Every call answered in your customers' language, at any hour.

AI voice agents answer your calls, resolve routine requests in your systems and hand the hard ones to your team. Queues shrink.

The phone is still where customers go when it matters

Customers still call about what worries them: a bill they cannot pay, a claim, a late delivery. Many of these calls are routine, yet they queue. DACH contact centres must cover several languages and dialects, after hours too, with staff who are hard to hire.

The technology has changed. Speech recognition now copes with Swiss German dialects and free speech, so callers say what they need instead of pressing one. Language models follow the conversation, look up a contract, move an appointment and reply fast enough for natural turn taking.

What separates a pilot from production is the work around the model: authentication, links to CRM, ticketing and telephony, clear rules for when a person takes over, and a record of every call for quality review. Most voice projects underestimate it. We build it.

What we build

  • Multilingual inbound voice agent

    Answers your hotline in German, Swiss German, French, Italian and English, understands free speech and replies in your chosen voice.

  • End-to-end self-service

    Routine requests settled during the call: address changes, invoice copies, payment extensions, bookings. The agent works in SAP or Salesforce.

  • Caller authentication

    Callers verified by customer number, name and address, or SMS code, before anything sensitive. Unverified callers get general information only.

  • Handover with context

    Calls that need a person reach the right team, with who is calling, why and what is done on screen.

  • Natural conversation handling

    Quick replies, callers can interrupt, silence and noise are handled, and numbers, dates and amounts are confirmed before any change.

  • Recording and quality review

    Calls recorded where you have a legal basis, transcribed and tagged. Quality teams review samples and feed corrections back.

How it works

Each call moves through five stages, all logged, so you see what happened and why.

  1. The call connects

    Telephony routes the call to the agent, which says it is an AI and detects the caller's language.

  2. Intent and identity

    The caller explains in their own words. The agent checks identity in your CRM before touching personal data.

  3. Resolve in your systems

    For routine requests, such as moving an appointment, it makes the change and confirms it with the caller.

  4. Hand over when needed

    Complaints, sensitive cases or low confidence go to a person, with transcript and summary. Callers never repeat themselves.

  5. Record, review, improve

    Each call is stored with transcript and outcome. Reviewed failures become test cases before the next release.

How we roll it out

  1. Discovery

    2 to 3 weeks

    We study recorded calls and volumes, then agree which call types to automate first and how success is measured.

    • Call intent and volume analysis
    • Target use cases with success metrics
    • Integration and data protection plan
  2. Pilot

    6 to 10 weeks

    Connected to your systems, the agent takes a few high-volume call types on part of real traffic, reviewed daily.

    • Production voice agent for first use cases
    • Live integrations and handover flow
    • Quality review dashboard
  3. Scale

    Ongoing

    We add call types, languages and hotlines in steps, improving from reviewed calls. Your team learns to own it.

    • Further use cases and languages
    • Regression test suite from real calls
    • Handover runbook for your team

In production elsewhere

Public examples from other organisations, not unus clients, showing what voice agents already do in European customer service.

Public case studyenersuisseEnergy services · Switzerland · 2025 Swiss German voicebot replaces keypad menu, connected to SAP 6.55 out of 7
The situation
enersuisse was set up in 2019 by several Swiss energy providers and serves around 800,000 end customers. Rising prices and supply questions brought more enquiries on more topics, staff turnover made onboarding harder, and callers faced a keypad menu.
What they deployed
With Bucher + Suter, enersuisse launched a voicebot in September 2025. It understands all Swiss German dialects and speaks standard German. In SAP it sends invoices, extends payment deadlines and verifies callers. Other calls reach the right employee, already briefed.
The outcome
The source reports that callers who complete an AI self-service journey give an average customer effort score of 6.55 out of a maximum of 7. Callers who prefer a person can still be put through directly.

Swiss German understanding is now workable, but the value comes from authenticated self-service in the back end, here SAP.

Source: Bucher + Suter, Customer Case Study: enersuisse activates voice agents in customer service, June 2026
Public case studyDeutsche Rentenversicherung HessenPublic pension insurance · Germany · 2025 Voicebot prequalifies calls before they reach staff 85,000 calls a month
The situation
DRV Hessen is the largest social insurance provider in Hessen, serving about 4.2 million insured people, 584,000 retirees and 117,000 employers. Calls on pensions, rehabilitation, prevention and retirement planning must each reach the right specialist team.
What they deployed
Deutsche Telekom's Conversational AI Suite runs on the Open Telekom Cloud, in GDPR-compliant data centres. The voicebot asks for the concern and insurance or case number and routes the call to the right team. Simple issues are fully resolved.
The outcome
According to Telekom, the system prequalifies roughly 85,000 calls each month automatically, and callers get through with virtually no wait times even at peak times.

Prequalification with a clean handover is a safe first step for a public body, and it runs on a German sovereign cloud.

Source: Deutsche Telekom, Quick, secure, personalized: The voicebot assists with inquiries to Deutsche Rentenversicherung Hessen, August 2025
Public case studyAUTO1 GroupOnline car trading · Germany · 2026 AI agents answer every call across nine European markets 9 countries, 7 languages
The situation
AUTO1's wirkaufendeinauto.de serves millions of people a year who want to sell their car. They call about appointments, documents, prices and next steps. The team and old IVR could not keep up with growth, and customers wanted their own language.
What they deployed
AUTO1 deployed Parloa AI agents that answer every call, recognise customers by phone number and book, change or manage car purchase appointments, or hand over to a person. A pilot in Germany and Spain went live in two weeks.
The outcome
The source states the agents were rolled out across nine countries and seven languages, including Germany, Austria, France and Italy, in nine weeks, and now handle millions of customer interactions.

Starting with one well-defined task, appointment handling, made a fast multi-country and multilingual rollout possible.

Source: Parloa, AUTO1 Group customer story, September 2026

These are public deployments by other organisations, summarised from the sources linked. They are not unus clients.

Built for European rules

Data stays in Europe

Recordings, transcripts and models run in EU or Swiss data centres or your own cloud tenant, under GDPR and the Swiss nDSG.

Callers know it is AI

Every call opens with the AI disclosure Article 50 of the EU AI Act requires. Callers can always ask for a person.

Know your risk class

Routing is rarely high-risk. Credit, life or health insurance pricing, benefits and emergency triage bring Annex III duties from 2 December 2027.

Security and human oversight

Callers authenticated, card and account numbers masked, every action logged. Supervisors can take over any call. We never infer staff emotions.

Questions buyers ask

Can it really understand Swiss German?

Speech recognition for Swiss German has become good enough for production: the enersuisse voicebot on this page understands callers across dialects. Names, numbers and addresses remain the hard part, so the agent reads them back for confirmation. Like enersuisse, most deployments answer in standard German. We test recognition on your own recorded calls during discovery and set the confidence thresholds at which the agent asks again or hands over.

How fast does the agent respond, and can callers interrupt it?

We stream speech recognition, the language model and speech synthesis, and tune the pipeline so the agent replies within the pause a person would leave in conversation. We measure response time on every call. Callers can interrupt at any time and the agent stops speaking and listens. For slower back-end lookups the agent says what it is doing, so the line never goes silent.

What happens when the agent cannot help?

It transfers the call to the right team through your existing telephony, with the transcript, verified identity and a short summary on the human agent's screen. You define the triggers: complaints, vulnerable customers, low confidence, repeated misunderstanding or a direct request for a person. Outside opening hours it can take a callback request or open a ticket instead.

Do we have to replace our contact centre platform?

No. We connect to what you run: Genesys, Avaya, Twilio, Microsoft Teams telephony or SIP trunks from your carrier, and CRM or ticketing systems such as Salesforce, ServiceNow and SAP. The voice agent becomes one more destination in your routing, so you can start with one number or one call type and expand from there.

Are calls recorded, and who can see them?

Calls are recorded and transcribed where you have a legal basis, with the notice and consent your jurisdiction requires. Recordings and transcripts stay in your chosen region, sensitive data such as card numbers is masked, and access is limited by role. Retention periods are yours to set, and deletion is logged.

How do we know it is working?

Before launch we agree the measures with you, for example share of calls resolved without a human, transfer reasons, wait times and customer effort scores. A dashboard tracks them by intent and language. Your quality team reviews samples of calls each week, and every failure they flag becomes a test the agent must pass before the next release.

Bring one process. We'll bring the loop.