An AI phone assistant answers calls around the clock, understands the request through speech recognition and resolves routine questions directly in the conversation. For online shops, this hits a sore spot: More than half of all classified support tickets revolve around order status, according to an analysis across 131 armincx merchants. These are exactly the calls an AI can take over. This guide shows you how the technology works, what it costs, which providers exist and where its limits lie.
What Is an AI Phone Assistant?
An AI phone assistant is software that answers incoming calls on its own, understands spoken requests and responds in natural language. To do this, it accesses your systems in real time, for example your shop system, CRM or ERP. It resolves simple cases within the conversation and hands over complex or emotional cases to your team, together with a transcript.
The difference to older phone technology is fundamental:
| Technology | What it does | Where it ends |
|---|---|---|
| Answering machine | Records messages | Every message creates a callback |
| IVR ("Press 1 for...") | Routes calls via menu | Rigid paths, long journeys to the goal |
| AI phone assistant | Understands and resolves requests | Needs training, data and clearly defined use cases |
Terms like voicebot, voice agent or AI voice assistant describe essentially the same technology. The differences lie in the details and in each provider's marketing.
How Does an AI Phone Assistant Work?
Three building blocks work together in every conversation. Speech recognition (ASR) converts spoken words into text. A language model understands the request, asks follow-up questions where needed and decides on the next step. Speech synthesis (TTS) turns the answer into a natural voice.
This classic pipeline creates pauses of one to two seconds per response. Newer speech-to-speech models process audio directly and react faster, but they are harder to control. Serious providers combine the technology with a human handover: If the AI detects frustration, a sensitive topic or a request outside its training, it connects the caller to a human and passes on the transcript as well.
Which Tasks Does an AI Phone Assistant Handle in E-Commerce?
The biggest lever lies in requests that repeat every day. In the armincx benchmark across 131 merchants, 54.6% of classified tickets fall under WISMO, meaning "Where is my order?". Measured against all tickets, it is 25.2%. The figure is directional, because only around 46% of tickets are assigned to a topic at all, and it applies across all channels, covering email, social and phone combined. The direction is still unmistakable: Order status is eating up your support.
On the phone, an AI phone assistant with a shop integration resolves two things above all:
- Order status: The caller states their name or order number, the AI pulls the live shipping status from the shop system and gives a concrete answer.
- Returns: The AI initiates the return directly in the conversation and hands the process over to the ERP.
Cancellations and address changes currently run through chat, email and WhatsApp at armincx. We tell you this so plainly because false expectations end a voice project faster than any technical question.
On top of that come product advice with CRM context, customer recognition via phone number and the handling of call peaks. During Black Friday, cross-channel total automation among armincx merchants rose to 35.2%, because workflows absorb the wave. Your phone line benefits from the same principle: The AI answers every call, even when 50 come in at once.
With clearly defined requests and a properly trained AI, the armincx voice agent answers up to 95% of incoming calls without a transfer. That is the target value for a good setup with well-maintained data. Plan the training phase realistically and you will get close to this figure.
What Does an AI Phone Assistant Cost?
Most DACH providers charge per minute, per conversation or via monthly licences. For orientation (as of July 2026): fonio starts at around €99 per month plus per-minute costs. VITAS starts between €49 and €98, and its Plus package with 1,500 conversations costs €299. Enterprise solutions such as Parloa are available on request only.
Whether it pays off depends on your call volume and the share of routine requests. A shop with 500 calls per month, half of them order status, quickly saves several working hours per week. A shop with 30 calls per month solves this more cheaply with good self-service on the website.
Which Providers Exist? An Honest Map Instead of a Ranking
Full transparency: armincx is our own product. A "top 10" list with us in first place would therefore be worthless to you. It is more useful to sort providers by use case, because they differ above all in who they are built for.
| Provider | Strong for | Less suitable for |
|---|---|---|
| fonio | Trades, estate agents, law and tax firms, medical practices: call answering plus appointment booking, fast self-service setup, natural voices | Shop processes like order status or returns, which are hard to map without custom API work |
| VITAS | Healthcare and call centres with high call volumes, structured requests, ISO 27001 | Online shops that need a native shop system integration |
| aaron.ai | Medical practices and clinics, deeply integrated with Doctolib | All industries outside healthcare |
| armincx | Online shops: order status and returns on the phone, connected to Shopify, Shopware, JTL and Xentral | Appointment booking for service businesses, practices and trades |
Beyond these four, there are enterprise platforms such as Parloa or Synthflow and developer platforms such as Vapi, where you build the assistant yourself via API. Both categories are aimed at large corporations or at teams with their own developers.
One note for context: All providers listed here offer GDPR compliance and EU hosting. Anyone advertising with that is naming a basic requirement. The real difference lies in whether the assistant can read your order data live.
GDPR and the EU AI Act: The Legal Rules for AI Telephony
You need to resolve three points cleanly before the AI picks up the phone:
- Disclosure obligation: Article 50 of the EU AI Act requires the AI to identify itself as an AI at the start of the conversation. This transparency obligation applies from 2 August 2026. A sentence like "You are speaking with the digital assistant of..." is enough.
- Recording: Call recordings require consent and a clear notice before recording starts.
- Voice data: Clarify data processing agreements, storage location and deletion periods. EU hosting simplifies the review considerably.
Done properly, the legal side is settled within a few days. Ignored, it costs you trust and, in the worst case, a fine.
The Limits: What an AI Phone Assistant Does Not Solve in 2026
Now for the part hardly any provider tells you. The armincx benchmark measures across all channels how much support work automation actually takes over. Fully on its own, the AI resolves around 5% of all tickets. Together with workflows, total automation sits at around one third. When the AI takes on a ticket, it brings 1 in 5 to a full resolution without human involvement.
These figures come from all support channels, with email dominating. We will publish phone-specific rates as soon as the data basis is large enough. To put the two key figures from this guide side by side: The up to 95% describes how many calls a well-trained voice agent can answer and handle on the phone. The roughly 5% measures the share of all tickets across all channels that the AI closes fully autonomously. The two values answer different questions. The speed effect, however, is clearly measurable: AI-resolved tickets are completed around 37% faster in the median, 1.9 instead of 3.0 days. And when the AI escalates, the first response arrives after 0.9 hours instead of 4.1 hours in manual operation.
In concrete terms: An AI phone assistant relieves your team on routine and speed. The complex, emotional and delicate cases remain human work, and that is exactly where your team wins its time back.
FAQ — AI Phone Assistants
What does an AI phone assistant cost per month?
DACH providers start between roughly €49 and €99 per month, plus usage-based costs per minute or conversation. The call volume at which it pays off depends on the share of your routine requests.
Does an AI phone assistant replace my support team?
It takes over routine requests such as order status and returns. Across all channels, total automation among armincx merchants sits at around one third of tickets. Complex cases are still handled by your team, with the transcript as groundwork.
How quickly is an AI phone assistant ready to use?
The basic setup takes hours to days with self-service providers. Add a training phase of a few weeks in which you review and sharpen the answers. The quality of your data determines the pace.
Does the AI have to say that it is an AI?
Yes. Article 50 of the EU AI Act mandates disclosure at the start of the conversation, binding from 2 August 2026. Serious providers anchor this notice in the conversation opening by default.
Which AI phone assistant suits an online shop?
One that connects natively to your shop system and can read order data live. armincx is built for exactly this case, with integrations for Shopify, Shopware, JTL and Xentral. For appointment booking at service businesses, fonio or VITAS are the better fit.
Conclusion: The 24/7 Employee, Honestly Assessed
An AI phone assistant pays off in 2026 for online shops with noticeable call volumes and a high share of routine requests. It resolves order status and returns directly on the phone, around the clock and even during Black Friday. The limits remain real: Around one third total automation is the proven state across all channels, and the rest is teamwork with a better starting point.
Choose your provider by use case. For appointment booking, the path leads to fonio or VITAS, for shop processes to a solution with a genuine shop system integration. You can see how order status and returns work on the phone with armincx on our AI voice feature page.
