Most vendors are happy to demo a voice agent taking a call. Getting a straight answer on what it costs per month — and whether that beats your current setup — is considerably harder.
This article publishes the numbers. Not precise quotes (those depend on your infrastructure, call volume, and requirements), but a real breakdown of the cost components, what drives them up or down, and a straightforward way to check the economics before you talk to anyone.
If you want the short version of the technology first, AI Voice Agents: What They Are and How Businesses Use Them has you covered.
The Four Cost Buckets in Every AI Voice Agent Deployment
Vendors usually quote a single monthly figure. That figure is made of four distinct components, and understanding which one drives your bill is what separates a good procurement decision from a nasty surprise six months later.
1. Telephony (per-minute costs)
Every call your agent handles touches a telephony layer — a VoIP carrier, a SIP trunk, or a cloud telephony platform (Twilio, Vonage, Bandwidth, and several others operate in this space). You pay per minute of inbound and outbound call time.
Realistic range: roughly CHF 0.01–0.04 per minute for inbound/outbound calls to Swiss landline numbers on a standard pay-as-you-go basis. Calls to Swiss mobile numbers are significantly higher (CHF 0.18–0.25/min on major platforms such as Twilio and Peoplefone). Volume commitments or enterprise SIP agreements can reduce these materially.
For a business handling 500 call-minutes per month — roughly 50 calls averaging 10 minutes — this line item is CHF 2.50–10. At 5,000 minutes, it’s CHF 25–100. Telephony is rarely the dominant cost; it becomes significant only at high call volumes or when you’re making a lot of outbound calls.
2. Speech and language model costs (per-minute or per-token)
This is the engine: speech-to-text converts the caller’s voice into text, a language model processes it and generates a response, and text-to-speech converts that response back into audio. You’re paying for all three, typically charged per minute of audio or per 1,000 tokens processed.
Realistic range for a full speech pipeline: CHF 0.05–0.25 per conversation minute, depending on the models chosen. High-quality, low-latency speech models cost more than lower-latency options with shorter vocabularies. Longer, more complex calls consume more tokens. Premium stacks using high-quality neural voices and frontier LLMs can reach CHF 0.30+ per minute.
This is usually the largest variable cost for voice agents. A business with 500 conversation-minutes per month might pay CHF 25–100 on model costs alone. At 5,000 minutes, that’s CHF 250–1,000.
3. Integration and build cost (one-time, amortised)
Before a voice agent answers a single call, someone has to build it: design the conversation flows, connect it to your calendar or booking system, CRM, or ticketing platform, configure fallback behaviour, test edge cases, and set up monitoring. This is where most of the real engineering work lives.
Realistic range for a custom agent:
- Simple agent (FAQs, collects a name and callback number, sends a notification): CHF 3,000–8,000 one-time
- Mid-complexity agent (books appointments, handles rescheduling, integrates with one external system): CHF 8,000–20,000
- Full-service agent (multiple intents, CRM + calendar + escalation routing, multilingual): CHF 20,000–50,000+
These are order-of-magnitude figures. The actual number depends on your existing systems, API quality, and how much the conversation flows diverge from a simple linear path. See What AI Agent Development Really Costs in 2026 for a broader breakdown of engineering investment across agent types.
Amortise the build cost over two years: a CHF 12,000 build spread over 24 months adds CHF 500/month to your total cost model.
4. Ongoing maintenance and tuning
A voice agent is not set-and-forget. Conversations drift — customers ask questions you didn’t anticipate, product lines change, phone numbers get updated. You need someone reviewing call logs, refining responses, and catching failure modes before they cost you customers.
Realistic monthly range: CHF 200–800/month for a light maintenance retainer on a stable agent, more if you’re iterating actively or expanding capabilities. Some businesses handle this internally once the agent is handed over; most benefit from a partner who monitors production quality on an ongoing basis.
The Other Side: What You’re Currently Spending
The cost of a voice agent only means something relative to the alternatives. There are three main comparators:
A human receptionist or answering team. In Switzerland, a part-time receptionist (around 40–60% FTE) covering extended hours might cost CHF 2,500–4,500/month in total employer cost, including mandatory social contributions. Full-time coverage across business and extended hours is higher still. The voice agent doesn’t call in sick, doesn’t take lunch, and handles simultaneous calls.
A professional answering service. Third-party answering services in Switzerland are typically structured as a monthly base fee plus a per-call or per-minute charge — exact Swiss market rates vary by provider and service level, so it is worth requesting quotes from providers such as Callpoint or similar. Quality varies; they don’t have access to your live booking system. See The AI Receptionist: 24/7 Phone Coverage for Small Business for a closer look at how the two compare.
Missed calls. This is the hardest to quantify, but often the most significant. If your phone rings after hours and nobody answers, a share of those callers books elsewhere or simply moves on. For a service business where a single new client is worth CHF 500–5,000, missing 10 calls a month that would have converted at even 15% is a real number.
A Break-Even Worksheet: Does the Math Work for You?
Run this for your own situation. The inputs are yours; the structure is the logic.
Step 1 — Estimate your total monthly AI voice agent cost
| Component | Your estimate |
|---|---|
| Telephony (call-minutes × per-minute rate) | CHF ___ |
| Model costs (conversation-minutes × per-minute rate) | CHF ___ |
| Build cost amortised over 24 months | CHF ___ |
| Maintenance retainer | CHF ___ |
| Total monthly cost | CHF ___ |
Step 2 — Estimate what you’re saving or earning
| Value driver | Your estimate |
|---|---|
| Receptionist / answering service cost replaced | CHF ___ |
| Value of calls answered outside current hours (volume × conversion rate × avg. customer value) | CHF ___ |
| Reduction in admin time for staff (hours × hourly cost) | CHF ___ |
| Total monthly benefit | CHF ___ |
Step 3 — Check the sign
If monthly benefit minus monthly cost is positive, the case holds. If it’s marginal, the most sensitive inputs are usually call volume and the value of captured after-hours enquiries.
Illustrative example: A 15-room boutique hotel takes 20 booking calls a week, 40% outside staffed hours. Average booking value: CHF 300. If the agent converts half those unanswered after-hours calls (roughly 8/week × 0.5 × CHF 300 = CHF 2,400/month) and the total agent cost is CHF 800/month, the break-even is clear. This is illustrative arithmetic; your conversion rate and booking value will differ.
What Drives Costs Up (and How to Control Them)
Several factors push costs above a vendor’s headline figure:
- Complex conversation flows. An agent handling eight intents with conditional branching costs significantly more to build than one handling two. Scope deliberately.
- Poor-quality integrations. Undocumented APIs in your booking system or CRM extend integration work and cost. Assess this early.
- Long average handle time. Model costs scale with minutes. Voice performs best on structured tasks (bookings, information, routing) — not open-ended support calls that run 15+ minutes.
- Multiple languages. Multilingual capability is achievable — Multilingual Voice Agents: One Phone Line, Four Languages covers the approach — but it increases testing scope and maintenance overhead.
When Voice Agents Are Not the Right Answer
Honest caveat: a voice agent is not always the right answer.
If you receive fewer than 20–30 calls per week, the amortised build cost dominates the math. If most calls require genuine human judgment — complex complaints, sensitive relationships, non-routine negotiation — automation creates a worse experience than a person. And if your team handles call volume comfortably within working hours with no meaningful after-hours demand, the problem is simply small.
Measuring the ROI of AI Agents: A Framework for SMBs provides a process-level framework for identifying which workflows justify the investment.
Getting to an Accurate Number for Your Business
The ranges in this article are honest approximations. Your real cost depends on specifics: your telephony setup, the systems the agent needs to integrate with, how many conversation paths matter, and what level of ongoing tuning makes sense.
The fastest way to get from “does this make sense for us?” to a real number is a scoped conversation. Our process optimisation practice includes voice agent deployments — we’ve built agents for bookings, after-hours enquiries, and inbound lead qualification.
If you want to run your own numbers with someone who can pressure-test the assumptions, book a 30-minute call with our team. We’ll assess your call volume, your current setup, and whether a voice agent is the right lever — or whether a different automation delivers a better return.