A voicebot, also known as an AI phone assistant, answers customer service calls, understands inquiries in natural spoken language, and resolves them using the company's knowledge base. It captures recurring processes such as meter readings, address changes, or cancellations through dialogue and forwards them to the appropriate department. It connects to the existing phone system, ideally as an additional channel within the same system that powers the website chat. This guide explains how this works technically, where a voicebot replaces touch-tone menus, and which seven questions you should use to compare providers.
At a Glance
Answering questions, capturing inquiries, triggering processes: These are the three core tasks of a customer service voicebot.
Natural language replaces touch-tone menus. Callers can still use the keypad to type digits, such as customer numbers, if they prefer.
Seamless connection to any SIP-enabled phone system, without replacing existing infrastructure. Routing remains fully under the company's control.
A single knowledge base for chat and phone eliminates duplicate maintenance.
Response time and fact-checking go hand in hand. Fast is good; verified is essential.
Billing is typically based on minutes, plus a booked capacity for concurrent calls.
What a Voicebot Delivers in Customer Service
A voicebot connects conversational AI to the phone system. Callers explain their issue in their own words; the voicebot identifies the intent and responds in spoken language. In customer service, it handles four key tasks.
Answering questions. Rates, deadlines, opening hours, documentation: The voicebot answers using the company's knowledge sources, such as websites, documents, and FAQs. Over the phone, it formulates shorter responses than in a chat, asks clarifying questions, and guides callers step by step.
Capturing inquiries. For structured processes, a dialogue guides callers through the necessary information. For a meter reading, this includes customer number, metering point, and reading; for a relocation, the new address, move-in date, and handover meter reading. The voicebot validates the plausibility of the data before processing it.
Triggering processes. Once the inquiry is captured, a workflow is triggered—such as sending an email with all the details to the relevant department or calling an API. Callers receive confirmation during the same conversation.
Routing to human agents. If the caller requests a human or if the inquiry requires one, the voicebot announces the transfer and connects the call. The receiving agent hears a brief summary including name, inquiry, and details beforehand, eliminating the need to ask again from scratch. Outside business hours or when lines are busy, the voicebot records the inquiry itself.
Voicebot Instead of Touch-Tone Menus
Traditional IVR systems rely on digits: "For sales, press 1." Callers who cannot find their issue within the menu structure end up in the wrong queue. A voicebot replaces the menu with an open question: "How can I help you?"
The keypad does not disappear entirely. For entering numbers—such as a ten-digit customer number—it is often faster and more accurate than speech. A robust voicebot therefore understands both speech and touch-tones (DTMF) and can also replicate a classic menu where needed.
Connecting to the Phone System
A voicebot does not require a new phone system. It is configured in the existing PBX like an additional extension and connects via SIP, the standard for internet-based telephony. This covers virtually all modern phone systems and carrier SIP trunks. The administrator responsible for the phone system simply provides the access credentials, without triggering an extensive IT project.
Call routing remains entirely in the hands of the company. Three setups are common:
Bot first: The voicebot directly answers every call on a designated number.
Bot as overflow: It only steps in if, for example, reception does not pick up within ten seconds.
Bot for specific areas: It serves only selected phone numbers, specific hours, or menu options, such as an out-of-hours emergency hotline.
Call Quality: Latency, Interruptions, Pronunciation
Over the phone, every pause is noticeable. Three factors determine whether a conversation feels natural.
Response time. High-performing systems respond in approximately half a second. More important than chasing the last millisecond, however, is ensuring the answer is fact-checked. An incorrect answer over the phone provides no visible source for the caller to cross-check. A proven approach is division of labor: A fast language model keeps the conversation going with greetings, clarifying questions, and confirmations, while the actual answer is retrieved and verified in the background. Short bridging phrases like "Let me look that up for you" prevent dead air.
Interruptions. People interrupt one another. A voicebot that supports mid-sentence interruptions (barge-in) and responds instantly to "Excuse me?" feels natural. Noise suppression and voice activity detection isolate the caller's voice from background noise, ranging from train station announcements to workshop environments.
Pronunciation. Brands, product names, and acronyms only sound natural if the company can define them. A pronunciation lexicon replaces terms with phonetic forms—turning "kWh" into "kilowatt hours"—or defines specific phonetic spelling.
One Bot for Chat and Phone
Many companies already operate a website chatbot. Implementing a separate voicebot product with its own distinct knowledge base doubles maintenance efforts: Every tariff change must be updated in two places, risking divergent answers.
A more sustainable approach is deploying a voicebot as an additional channel within the same system. Knowledge, dialogues, and workflows are maintained once and apply to both chat and phone. Only the output format adapts: links and buttons for chat, concise spoken sentences for voice.
Seven Questions to Evaluate Providers
Knowledge base: Does the voicebot use the same knowledge base as the chat, or is it maintained separately?
Integration: Does it work with your SIP phone system, and who provides which credentials?
Response and verification: How fast does it respond, and what sources back up the facts in its answers?
Processes: Which workflows can it handle via dialogue, and how does it push data into your systems?
Human handover: When and how does it transfer calls to staff, and what context is passed along?
Data privacy: Where are speech recognition and speech synthesis processed, and can call recording be toggled? For details, see the article on GDPR-compliant AI phone assistants.
Operations and costs: What is the billing model, and can you monitor utilization, call history, and topics in the analytics dashboard?
How Mercury Voice Powers the Voice Channel
Mercury Voice makes an existing Mercury.ai bot accessible over the phone. Adding a new voice channel is all it takes—knowledge, conversational dialogues, and workflows automatically apply to phone calls. A single project can support multiple voice channels, such as general customer service and a technical hotline, each with its own voice and greeting.
Integration works via SIP with any SIP-compliant phone system. Mercury Voice understands speech and touch-tones, supports barge-in, and typically responds in around 500 milliseconds (or up to one second for complex queries). The answer is generated using the same retrieval and fact-checking mechanisms as the chat. Voices are powered by Amazon Polly and optionally Cartesia, while a pronunciation lexicon defines how brand and technical terms are spoken. The service launches with European languages.
Calls appear in the analytics dashboard with KPIs, call history, and summaries as a dedicated channel alongside chat. Pricing is based on call minutes and a booked volume of concurrent calls. Learn more on the Mercury Voice overview page.
When an inquiry requires a human agent, Mercury Voice transfers to a designated phone number, complete with an announcement to the caller, a short briefing for the agent, and support for schedule-based routing by weekday. Outside these hours or during line busy states, Mercury Voice captures the inquiry autonomously.
Frequently Asked Questions
What is the difference between a voicebot and an IVR?
An IVR guides callers through rigid touch-tone menus and pre-recorded prompts. A voicebot understands natural spoken language, holds a dynamic conversation, and routes callers directly to an answer or process without complex menu layers.
What is an AI phone assistant?
An AI phone assistant is a voicebot that handles phone calls using artificial intelligence. It answers questions based on company knowledge, collects inquiry details through natural conversation, and triggers backend workflows.
Do I need to replace my phone system?
No. A voicebot like Mercury Voice is simply configured as an additional extension within any SIP-enabled phone system. You only need the SIP credentials of your PBX.
How fast does a voicebot respond?
Mercury Voice typically responds in around 500 milliseconds, and up to one second for complex queries. A fast language model maintains conversational flow while the actual answer is validated in the background.
Which languages does Mercury Voice support?
Mercury Voice launches with European languages, including German (with Austrian and Swiss variants), English, French, Spanish, and Italian. Additional languages are enabled following quality assurance.
How is a voicebot billed?
Billing is typically based on minutes of conversation. Mercury Voice adds a booked capacity for concurrent calls, with clear visibility into utilization and peak loads in the dashboard.
Conclusion
A voicebot offloads routine calls from your service team and keeps your hotline available even during peak demand. Its success depends on three pillars: a shared knowledge base with your chat, an integration that leaves your phone system intact, and answers that are both rapid and fact-checked. Use the seven questions above to directly evaluate providers during product demos.
Looking to automate your hotline? Talk to our team and experience Mercury Voice in a live demo.
About the Author
Dr. Hendrik ter Horst
Chief Product Officer @Mercury.ai
Dr. Hendrik ter Horst is Chief Product Officer at Mercury.ai. He holds a PhD in Computer Science from the CITEC Institute at Bielefeld University, with a focus on machine information extraction and data processing. For ten years, he has been working on dialogue-based AI systems, overseeing the conception, implementation, and validation of AI-powered applications from initial idea to production deployment.
