Many B2B companies still lose expensive leads on the phone because no one is available or callbacks come too late. Artificial intelligence or a voice bot for sales closes this gap in real time: it immediately accepts customer inquiries, qualifies them, and transfers them cleanly to CRM or sales. In this interview, Martin Hartig, responsible for the development of the matelso AI Sales Agent, explains why this approach is so relevant for SMEs, what technical hurdles there were, and how to use AI and intelligent voice assistants in a way that really generates more deals – and in GDPR compliance.
Summary
- Why are seamlessly integrated AI voicebots or chatbots necessary in sales today?
- Why does sales need an AI voicebot today—what specific use cases does it solve?
- As a voicebot provider, how do you measure whether such bots really work in sales and evaluate calls automatically and efficiently?
- What does “AI-first, human handover” mean in your voicebot—and why were you able to reach a competitive level so quickly in terms of automation and AI architecture?
- Frequently asked question: How does conversational AI work at your company, and what can the matelso voicebot already do in terms of calls and automation?
- How deeply integrated is the AI voicebot in the matelso platform—especially for use cases in the areas of GA4, CRM, and reporting?
- How do you ensure that the bot not only chats nicely, but also sounds natural and works effectively in sales?
- How do you keep the complex issue of voice latency in spoken dialogue below the target value (including barge-in) and prevent hallucinations in live calls—what mechanisms and monitoring KPIs do you use?
- Many SME customers ask first about powerful data protection. What can you say about your AI sales agent in this regard?
- What insights have you gained from the development process—and what’s next?
Why are seamlessly integrated AI voicebots or chatbots necessary in sales today?
Implementing an AI voice bot in sales efficiently and seamlessly closes the biggest gap between marketing and sales: the moment of initial telephone contact. Those who respond more quickly and provide appropriate answers will gain more from their paid leads.
Why does sales need an AI voicebot today—what specific use cases does it solve?
Especially in B2B and particularly with decentralized sales structures—headquarters does marketing, branches are supposed to call back—too many leads are still being lost. Calls end up nowhere, employees are busy, call forwarding is not properly regulated. This is expensive because the leads were hard-won and cost a lot of money through PPC campaigns, websites, and other digital channels. Our approach at matelso platform is: no lead should be lost. Until now, we were able to route and forward phone calls internally, forward them via our telephone system, or transfer them to the matelso ServiceCenter – now the voice bot has been added, which answers the call immediately and thus relieves the sales team. This allows it to capture the moment when the intention to buy is at its highest. For sales, this means more contacts reached, more appointments, and less wastage. And you can see in the market that many providers are positioning the voice assistant for exactly this purpose: lead qualification and appointment booking.
As a voicebot provider, how do you measure whether such bots really work in sales and evaluate calls automatically and efficiently?
Only when bot data ends up in CRM or GA4 does the added value for marketing and sales become apparent. AI alone is not enough—it must be measurable.
We have two levels for this:
- Hard sales KPIs: Availability, qualified call rate, appointment rate, and ultimately conversions—linked to CRM or GA4 so that marketing can see that its leads are being followed up on.
- Bot KPIs: Does the bot understand the caller? Does it classify correctly? Does the conversation lead to an appointment? Ultimately, this is the “success rate per call.” Customer acquisition costs (CACs) and average handling time (AHT) also play a role: if the bot qualifies faster and humans only take on the valuable cases, acquisition costs and processing time decrease.
What does “AI-first, human handover” mean in your voicebot—and why were you able to reach a competitive level so quickly in terms of automation and AI architecture?
“AI-first” means that the bot always goes first and tries to resolve the issue—making appointments, collecting data, understanding needs. This is particularly relevant during periods of high call volume.
“Human handover” means that the customer is always transferred to a human agent, either automatically in the case of complex issues or directly upon request. With us, the caller can even switch agents at the touch of a button. These clear escape routes maintain high call quality, increase process efficiency, and improve response times.
We were quick to succeed because we didn’t want to build an all-purpose tool. We solved a specific sales problem: immediately picking up incoming leads, asking structured questions, and handing them over cleanly. Less configuration effort on the part of the customer means more quality in our hands. Other players also cite this focus on sales use cases as a success factor.
Frequently asked question: How does conversational AI work at your company, and what can the matelso voicebot already do in terms of calls and automation?
Lead tracking with the matelso platform – Book a live demo
More leads, less waste coverage, lower costs—lead tracking makes digital marketing more efficient and profitable! Benefit from up to 30% lower CPLs and 40% less waste coverage!
How deeply integrated is the AI voicebot in the matelso platform—especially for use cases in the areas of GA4, CRM, and reporting?
Completely. The bot is not a foreign body, but another “agent” in the communication flow. Just as conversations from the matelso ServiceCenter are already visible in the customer context today, bot conversations are also fully documented: who called, which campaign the contact came from, what was said, what the result was. We can send this data to GA4 or import it into the CRM as usual. This is powerful for sales managers because, for the first time, they can see the entire process:
Google Ads → Call → Bot qualifies → Appointment set.
This end-to-end view is exactly what modern sales and marketing setups require.
How do you ensure that the bot not only chats nicely, but also sounds natural and works effectively in sales?
A voice bot must be able to sell, not just talk. That’s why we manage roles, knowledge, and handover processes very closely.
We build in layers. At the bottom is the telephony stack. On top of that comes speech-to-text (STT) and text-to-speech (TTS)—in our case, Azure for input and ElevenLabs for output, directly linked to the media stream. Above that is a system prompt that we control: role, conversation flow, order of questions, appointment logic, termination. Then the bot only receives the knowledge sources it really needs – e.g., from a knowledge base or the CRM. This keeps it close to the context. We check the quality twice: automatically via logging and metrics – and manually via transcripts, which we have sent to us afterwards. Sales managers notice very quickly when a bot starts to ramble. That’s why we deliberately keep it sales-oriented.
How do you keep the complex issue of voice latency in spoken dialogue below the target value (including barge-in) and prevent hallucinations in live calls—what mechanisms and monitoring KPIs do you use?
With voice, latency determines naturalness. We aim for very short response times and stream the output so that something can be heard immediately. Filler phrases such as “one moment, I’ll check that quickly…” bridge the time while the LLM is still working. Barge-in, i.e., interruption by the caller, is an integral part of the setup. To combat hallucinations, we rely on strict prompts, RAG (responses only from permitted sources), and logging of response confidence. If the confidence level becomes too low, the call is transferred to a human. And we monitor this – so we can see if latencies are increasing or the recognition rate is decreasing.
Many SME customers ask first about powerful data protection. What can you say about your AI sales agent in this regard?
We keep as much as possible in our own infrastructure and separate very precisely which data is allowed to go where. It is not possible to do without third parties entirely, because STT/TTS services are simply more efficient at present. However, we log transparently, adhere to data minimization, and make conversations auditable. Important for companies: When AI speaks in sales, it must be possible to understand why it said something—especially when it comes to personal data. That’s why we build the monitoring and logging level in such detail.
A stable audio pipeline and low latency are the basis for any productive voice project. Only then is it worth fine-tuning AI, prompts, and dialogue design.
What insights have you gained from the development process—and what’s next?
- The real magic doesn’t happen in the LLM, but in the audio pipeline. If you don’t have telephony under control, you won’t get a real-time experience.
- Focus beats feature list. A bot that reliably answers calls, qualifies leads, and makes appointments brings more to sales than an assistant who can do “a little bit of everything.”
- Monitoring is a must because models change and you have to keep up with call quality.The next steps are clear: outbound scenarios, i.e., the bot also proactively calls leads, and omnichannel extensions. The market shows very clearly that sales teams want exactly that: automated qualification, automatic entry into the CRM—and only involving humans when it really becomes relevant for sales. Thank you for the interview, Martin.
The expert – Martin Hartig
Martin Hartig is Head of Solution Management & Consulting at software provider matelso. In this position, he is responsible for the overall development of the company’s SaaS solutions—from lead management and MarTech solutions to AI applications.
