Choosing your AI advisor in Switzerland: four criteria

To choose who guides you through AI, four verifiable criteria are enough: a measurement logic rather than a catalogue of services, FADP compliance built in from the scoping stage, demonstrable technological independence, and the capacity to measure what is promised. In the interest of honesty: I sell exactly this kind of engagement. So take this grid for what it is — criteria to turn against any provider, myself included.

Article published in March 2026, revised on May 25, 2026, updated on July 7, 2026.

Why sorting through the options has become difficult

The Swiss market for artificial intelligence consulting has grown denser as generative AI became a boardroom topic. Generalist strategy firms that added an AI practice, digital agencies repositioned overnight, specialized independents, technology integrators expanding their offering, academic spin-offs: the diversity of supply is not a problem in itself. It becomes one when the executive has no framework to distinguish a serious approach from an opportunistic offer.

The structural context actually works in your favor. Switzerland has a recognized academic ecosystem in artificial intelligence — federal institutes of technology, active research institutes, national coordination structured since 2024 between higher education institutions and economic actors. This fabric produces competent providers in sufficient numbers. The decision problem is therefore not scarcity. It is qualification.

First criterion: what are they trying to sell you in the first hour?

A serious approach begins by stating what it does not sell. A provider who presents AI as an emergency to be handled through a string of proofs of concept is selling a grammar of urgency that holds up neither over the life of an engagement nor against two years of real-world experience. A provider who announces a quantified return on investment before measuring your starting point is selling a promise rather than a method.

What you should be looking for is a measurement logic: a diagnostic that establishes your current situation across the relevant scope — visibility, productivity, automation, internal use of models —, an explicit prioritization of corrections, a stable cadence of re-measurement. This is exactly the logic behind a solutions scoping: your actual workflows first, the tools second.

The test happens within the first hour of conversation. Is your contact selling services, or qualifying your situation? If they are selling, they have not yet begun to understand you. If they are qualifying, they have the posture that will carry the engagement.

Second criterion: FADP built into scoping, not bolted on at the end

The revised Federal Act on Data Protection, in force since September 1, 2023, imposes demanding requirements on the processing of personal data — particularly in cases of international transfers or automated decisions^[1]. Generative models accessed via an API hosted outside Switzerland raise precise questions, and a serious engagement addresses them at the scoping stage, never as a final checkbox.

Concretely, depending on the sensitivity of your data, the proposed architecture can combine Swiss hosting for the most sensitive data, open models deployed on controlled infrastructure when autonomy matters, and public APIs only for what touches no personal or strategic data. The Swiss ecosystem allows for this layered architecture: several providers offer infrastructure domiciled in Switzerland, and the major international vendors operate Swiss regions. The choice is a matter of risk qualification by use case — ideological preference has no place in it. A provider who cannot conduct this qualification should not be leading the engagement.

Third criterion: could you switch models in six months?

The model market moves at a pace that renders any fixed technology bet obsolete. A provider who ties their entire practice to a single vendor weakens the companies they advise. Operational independence can be verified: the capacity to compare available models for your use case, to move a system from one provider to another without a full rebuild, to mobilize open models when the context calls for it.

The practical test I would suggest: ask, for a specific case, why this model rather than another — and what would happen if you had to migrate in six months. A well-reasoned answer signals a real practice. An evasive answer signals a disguised single-vendor dependency.

Fourth criterion: do they know how to measure what they promise you?

For engagements touching your visibility in search engines and AI assistants, an additional criterion has applied since May 2026: demand a codified measurement of citability, rather than content produced on a bet that it will be cited. The official doctrine published by Google on May 15, 2026 clarified that the core ranking systems also govern the generative features of search^[2]. This clarification disqualifies the narratives that sell AI optimization as a discipline separate from SEO. It does not disqualify the measurement of citability, which remains a strategic matter.

For this purpose I have stabilized the Score GEO™, fully laid out in MCVA Cahier No. 1. Other methods exist and can legitimately coexist. Which instrument is chosen matters less than one condition: that one exists, and that it is transparent and reproducible.

What this grid does not cover — and when not to choose me

This grid filters for methodological posture; it says nothing about sector expertise, the quality of the relationship, verifiable references, or operational reliability — check those elsewhere. And it deserves an honest addendum. If your core need is heavy integration of an existing system, an integrator will serve you better than a consulting firm. If you have a daily volume of AI work to absorb in-house, a hire is worth more than an engagement. The right provider is the one whose shape fits your need — and whichever it is, the four criteria above still apply.

Before signing anything, know what you want to automate, produce, or measure: that is the purpose of a usage diagnostic, not a sales brochure. And to map out the full landscape before choosing who accompanies you, start with What should a Swiss SME do about AI in 2026?.

Key takeaways

— A serious engagement qualifies your situation before selling, and builds FADP compliance into the scoping stage of the mandate. — Technological independence can be tested: why this model, and what happens if migration is needed in six months? — Since May 2026, anyone promising visibility in AI systems must know how to measure it, with a transparent and reproducible instrument.

FAQ

Firm, independent, or integrator: does the legal form matter? No. Every form can carry out a serious practice, and every form also shelters opportunistic offers. The four criteria — measurement, FADP at scoping, technological independence, measurability of promises — apply to all of them. Size changes execution capacity, never the required posture.

How much does AI consulting cost in Switzerland? Ranges vary too widely — volume, sector, level of outsourcing — for a general figure to be honest. Instead, demand a quote structured by deliverable, excluding VAT, with a measurement of the starting point. Be wary of a price quoted before any qualification of your situation: it signals a standardized offer that will adapt poorly.

Does the provider have to be Swiss? No, but they must master the FADP and Swiss hosting options, because that is where a real share of the risk plays out. Proximity helps for workshops and change management. The decisive criterion remains the capacity to qualify data risk by use case.

What is the first question to ask in an interview? "How will I know, with figures to back it up, that your intervention worked?" A serious answer describes a baseline measurement, tracked indicators, and a cadence of re-measurement. If the answer is a catalogue of services, you have your answer.

And for your next engagement? The AI Usage Diagnostic: sixty minutes to lay out your actual workflows, identify what deserves custom work, what stays in SaaS, and what needs no AI at all. Book a diagnostic

Sources

[1] Federal Act on Data Protection (FADP), revision of September 25, 2020, in force since September 1, 2023. www.fedlex.admin.ch/eli/cc/2022/491/fr [↩]

[2] Google Search Central, Optimizing your website for generative AI features on Google Search, published May 15, 2026. developers.google.com/search/docs/fundamentals/ai-optimization-guide [↩]

Jérôme Deshaie is CEO and founder of MCVA Consulting SA, an augmented agency based in Valais. Fifteen years serving major international brands, now working directly with Swiss SMEs. Background.