Skip to content

Voice AI and Semantic IVR: The End of Press 1 in Customer Service

How voice AI replaces the numeric IVR menu: the customer speaks, the AI understands, resolves or hands off to a person, on WhatsApp, on the web and by phone.

Marlos Carmo

Founder of Tolky

Updated on November 16, 2026

9 min read
Voice AI and Semantic IVR: The End of Press 1 in Customer Service

Share

Everyone knows the scene: call a company, hear six options, press the wrong number and go back to the start. The traditional IVR organized phone service for decades, but it forces the customer to adapt to the company's menu, and not the other way around.

Voice AI flips the logic. The customer says what they need in their own words, and the AI understands the intent. This post shows how that works in practice, what Tolky delivers and how to adopt it with care.

Updated in November 2026.

What is a semantic IVR?

A semantic IVR is a voice service in which the customer speaks in natural language and the AI understands the intent, with no numeric menus. Instead of "for a duplicate invoice, press 2", the customer says "I need a copy of my invoice" and the conversation goes on from there.

The difference from the traditional IVR lies in what the system understands. A menu IVR only recognizes keys; a semantic IVR recognizes speech, with variations, accents and interruptions. What does this is an AI voice agent, which combines speech recognition, a language model and speech synthesis. For the base concept of an agent, see what an AI agent is.

How does voice AI differ from a traditional IVR?

Voice AI differs from a traditional IVR because it understands free speech, accepts interruption and resolves or transfers with context, instead of just routing by key. The table compares the two models.

AspectMenu IVR ("press 1")Semantic IVR (voice AI)
How the customer interactsKeys and fixed optionsNatural speech
Changing subject midwayBack to the startThe conversation continues
InterruptionMust hear the whole menuThe customer can interrupt the AI
ResolutionRoutes to a queueCan resolve and look up systems
Handoff to a personTransfers without contextTransfers with the reason and transcript
RecordAlmost noneTranscript in the customer history
MaintenanceRedesign the menu treeUpdate the knowledge base and the AI's scope

The AI is not magic: it needs a reviewed knowledge base and clear limits, just as in text. See how to train an AI agent on your knowledge base.

Where does Tolky's voice work?

Tolky's voice works in three places: WhatsApp calls, voice on the website and regular telephony. Telephony is available on activation, choosing the number and agent that answers.

  • WhatsApp: the customer taps the call button in the conversation and talks to the same AI that serves by text.
  • Web: voice straight in the browser, with nothing to install.
  • Telephony: the AI answers on the number you choose, on activation.

One point that often raises questions: for the company to call through WhatsApp, the contact must give permission. Tolky sends the permission request, shows the status in the conversation and, for those who did not authorize, regular telephony is the path. When the customer is the one calling, there is no such step. The voice service solution details each channel.

How is a call handled, end to end?

A call is handled in four steps: choosing channels, setting the voice, the conversation with the AI and the team stepping in when needed. The walkthrough below summarizes the flow.

  1. Choose the channels: WhatsApp, telephony and website voice, each number with its own agent.
  2. Give the AI a voice: choose the voice, regional accent, opening greeting and answer length.
  3. The AI converses and resolves: it listens, understands, looks up systems and answers, accepting interruptions.
  4. The team steps in when needed: the AI transfers to an agent, and the full transcript stays in the history.

There are two operating modes, depending on the call. In cascade mode, the platform transcribes, thinks and synthesizes the voice. In real-time mode, the AI listens and speaks at the same time, for more fluid conversations. Before publishing, you choose the voice with a preview.

What does real-time transcription change?

Real-time transcription converts speech into text while the person is still speaking, and that changes three things: the AI understands the call, the team follows live and the history becomes searchable. At Tolky, the transcript follows the language of the speaker and sits in the contact history, alongside text conversations.

During a call, the team sees calls in progress, filters by AI or person and can listen in. When an operator joins the audio, the AI goes silent. If the owner does not answer, the call can ring other operators. This unified history is what separates an IVR from a contact center that knows the customer: what was said on the phone shows up on WhatsApp, and vice versa. See also conversational memory.

When does it make sense to swap "press 1" for voice AI?

It makes sense when call volume does not fit the team at peak hours and a good share of calls are repetitive, with answers the AI can give. The table shows common scenarios.

ScenarioWhat the AI can doWhen to hand off to a person
Contact center with call peaksAnswer the first call and resolve the repetitiveOut-of-scope or sensitive cases
Clinics and scheduled servicesConfirm and reschedule by voice, with no queueSituations needing human judgment
Companies that live on WhatsAppTake the WhatsApp call with the same AI as textWhen the customer asks for a person
After-hours serviceAnswer and log for the teamCallback on the next business day

For handoff design, see AI to human handoff and the term handoff in the glossary. For the text channel, see AI contact center.

A practical tip for the first weeks: keep a visible option to reach a person at any time, and review the calls where the customer asked for one. Those calls show which topics the AI still cannot cover and are the best source for the next update of the knowledge base.

What about voice campaigns?

Voice campaigns are outbound calls: the AI calls your list, follows a script and pursues an objective, through WhatsApp or telephony. You write the objective, generate the script with AI, rehearse the conversation and choose channel, hours and attempts.

A few controls matter: time windows, retries with an interval, number protection with gradual calling and a do-not-call list, and transfer to a person during the call. Every call begins with the transcription notice, which cannot be turned off. Before starting, Tolky estimates the credits, and you set a daily spending cap. Two items stay out of this page: A/B script testing and scheduling activities, which are coming soon or still simulated. See voice campaigns.

How do you measure whether voice AI is working?

Voice AI is working when repetitive calls end resolved, handoffs happen for the right reasons and the customer does not have to repeat what they already said. Measure this with a few indicators, before and after publishing.

IndicatorWhat it showsHow to track
Calls resolved by the AIHow much of the repetitive work was absorbedConversations closed without transfer
Reasons for transferWhere the AI or the base failsReading the transferred transcripts
Time to first responseWhether the wait decreasedCompare with the previous period
Transcription qualityWhether names and numbers are understoodWeekly sample of calls
Customer satisfactionHow callers perceived itNPS at the end of the service

Use the same criteria as for text: see AI resolution rate: how to measure it and AHT and first response SLA. Listen to a sample of calls every week at the start. The transcript in the history makes this review easier, since you can search for the passage instead of listening to everything.

What should you keep in mind before adopting voice AI?

Before adopting, consider six points: scope, knowledge base, handoff to the team, privacy law, WhatsApp permission and customer expectations. A careful rollout avoids the frustration of a new IVR that behaves like the old one.

  • Scope: start with a few frequent topics.
  • Reviewed base: the AI only answers well what is documented.
  • Clear handoff: the customer must be able to reach a person.
  • Notice and privacy: the call is transcribed, and the customer is told. Confirm with your legal team the rules that apply to your sector.
  • WhatsApp permission: respect the contact's authorization status.
  • Testing: listen to the AI in simulated calls before publishing.

Voice also needs honesty about limits: the AI introduces itself as an AI assistant, and the team remains part of the service. Tolky aggregates more than 5 million AI messages per month and more than 67% AI resolution, platform figures, not a promise for your voice operation.

Frequently asked questions

Does voice AI work on my phone number?

Yes, on telephony, on activation: you choose the number and the agent that answers. It is also possible to answer calls on WhatsApp and through voice on the website.

Can the customer interrupt the AI?

Yes. The customer can interrupt the AI, as they would a person, and the transcript follows the language of the speaker.

Can a person join the call?

Yes. The operator joins the audio, and the AI stays silent while they speak. The AI can also transfer to an agent when the conversation calls for it.

Is the call recorded?

The transcript is saved in the contact history, alongside text conversations. During the call, the team follows the live transcript and can listen in.

Do I need permission to call through WhatsApp?

For the company to call, yes: WhatsApp requires the contact's permission. Tolky sends the request and shows the status in the conversation, respecting WhatsApp's limits.

Does voice AI replace the contact center team?

No. It handles the repetitive calls and transfers what needs judgment, with the transcript. The goal is AI with humans.


Conclusion

A semantic IVR swaps the menu for a conversation, and Tolky's voice AI brings it to WhatsApp, the web and telephony on activation, with the transcript in the history and the team available when the customer needs it. Shall we talk?

See the plans, explore voice service or talk to the team →

Share

Tags

  • voice AI
  • semantic IVR
  • AI voice agent
  • voice customer service with AI
  • #conversational AI CRM
  • IVR with artificial intelligence
  • WhatsApp calls with AI
  • real-time transcription

About the author

Marlos Carmo

Founder of Tolky

Marlos Carmo is an AI entrepreneur and founder of Tolky, the conversational-era infrastructure and AI CRM that unifies intelligent service, multi-channel support (such as WhatsApp and voice), live CRM, and operational intelligence in a single ecosystem. He is a finalist for the SXSW Innovation Awards and a member of Francesco's Economy, a global network of young entrepreneurs focused on innovation and social impact. He works connecting Artificial Intelligence and digital transformation in projects for large organizations.

Next step

Want to bring this to your operation?

Tati, Tolky's AI assistant, understands your scenario in a few questions, suggests the best path and connects you with someone from the team, with full context.

Read also