What does AI automation cost for a Swiss SME? Honest prices June 2026

What this is about

The question "what does AI cost in our office?" cannot be answered with a single number. It splits into three line items: one-off project cost, monthly infrastructure, and per-query token cost. Looking at only one item leads to bad planning. Knowing all three lets a 5-FTE fiduciary office plan with roughly CHF 20,000 for the first year – through a stepped path from the speedcheck via the KI-Fundament to full onboarding with twelve months managed service.

Prices below reflect June 2026 in the Swiss market for typical fiduciary/SME setups with 3 to 15 employees. They are not list prices of a software, but consulting and implementation fees plus the transparent third-party costs the client pays directly. We name them openly because opaque "on request" pricing is the biggest brake on pragmatic AI adoption in Switzerland. TYTOS Switzerland cites a comparable range of CHF 15,000 to 30,000 setup for a 20-FTE SME in its 2026 guide; our smaller packages sit below that because they focus on a single bottleneck instead of full coverage.

Why the breakdown matters

Two mistakes happen regularly in Swiss SMEs. First: buying the wrong package. A 4-FTE office with 60 clients books a CHF 90,000 platform with custom model training because a vendor sold "Enterprise AI". The money is gone, the office works three months later as before. Second: the shock of hidden recurring costs. Another office starts with a "free pilot" and discovers, after six months, that monthly token spend has reached CHF 800 because nobody configured routing.

A clear breakdown defends against both. A KI speedcheck at CHF 590 opens the door and is fully credited; a readiness check (audit-light) at CHF 1,700 decides whether a CHF 4,800 KI-Fundament even makes sense. A CHF 7,800 AI backend pilot is not a leap into the unknown but a clearly bounded 4-to-6-week scope with a defined output artefact. A full onboarding over twelve months is the choice for offices that do not want to run operations themselves.

The link to value capture is in the ROI calculator (see roi-rechner-ki-projekt). This page is the honest cost side – what you must take from cash flow before the savings show up.

Packages and line items in detail

KI Quick Scan – CHF 590, door-opener. A compact first check: one focused workshop plus a short written assessment with the three most obvious automation angles and a recommendation on whether a deeper readiness check is worthwhile. Fully creditable against any follow-up package. The cheapest entry – for offices that first want an honest outside view without a big budget.

Readiness Check (Audit Light) – CHF 1,700, 2 days. Two workshops of 3 hours each, one with management, one with case handlers. Output: a written report with three prioritised automation candidates, effort/benefit estimate per candidate, data-protection prep recommendations. No code. Suitable as a decision basis for the next quarter.

KI-Fundament – CHF 4,800, 5 days. Readiness check plus setup of an EU-hosted server (typically Hetzner Helsinki or Falkenstein, around CHF 30 to 60 per month directly on the client account), installation of one n8n workflow (e.g. email triage or document recognition), a 2-hour team training. Delivers one productive workflow at the end of the week.

AI Backend Pilot – CHF 7,800, 4 to 6 weeks. Complete backend pipeline: multi-LLM gateway (LiteLLM), vector DB (Qdrant) with initial ingestion, RAG over existing guidelines/manuals, 2 to 3 n8n workflows, monitoring (Grafana + Sentry). Includes the data-protection impact assessment as a finished document. Not included: training, managed service.

Full Onboarding – CHF 14,500, Pilot + 12 months managed. The pilot above plus twelve months Managed Service Pro (patching, monitoring, monthly update call, minor adjustments). For offices without an in-house IT person.

Server rent – CHF 30 to 120 per month, directly on client account. We do not resell the server – it runs on the client's Hetzner/OVH account. Sovereignty stays with the client; no lock-in. Typical configurations: CX22 for audit stage (CHF 7/mo), CX32 for standard (CHF 14/mo), CCX23 for offices with local Ollama (CHF 60/mo), dedicated AX42 for GPU needs (from CHF 120/mo).

Token cost – typical CHF 1 to 25 per month. Example RAG with 200 queries per month, Mistral Small 3.1 (May 2026: USD 0.20/1M in, USD 0.60/1M out): about CHF 1 per month. Heavy usage at 2,000 queries with larger contexts: up to CHF 25 per month. Anthropic Claude or OpenAI GPT-4o sit 5 to 10 times higher, hence multi-LLM routing.

Managed Service – CHF 600 / 1,200 / 2,200 per month. Three tiers. *Basic* (CHF 600): patching, backup check, monitoring alerts. *Pro* (CHF 1,200): Basic plus a 60-minute monthly update call, up to 4 hours of adjustments. *Plus* (CHF 2,200): Pro plus guaranteed 4-hour weekday response, quarterly security review, ROI reporting.

Total example, 5-FTE fiduciary, first year. KI-Fundament CHF 4,800 (credits the speedcheck and readiness check) – that is CHF 4,800 for the entry stages – plus AI backend pilot CHF 7,800 – project total CHF ~12,600. Plus 12 months server (CX32: CHF 168) plus token (CHF 1 x 12 = CHF 12) plus Managed Pro 12 months (CHF 14,400). Year-1 total: ca. CHF 27,180. If the office stays on Managed Basic: about CHF 20,000. Both numbers sit inside the range Deloitte 2026 describes with break-even at 2 to 4 months.

Booking and billing flow

1. 01Free 30-minute first call: we listen, ask about the three biggest time drains, propose the right package.
2. 02Start with the KI speedcheck (CHF 590), book a readiness check (CHF 1,700) or go straight to the KI-Fundament (CHF 4,800) – speedcheck and readiness check credit if you upgrade within 60 days.
3. 03Contract and privacy declaration in writing, 50% deposit before project start, balance on acceptance.
4. 04Server account opened in the client's name – Hetzner/OVH/Infomaniak, your choice. Payment flows directly between client and provider.
5. 05Token account at Mistral/OpenAI/Anthropic also in the client's name – we set budget alerts at USD 10/20/50 per month as a safety net.
6. 06Managed Service starts month 1 after pilot acceptance, terminable with 1 month to month-end – no 12-month lock-ins.
7. 07Monthly invoice in CHF, VAT applicable (Swiss establishment), net 30. Quarterly or annual prepayment grants 5% rebate.

Which package fits which profile

The KI speedcheck suits offices on a tight budget that want a first outside view with minimal commitment before deciding on a bigger step. The readiness check (audit-light) suits offices that do not yet know where their biggest time drains lie and want a thorough outside perspective first – before they invest. The KI-Fundament suits offices with a concrete bottleneck (e.g. email flood, manual document capture) and a clear expectation of the first productive use. The AI backend pilot suits offices that want to automate several workflows and are willing to invest in their own infrastructure rather than a SaaS subscription – typically from 5 FTEs upward.

Full onboarding is the path for offices with established client processes that know they have neither time nor appetite to maintain an n8n instance themselves. The client pays for the calm of seeing the system run every month without further input – and receiving once a month a short email saying "all green plus three improvement suggestions".

Server rent and token cost are not optional; they are part of every package from the KI-Fundament onward. They are however transparent and belong to the client – if the consultant is changed, the infrastructure stays in place.

When not to book

Do not book if the office has fewer than 3 FTEs and no recurring standard processes – setup overhead exceeds the benefit. Investment in a good bookkeeping platform (Bexio, Klara, Run my Accounts) creates more value than AI here.

Do not book if management positions AI as a tool for staff cuts. The packages are designed to free up case handlers and allow more clients per person – not fewer people. Starting with the wrong expectation typically kills the project after three months and the investment is lost.

Do not book if the data-protection groundwork (DPA with the LLM provider, DPIA, updated privacy notice) cannot or will not be done. We supply the documents in the pilot, but management must sign them. Without that signature the office sits in the revDSG risk zone.

Trade-offs

FAQ

Related topics

Sources

1. TYTOS Schweiz – KI-Automatisierung für Schweizer KMU Leitfaden 2026 · 2026-03
2. schnellstart.ai – 5 Prozesse zum Sofort-Automatisieren (Frauenfeld Case) · 2026-03
3. Mistral La Plateforme – Pricing Mai 2026 (Small 3.1, Large 3, Medium 3) · 2026-05
4. AI Automation Hub Schweiz – Kosten und ROI Berechnung · 2026-02
5. Hetzner Cloud – Preise CX22/CX32/CCX23/AX42 · 2026-05