Three quarters of customer contact still happens by phone: what voice AI changes about that

During OpenAI's DevDay keynote on 29 September, the company showed a new AI assistant called dots. In the demo, a dot named Alfred ends a phone call with the words 'Done, you're live', the moment a website goes live after its design was approved earlier in the conversation. Dots can be reached through ChatGPT, text message, Slack, Teams, and a plain phone call, runs continuously on its own cloud environment, and works independently on whatever task you give it.

It is a notable moment, not because an AI assistant can now make calls, that already existed in more limited form, but because one of the largest consumer platforms now treats phone reachability as a given alongside chat and email. For organisations still running a phone menu from the previous decade, that is a signal worth taking seriously.

The phone never actually went away

Many organisations have spent recent years investing mainly in chat, web forms, and self-service portals, assuming call volume would naturally decline. Industry publication CustomerFirst wrote in mid-September that an estimated three quarters of customer interactions in Western Europe still happen by voice, even as many organisations keep focusing primarily on digital channels.

That gap between attention and actual usage explains why phone wait times at many companies have barely dropped despite years of digitisation plans. Customers tend to call once a question becomes complex, emotionally charged, or simply needs resolving faster than a form allows. That behaviour does not change just because a chatbot sits on the website.

From phone menu to conversation

What is changing is the technology behind the phone line itself. The traditional phone menu, press 1 for billing, press 2 for support, is gradually giving way to voice technology that understands a question in plain language and routes or resolves it directly. Vendors that have worked on voice recognition for bots since 2018, such as AudioCodes, note that generative language models are now making these conversations noticeably more natural, including in Dutch.

The applications that pay off first are strikingly recognisable: booking or rescheduling an appointment, verifying identity before transferring a call, checking the status of an order or a report, and automatically logging a summary of the call afterwards. Tasks that used to cost a staff member minutes, and that a well-trained voice model now handles flawlessly within seconds.

Contact centres are moving at the same pace

Consumer platforms such as dots are not the only ones moving this way. Providers of business phone and contact centre software are expanding their AI functionality at the same pace, including voice agents that answer incoming calls and run outbound calling campaigns on their own, now in dozens of languages. The two developments reinforce each other: customers who already call and email an AI assistant at home expect the same ease once they get your business on the line.

Why staff are not redundant yet

Still, fully automating the phone line is not a good idea for most organisations yet. Research into customer preferences repeatedly shows that Dutch customers are more hesitant than average when a company offers only AI for customer contact, especially for sensitive or financial matters. An angry caller usually wants, above all, to feel heard in that moment, something voice technology still does not handle well.

The most promising approach is therefore hybrid: AI takes over the predictable, repetitive calls, leaving staff free for the conversations that need nuance, empathy, or a difficult decision. It also matters that a call can always be handed to a person easily and without frustration once the AI reaches its limit.

A practical point IT and management need to settle together is where the voice data goes. A call processed by an AI model is almost always transcribed and stored as well. For customer calls involving personal data, that means knowing in advance which vendor processes that data, for how long, and whether it stays within the EU.

What you can do this month

Start small and concrete. Map out which types of calls your main line receives most often, and which of those are predictable and low-sensitivity, think of opening hours, appointments, and status questions. Ask your current phone or VoIP provider what voice AI their platform already supports, and what a pilot for one or two of those call types would cost.

Then agree with your team how the handover to a staff member works once a call turns out more complex than expected, and tell customers clearly when they are speaking with an automated system. That transparency prevents the irritation that follows when someone assumes they are talking to a person and later finds out otherwise.

Your organisation's phone line is quietly gaining a new layer of technology this year, whether you steer that actively or not. Want to map out where your current telephony is ready for voice AI and where it is not, or help choosing a platform that grows with you? Zarioh is happy to think through the next step for your telephony and customer contact.