Services · Building block

Every call gets answered and becomes a case.

The AI phone assistant picks up, says an AI is speaking, and creates the case in your CRM or ticket system. Your team makes the decisions.

Case T-26-04182Ticket system
  1. CallerMs Brandt, no. 40213identified
  2. RequestWater in kitchen, flat 3.2captured
  3. Urgencyhighon call
  4. Next stepPut throughhanded over
Service lineMon 18:42
Ms Brandt+49 521 3389 041201:48
  1. AI assistantKramer Property Management, you are speaking to an AI assistant. This call is transcribed. How can I help?
  2. Ms BrandtWater is coming out of the wall in my kitchen. Flat 3.2.
  3. AI assistantMs Brandt, customer number 40213, is that right? I am putting you through to the on-call service now.
  • Calls no longer ring into the void

    The agent picks up when nobody is free, over lunch or after hours. A ringing phone becomes a case.

  • Fewer routine questions

    Delivery date, technician visit, status: the agent answers straight from the system. Your team takes the calls that really need them.

  • Everything is in the system

    Summary, transcript and category land in the CRM or ticket system. No sticky note that goes missing.

  • Open about being AI, legally sound

    In its first sentence the agent says an AI is speaking and whether the call is transcribed. No audio without consent.

Where it gets stuck today.

  1. Nobody picks up.

    In the workshop, the warehouse or out on site, the phone rings into the void. The caller does not leave a message, they ring your competitor.

  2. The same three questions, all day.

    The colleague who also does the invoicing spends half her day on delivery dates, technician visits and where a case stands.

  3. The note disappears.

    The call is taken, the request goes on a sticky note. It shows up in the system when the customer calls again.

From call to case

Answered, logged, passed to the right person.

The agent settles name, request and urgency and reads numbers back to confirm them. The case then sits in your system with its transcript, and anything that needs a decision moves on with the full picture.

Service lineMon 18:42
Ms Brandt0521 3389 041201:12
  1. AI assistantKramer property management, this is an AI assistant. The call is transcribed.
  2. Ms BrandtWater is coming out of the wall in my kitchen. Customer number 40213.
  3. AI assistantLet me repeat: 4 0 2 1 3, Lindenstrasse 12. Is that right?
Ticket T-26-04182Ticket system
CustomerUlrike Brandt, 40213
PropertyLindenstrasse 12, flat 3.2
CategoryWater damage
Urgencyhigh, emergency word
SummaryWater leaking from the kitchen wall.
Create case
Hand-overT-26-04182
  1. On callMr Yilmaz, mobilereached
  2. StatusSummary read outhanded over
  3. Transcripton the casesaved

Hand-over

When the agent passes the call to a person.

You set the hand-over rules before the first call. Nobody has to tell their story twice.

  • Emergency words go straight through

    Water damage, standstill, injury: such calls go to the on-call service without a detour.

  • Ask for a person, get a person

    One sentence is enough. Depending on the time, the agent puts the call through or adds a callback with notes to the list.

  • Unsure means hand over

    If it does not understand a detail with certainty, or the call is about goodwill, prices or commitments, it hands over instead of guessing.

Transcriptlive
  1. Kramer property management, this is an AI assistant. How can I help?

  2. Caller

    I would like to speak to someone, please. It is about a credit note.

  3. Of course. I am passing your request on with what we have so far. Ms Lenz will call you back by 11.

    Callback list: goodwill request
Hand-over rulesService line
  1. Emergency wordWater, standstill, injuryput through
  2. Goodwill, pricesany commitmentto the team
  3. Not recognisedafter one follow-upcallback with notes

Approaches

How an AI phone assistant gets built.

On every route it connects to your phone system over SIP, whether 3CX, NFON, Placetel or another. Check response time and German speech quality on each route. And for all of them: the AI is disclosed in the first sentence, recording only with consent.

  • Dograh, run by us

    Our default

    Our stack, open source, in Germany

    Fits when
    calls and recordings should not sit with a US platform. Dograh is an open alternative to Vapi or Retell: flows in an editor, language, speech recognition and voice models of your choice, connected to your phone system over SIP.
    Mind
    German voice quality depends on the speech model chosen; we test with real calls before it goes live.
  • Voice platform: Vapi, Retell, ElevenLabs Agents or similar

    Managed, specialist provider

    Fits when
    it has to be quick. Speech recognition, model, speech output and interruptions come wired up; your phone system connects through a SIP trunk.
    Mind
    Check data location per provider; EU data residency is sometimes only in the enterprise tier. Billing is per call minute.
  • Realtime speech model directly: OpenAI Realtime API or similar

    Built on the model interface

    Fits when
    conversation design and integration should be fully in your hands. A speech-to-speech model answers without a detour through text; according to OpenAI, the Realtime API takes calls directly over SIP.
    Mind
    Handover, logging and operations are yours to build. Test voice and German pronunciation on real calls.
  • Voice bot module of the contact centre suite

    Extension of existing software

    Fits when
    you already run contact centre software whose vendor offers a voice bot. Routing, queues and reporting stay in one place.
    Mind
    Ties you to the vendor. How well the module understands German and how open it is to CRM and ERP: check on your own calls before buying.
  • Classic phone menu (IVR)

    No AI, keypad or keyword

    Fits when
    it is only about routing to the right place. Predictable, cheap and already included in many phone systems.
    Mind
    Takes no request and opens no case. Long menus lose callers.

Tasks and limits.

What the system takes on

  • Answers and says in its first sentence that an AI is speaking and whether it transcribes
  • Asks for the mandatory details and reads numbers and names back
  • Creates the case with summary and transcript in your system
  • Gives simple answers from the system, such as delivery or appointment status
  • When unsure it asks again or hands over instead of guessing
  • Measured weekly: solved alone, handed over, dropped calls, a sample of wrongly created cases

What your team takes on

  • Sets the greeting, the mandatory details and the emergency words
  • Decides anything involving money or liability: goodwill, prices, commitments
  • Takes the unclear, emotional and delicate calls
  • Sets the retention period and brings in the works council before the build
  • Limit: no agent that poses as a person, no audio stored without notice and consent
  • Limit: not worth it with few calls, or where every call is real advice

What it connects to.

Works with the systems you already use

AI phone assistants

An AI phone assistant is an AI agent on the phone: it answers, understands the request, opens the case in your system and hands over when a person has to decide.

Which one when?

  • AI agents

    If the request comes by email or form, it is a job for AI agents. Same build, different channel.

  • Workflow automation

    If all that follows the call is a fixed routine, such as a confirmation and an ERP entry, workflow automation is enough.

  • AI knowledge systems

    If the agent should answer from manuals or policy terms, with the source, add a knowledge system.

The way there.

From the first call to a system in operation. Each phase stands on its own; none ties you to the next.

  1. Intro call

    free

    30 minutes. We work out where AI fits for you and whether an AI strategy is worth it.

  2. AI strategy

    €1,990

    AI strategy for mid-sized companies: which of your processes suit AI, what the build costs and brings back, and which ones are not worth the effort.

  3. Pilot & implementation

    from €5,000

    AI pilot at a fixed price: one use case, fixed scope and acceptance criteria in writing up front. Built in your systems, tested with your data.

  4. Operation & improvement

    from 20% a year

    Running AI systems: hosting in Germany, monitoring, model and prompt maintenance, ongoing measurement, and the next use case on the same foundation.

Further reading

All posts

Frequent questions

Common questions about it.

The questions that almost always come up.

What does an AI phone assistant cost?

In our example calculation from list prices, a minute of conversation costs roughly 4 to 11 US cents; ready-made subscription products run from 50 to several hundred euros a month. The connection to your software, maintaining the answers and operation usually cost more than the minutes. The full calculation is in our blog post on cost and law.

Does the caller have to know an AI is speaking?

Yes. Under Article 50 of the EU AI Act, applicable since 2 August 2026, it must be clear at the latest at the first interaction. So the notice is in the first sentence, not in the small print. We do not build agents that try to sound human, but a service line that gets to the point.

May we record calls?

Only with the consent of everyone on the call. In Germany, recording the privately spoken word without consent is a criminal offence under section 201 of the Criminal Code. The agent announces it at the start and puts the caller through if they decline. Often a transcript without stored audio is enough.

What happens when the AI phone assistant gets stuck?

It hands over: to a person, the on-call service, a callback list or voicemail, by your rule. Anyone who asks for a person gets one.

How well does an AI phone assistant cope with dialect and technical terms?

Well with standard German on a quiet line, noticeably worse with strong dialect, workshop noise or part numbers. So we test against your own recordings before building and have numbers read back.

How fast does the AI phone assistant answer?

In conversation people expect an answer within a fraction of a second. Speech recognition, model and speech output together take a little longer. We measure response time in operation and keep it short enough that the conversation does not stall.

Where do the call transcripts end up?

Where the case already lives: on the contact in the CRM, on the ticket. A transcript that exists only as a file does not get read.

Can an AI phone assistant be run in line with the GDPR?

Yes. Processing in German data centres or your own infrastructure, under a data processing agreement under Article 28 GDPR, and only with model access that does not train on your data. Voice recordings are personal data, so the retention period is fixed before the first call.

The next step

Let us find out where AI pays off for you.

30 minutes, free and without obligation.