Insights
What is Voice AI? Evolution, Technology, Added Value
8 min.Reading time
Voice AI is the evolution of classic voice dialogue systems with generative AI. Where rigid touch-tone menus used to guide callers through tree structures, modern Voice AI understands natural language, answers questions from a knowledge base, and executes processes autonomously. This article explains the evolution from IVR to Voice AI, the four technology building blocks, and why Voice AI is a no-brainer for businesses today.
"Press 1 for sales. Press 2 for support. Press 3 for..."
Everyone knows this voice from the phone. And hardly anyone likes to hear it. Anyone who calls and ends up in a touch-tone menu is frustrated before the conversation has even begun.
Voice AI is the answer to this frustration. It is the evolution of classic voice portal systems using generative AI. Where previous systems guided callers through rigid tree structures, modern Voice AI understands natural language, answers questions from a knowledge base, and executes processes autonomously. By 2027, 50% of all service requests are expected to be resolved by AI, up from 30% in 2025 (Salesforce State of Service, September 2025).
This article explains what Voice AI is, how it works, and why it is a no-brainer for businesses today.
1. From touch-tone to natural language
The history of voice technology is longer than many think. As early as 1911, a toy dog named Rex jumped out of his kennel when his name was called. That was the first recorded example of automatic speech recognition. From the 1970s onwards, touch-tone IVR systems shaped customer service on the phone. Speech recognition, if used at all, was limited to keywords and digit strings. The systems were clumsy and frustrated callers.
The next stage began in the early 1990s. Researchers at the Erlangen Chair for Pattern Recognition developed Evar and FränKi, the world's first natural-language dialogue systems. Evar was an information system for Deutsche Bahn. FränKi knew which cinema was showing the latest James Bond. Out of this team, Sympalog Voice Solutions GmbH emerged in 2003, which continues to develop voice portal systems to this day.
The breakthrough for acceptance did not come from customer service, but from everyday life. Siri, Alexa, and Google Assistant made conversations with a computer a matter of course. At the end of 2022, OpenAI released ChatGPT. By mid-2025, the voice experiments had turned into widespread products: Amazon launched Alexa+, Apple showed a Siri with goal-oriented conversation flow, and Google released Gemini Live for real-time voice conversations.
Three terms are important for differentiation:
IVR System
An IVR (Interactive Voice Response) system is a voice menu that guides callers through a fixed dialogue tree via keypress or simple voice commands. Touch-tone IVRs are the oldest form and are still widely used today.
Voice Portal System
A voice portal system (also called a voicebot or dialogue system) is software with which people communicate with a computer using spoken language. Modern systems rely on AI and understand full sentences, ask follow-up questions, and react flexibly to the caller's concern.
Voice AI
Voice AI is the current generation of voice portal systems that utilizes generative AI and Large Language Models to conduct natural conversations, generate answers from a knowledge base, and execute processes autonomously. The use of LLMs and the ability to operate without predefined dialogue trees are the crucial characteristics.
2. How Voice AI works: The technology behind the telephone
A modern Voice AI system consists of four technology components that work together seamlessly.
Speech-to-Text (STT): Converting speech to text. The first step is speech recognition. Modern speech recognition systems like Whisper belong to the class of Automatic Speech Recognition (ASR) models. They are extremely error-tolerant, multilingual, and recognize speech even with highly varying audio quality, dialects, or background noise. Speech recognition converts speech into text. It does not yet understand the meaning behind the statements.
Natural Language Understanding (NLU): Recognizing the intent. The second step is understanding. Put simply, NLP (Natural Language Processing) stands for searching for words or phrases in a question. However, as soon as questions become more complicated, NLU is essential. The AI looks at the entire question, checking the subject, object, pronouns, and action. This allows the system to correctly classify queries like "I want to exchange the goods. But I have already unpacked them."
Enterprise Search and Knowledge Base: Finding the answer. The third step is knowledge retrieval. The system searches a knowledge base for the appropriate answer: product database, FAQ system, CMS, CRM, DMS, or a combination thereof. This is where the quality of any Voice AI is determined. A Voice AI is only as good as the knowledge base it accesses.
Text-to-Speech (TTS): Speaking the answer. The fourth step is speech output. Modern TTS voices sound so natural today that they are suitable for productive use. Cloud-based TTS solutions have become affordable despite their high quality.
Latency as a critical factor. On the phone, any pause is fatal. Modern real-time voice systems achieve an end-to-end latency of under 300 milliseconds. The Red Phone, a London-based provider of natural language routing, operates with under 600 ms latency at 99.9% uptime in over 30 languages.
The crucial difference to the past: Previously, dialogues and grammars had to be maintained manually. Today, small training sets or zero-shot classification with LLMs are sufficient. Large Language Models like GPT, Claude, or Llama guide dialogues dynamically based on the conversation context, without predefined dialogue trees.
3. Why Voice AI is a no-brainer today
Three developments make Voice AI absolutely attractive for companies today.
The technology is mature. Speech recognition reaches 90% accuracy. Text-to-speech voices sound natural. Latency is under 300 milliseconds. 89% of service professionals say that conversational AI increases the self-service resolution rate (Salesforce State of Service, September 2025). The technology is proven in productive use worldwide.
The costs are low. A voice portal system does not need thousands of daily calls to be profitable. Newcall, an AI telephone assistant for tourist information offices in the DACH region, is available starting at €650 per month (new-call.de, 2026). Fonio, the volume player in the German market with over 9,500 customers, can be set up in minutes (OMR, 2026). What used to be a major IT project is now a SaaS subscription.
The added values are real. Voice AI offers 24/7 availability, multilinguality, scaling during call peaks, and insights from call data. Newcall answers guest inquiries in over 25 languages. The Red Phone replaces rigid IVR menus with natural language routing in over 30 languages. 70% of travelers would skip the reception during hotel check-in. Among Gen Z, this figure is as high as 82% (Mews Survey, June 2025). Guests want information immediately, in their language, and without waiting.
Those who still operate touch-tone IVR today are losing callers. The question is not if, but when and how you introduce Voice AI.
4. Where Voice AI is already working today
Voice AI can be used across all industries. Wherever standardized processes and connected systems exist, it can handle requests independently.
Industry | Specific Requests | Added Value |
|---|---|---|
Tourism & Transport | Opening hours, event information, reservations, travel info | 24/7 multilingual guest support, relieving tourist info in peak season |
Financial Services | Account balance, transaction inquiries, transfers, branch info, PIN management | Self-service for standard inquiries, authentication by voice |
Energy Utilities | Meter readings, outage reports, tariff changes, billing questions | Buffer during call peaks caused by outages |
E-Commerce & Retail | Order status, returns, Click & Collect, product availability | Support for completing purchases over the phone |
Administration & Public Hotlines | Appointment scheduling, information, form requests | Round-the-clock availability without staff expansion |
Healthcare | Appointment scheduling, prescription requests, follow-up information | Relief for practice staff |
A concrete example from tourism: A guest inquiry at 11 PM asking for the nearest pharmacy. No tourist information office is staffed at this hour. A Voice AI connected to the destination data answers the question immediately, in the guest's language. Newcall does exactly that, with a go-live within one week (new-call.de, 2026).
Voice AI closes the gap because it meets the caller right where they already are: on the phone.
5. Voice AI needs a central knowledge base
Voice AI on the phone only works as well as the knowledge base it accesses. The best language model will not deliver useful answers if the company data is incomplete, outdated, or unstructured.
branchly does not offer its own Voice AI. branchly is the central knowledge base and dashboard that connects Voice AI solutions like Pendura, The Red Phone, Fonio, or Newcall with the website AI. The voice technology comes from the partner, the database and the insights come from branchly. All four providers are direct partners of branchly and can be connected natively to the branchly platform.
The principle is simple: You set up your content once in branchly as a knowledge base. Website texts, FAQs, product data, PDFs, event calendars, external databases. This knowledge base then applies to every module: the chat on the website, the hybrid search, the Advisor, the Navigator, the Forms, and the Voice Agent on the phone.
What changes when all interfaces come from a single source:
No conflicting info. Phone and website feed from the same knowledge base. A customer gets the same answer on the phone as in the chat.
No double maintenance effort. You maintain one knowledge base, not two. Every update is instantly active across all channels.
One compliance standard. GDPR, EU AI Act, and WCAG/BFSG are solved once, not per channel. One platform, one DPA, one audit trail. Microsoft Azure EU hosting for LLMs, OVH Cloud for user data. Nothing leaves the EU.
One dashboard. All voice and web insights in one interface. Call topics turn into FAQ candidates. Frequent questions turn into content. Data turns into recommendations.
branchly has processed over 40 million AI-powered sessions and natively supports 101 languages. That is the difference between "buying Voice AI" and "strategically implementing Voice AI".
How to properly implement Voice AI in five steps is described in the article How to properly implement Voice AI in 5 steps. Why isolated Voice AI is not enough is covered in Why isolated Voice AI is not enough.
Summary
Voice AI is the evolution of classic voice portal systems using generative AI. The evolution led from touch-tone IVR systems of the 1970s through the first natural-language dialogue systems of the 1990s to today's systems that understand natural language, generate answers from a knowledge base, and execute processes autonomously.
The technology is mature, the costs are low, the added values are real: 24/7 availability, multilinguality, scaling during call peaks, and insights from call data.
The crucial realization: Voice AI on the phone only works as well as the knowledge base it accesses. branchly does not offer its own Voice AI, but is the central knowledge base that connects Voice AI partners with the website AI. One database for six modules. One platform, one dashboard, one compliance standard.
Those who still operate touch-tone IVR today are losing callers. Those who implement Voice AI strategically gain data that drives their marketing forward.
