it, so check with your supplier.
- Cisco. Versions 12.5 and 12.6 of Cisco’s contact center products, including Cisco Unified CVP, its IVR platform, reached end of sale on December 31, 2025. Software maintenance ends December 31, 2026, and the last day of support is December 31, 2027.7
- Genesys Engage on-premises. Genesys has set end of maintenance and support for December 31, 2028, with last full-year renewals at the end of 2027. Some components retire earlier.8
If one of these platforms sits behind your phone line, check which version you run and when its support ends. Knowing the date early leaves time to plan the change instead of rushing it.
The law now has a view on who callers talk to
In September 2026, California enacted AB 1609, the Right to Human Customer Service Act. Starting January 1, 2027, businesses with more than $500 million in annual revenue must let California customers ask for a person through a simple command, menu option, or spoken request, at least during business hours. They must make a good-faith effort to connect that person within 15 minutes or book a callback within one business day. Once a caller reaches someone, holds should stay under 15 minutes at a time and an hour in total. A chatbot can’t be presented as a human, and it has to say it’s AI if a caller could reasonably mistake it for a person. Hospitals’ health care communications are exempt.9 Maine already bars businesses from using an AI chatbot in a way that could lead a reasonable consumer to think they’re talking to a person, unless the business clearly discloses it, and its law covers voice as well as text.10
At the federal level, the FCC ruled in 2024 that AI-generated voices count as “artificial” voices under the Telephone Consumer Protection Act, so outbound calls that use them need the same prior consent as any prerecorded call.11 On September 30, 2026, the FCC also voted to update its rules on how consumers opt out of robocalls. Once the changes take effect, a business will be able to name an automated, interactive voice or keypad opt-out as its official opt-out channel for calls, as long as it tells callers about it on the call.12 And a proposed FCC rule, pending since August 2024, would require outbound calls that use AI-generated voices to say so at the start of the call. It would not cover AI that answers inbound calls.13
For anyone running AI on a phone line, the law and the customer research point the same way: tell callers when they’re talking to AI, and keep the path to a person short and easy to find.
AI’s economics are getting a reality check
Spending is moving fast. In a Gartner survey of 199 customer service leaders in spring 2026, AI spending was up 38 percent while total service budgets rose just 2 percent, with money moving away from labor and overhead.14
The same analysts are tempering expectations. Gartner predicts that by 2030, the cost per resolution for generative AI will exceed $3, “higher than many B2C offshore human agents.”15 It also predicts that more than 40 percent of agentic AI projects will be canceled by the end of 2027, citing “escalating costs, unclear business value or inadequate risk controls.”16 And 85 percent of service leaders told Gartner they are expanding what their human agents do, while only 31 percent have carried out or plan AI-driven layoffs through the first quarter of 2027.17
None of this argues against AI on the phone line. It argues for starting with the calls where AI clearly helps, like the common, answerable questions that flood the menu, and measuring cost per resolved call instead of assuming a headcount saving.
Four numbers to pull from your own phone system this quarter
Most phone systems already report these. Pull them by location if you can, since the averages hide the sites that are struggling.
- The share of callers who press out of the menu. Compare it to the 32 percent industry average, and look at which menu options people leave from.
- Total wait, including menu time. Add the time spent in the menu to the time spent on hold. That’s the wait your callers actually feel.
- Abandonment, by location and by hour. The 5 to 7 percent average hides the sites and times where it’s far higher.
- Repeat calls. Count how often the same number calls back within seven days. It’s a rough measure, but it’s honest, and it will tell you more than an internal resolution score.
Computer Instruments has built voice automation since 1979, through every shift in how callers talk to a phone system: touch-tone, press-or-say, speech recognition, and now conversational AI. The AI Concierge lets callers say what they need in their own words instead of working through a menu. It answers common questions from your own website and documents, around the clock. It can tell callers at the start that they’re talking to AI, and it hands them to a person as soon as they ask. It works alongside the phone system you already run, so you don’t have to replace anything to start. If your own press-out rate, wait times, abandonment, or repeat calls are telling you it’s time for a change, we’d be glad to talk through your options with you.
