Think of the last ten calls that came in while nobody could pick up. Lunch break, a wave of sick days, after hours, a spike at half past nine in the morning. Some of them were routine: opening hours, a status check, "can you call me back?". But one or two were real enquiries with a budget behind them. And by the third ring, those had gone somewhere else. This is exactly where a voicebot comes in: it picks up when your team can't, understands the request and either handles it itself or passes it on, neatly pre-qualified.
The good news: in 2026 the technology has grown up. A modern voicebot no longer sounds like the phone menu of 2015 ("For accounts, press three"); it holds a real conversation. The slightly less comfortable news: providers, pricing models and data protection obligations vary a lot, and that is exactly what decides whether the project pays off or turns into an expensive experiment. This overview sorts out the three questions that come up in every first conversation: Which providers are there, what does it cost, and what do I need to settle on data protection?
What a voicebot is — and what it isn't
A voicebot is software that understands spoken language on the phone (or in an app) and replies in natural language. The caller states their request, the bot listens, recognises the intent and responds. At its best, so smoothly that the conversation doesn't feel like filling in a form.
Technically, three building blocks work together in real time: speech-to-text turns what is said into text, a language model (LLM) understands the intent and phrases the answer, and text-to-speech speaks it back. The decisive quality factor is not the voice but the latency. Good systems reply in under 800 milliseconds from the moment you stop talking. Otherwise you get those awkward pauses where both sides start speaking at once. This is achieved through streaming: the bot transcribes while you are still talking and starts speaking before the full reply has been generated.
Important for setting expectations: a voicebot is not a chatbot with a voice. A chatbot on a website has time; the user reads and types. On the phone every second counts, misunderstandings can't simply be "read again", and background noise, dialects and interruptions come on top. Voice is the more demanding channel. That is exactly why it pays to look closely at the providers instead of simply picking the cheapest tool.
AI phone assistant or voicebot – what's the difference?
Both terms describe the same class of technology: software that understands spoken language, recognises the intent behind it and replies in natural language. "Voicebot" is the umbrella term and also covers voice assistants in apps, on devices or in cars. An AI phone assistant is the variant built for telephony: it is connected to your phone number, takes incoming calls, books appointments, pre-qualifies leads and hands over to a human as soon as the request calls for it.
In practice, then, the two differ less in technology than in scope. An AI phone assistant comes with the connection to your phone system, forwarding to extensions and handover to your CRM or calendar out of the box; a general voicebot has to be configured or integrated for that first. If you hear both terms in conversations with providers, don't ask about the label; ask about exactly these points: which phone numbers, which handovers, which systems.
For data protection the label makes no difference: as soon as calls are processed, the obligations described below – from telling callers that an AI is speaking to the data processing agreement with the provider – apply to both equally.
Where voicebots really deliver in B2B
The most common mistake is to see a voicebot as a replacement for the whole team. A more realistic and more profitable role is that of a first-level filter: it absorbs the volume that doesn't need a human and passes on the rest with added context. Four use cases that regularly pay off in B2B:
- Call answering around the clock. No more missed calls outside core hours. The bot records requests, answers recurring standard questions and makes sure nobody hits a dead end at eight in the evening.
- Lead pre-qualification. Incoming enquiries are captured in a structured way (request, company size, urgency) and land cleanly in the CRM before a sales rep even calls back. That saves the first two minutes of every conversation.
- Appointment booking. The bot checks calendars and books the slot itself. Especially valuable in industries with high call volumes and plannable slots.
- Structured data capture. Meter readings, status checks, address changes, damage reports: anything where the caller provides information that a person would otherwise type in.
The configuration differs noticeably between B2B and B2C. In B2C a lot revolves around order status, returns and high volumes. In B2B it is more about qualification, scheduling and a clean handover to a named contact person. So if you introduce a voicebot, don't ask "can the system talk?" but "does it fit my three most common reasons for calls?".
Voicebot providers at a glance
There is no single best voicebot. There is the one that fits your company size and your use case. The German-speaking market can be roughly divided into three groups, plus a fourth option that doesn't appear in any provider list.
| Segment | Example providers | Typical fit |
|---|---|---|
| Enterprise / corporate | Cognigy (part of NiCE since 09/2025), Parloa | High volume, many systems, a dedicated team for maintenance; often multilingual and deeply integrated into the contact centre |
| Mid-sized companies / SMEs | Vitas, fonio, voiceOne | Fast setup, preconfigured industry templates, predictable monthly costs |
| Industry specialists | Aaron.ai (healthcare) | In-depth templates for one industry, less configuration effort, but less flexible |
| Custom solution | Tailor-made voice agent | Very specific processes, tight connection to existing systems and automations |
The table gives a rough map of the field – and the field keeps shifting: in September 2025 the US group NiCE acquired the Düsseldorf-based enterprise provider Cognigy for around 955 million US dollars. A clear signal of how strategic voice AI in customer service has become. The actual selection logic cuts across all of this, though: it depends less on a provider ranking than on how closely the bot has to connect to your existing systems. An SME tool with an industry template is live within a week and easily covers call answering and appointment booking. But as soon as the bot needs to look into the ERP live, map unusual processes or work together with existing automations, the calculation shifts towards an enterprise platform or a tailor-made custom solution. We cover exactly this fourth option further down.
AI phone assistant: what a voicebot costs in 2026
This is where it gets concrete and confusing, because at least three pricing models exist side by side. Anyone comparing quotes is often comparing apples with oranges. The three building blocks to watch:
| Cost block | Typical range (guide) | What matters |
|---|---|---|
| One-off setup / configuration | approx. €2,000 – €15,000 | Complexity of the requests, number of system integrations, languages |
| Monthly licence / operation | from approx. €500, enterprise considerably more | Feature set, parallel lines, support level |
| Usage per minute | approx. €0.12 – €0.55 | Call volume; some plans include free minutes |
Lean SaaS tools for SMEs sometimes start at just a few hundred euros a month, while a comprehensive enterprise solution with deep integration can reach five figures in the first year. Typical industry estimates put the total for the first year at around €10,000 to €50,000, depending on complexity. These figures are a guide, not a fixed price: a bot that only answers calls and notes callback requests costs a fraction of one that looks into the ERP live and triggers orders.
A more honest yardstick than the list price is the counter-calculation: what does a missed qualified call cost you today? If a single won order per month covers the licence, price is a secondary question. If your volume is low and most calls need a human anyway, a voicebot can quickly cost more than the status quo. Do the maths with your real numbers, not with the ROI promises from provider brochures.
Voicebots and data protection: what you need to settle
Sounds like red tape? It isn't, if you clarify four things up front instead of after go-live. As soon as a voicebot takes calls, it processes personal data, and the GDPR applies in full. Voicebot data protection is therefore not a box you tick at the end but a selection criterion from the start. These four levers decide:
Transparency and consent. Callers must be told clearly at the start that they are talking to a bot and that voice data is being processed. As soon as you record, store or use conversations for training, you generally need consent. Silent recording is not an option. And since this summer it is no longer just a GDPR question: from 2 August 2026, Article 50 of the EU AI Act explicitly requires that people are informed when they are interacting with an AI – unless that is obvious from the context. Violations can be fined up to 15 million euros or 3 per cent of worldwide annual turnover. Important for your planning: this transparency obligation applies on schedule, unlike the high-risk obligations, whose start the Digital Omnibus has postponed to December 2027. What else the EU AI Act means for SMEs is covered in our EU AI Act overview.
Voice recordings are sensitive. A recorded voice is personal data. If its acoustic features are used in a way that uniquely identifies a person, it can even count as biometric data, which is subject to stricter requirements. The pragmatic consequence: record only what you really need, and set deletion periods from the start.
Data processing agreement (DPA). In the vast majority of cases the voicebot comes from an external service provider. Then you need a data processing agreement that covers the subject matter, types of data, technical measures, sub-processors and deletion. Ask for it early; a reputable provider will hand it over without being asked.
Technical measures and hosting. Encryption of voice data, access controls, logging, backups. For many companies the server location is also decisive: hosting in Germany or the EU makes compliance considerably easier and is often mandatory in regulated industries. Ask specifically where the data is stored and which language model runs in the background.
If you operate in particularly sensitive areas (healthcare, finance, legal advice), you should also ask the provider about industry-specific experience. Of course, this overview does not replace a detailed legal assessment; when in doubt, discuss the specific setup with your own data protection officer.
How to approach the rollout
The road to a productive voicebot is shorter if you start small. Pick the one use case with the highest volume and the clearest process, usually call answering with callback capture, and automate only that first. A narrowly scoped bot that does one thing reliably beats the all-rounder that works 80 % of the time everywhere and, for exactly that reason, is trusted by no one.
The bigger strategic question is standard SaaS versus a custom solution. An off-the-shelf tool goes live quickly and is cheap to start with, but hits its limits as soon as the bot needs to reach deep into your systems, map unusual processes or work together with existing automations. If you already rely on process automation with n8n & co., you can connect the voice agent directly to your existing workflows. Then the phone channel doesn't become an isolated tool but a building block in the same system. This bridge between phone, CRM and back end is exactly where tailor-made voicebot development pays off compared with an off-the-shelf product.
Whichever route you choose: define up front how you will measure success. Reachability rate, automatically resolved calls, number of qualified leads. Without a target metric, you won't be able to say after three months whether the investment pays for itself. And that brings us back to the beginning: the one or two real enquiries that today go elsewhere by the third ring are the whole reason for the first bot. If it does nothing more than reliably catch those, it has already paid off. A voicebot doesn't run itself, but for the right call reasons it is one of the most rewarding automation levers SMEs have in 2026.
Frequently asked questions
How much does an AI phone assistant cost for businesses?
The ranges from this article apply here too: a one-off setup of typically around €2,000 to €15,000, a monthly licence from about €500 and usage prices of around €0.12 to €0.55 per minute; in the first year most projects land between €10,000 and €50,000. Exactly where you end up depends on call volume, the number of system integrations and the languages you need.
Which voicebot providers are suitable for SMEs?
For SMEs, tools such as Vitas, fonio or voiceOne with preconfigured industry templates and predictable monthly costs are a good fit. Enterprise platforms such as Cognigy or Parloa are aimed at high volumes with a dedicated maintenance team, industry specialists such as Aaron.ai at individual sectors. If the bot needs to reach deep into the ERP or into existing automations, a tailor-made custom solution is the fourth option. What matters is not the ranking but how tightly it has to integrate with your systems.
Want to connect your phone channel directly to your CRM, calendar and workflows? Then have a voice AI phone assistant built – we build the voice agent around your most common reasons for calls.
As of July 2026. The provider landscape and price ranges change constantly; the transparency obligations under Art. 50 of the AI Act apply from 2 August 2026.