How to choose an AI consulting partner

Fadel Dia-Eddine· Co-Founder & Product Lead6 min di lettura

To choose an AI consulting partner, first define the work you need help with: assessing a use case, developing the software, integrating it with existing systems or operating it after launch. Then compare providers against that scope. A workshop, a model evaluation and a production implementation require different deliverables and skills.

For Swiss companies, the question is increasingly practical. In AXA's 2025 survey of 300 SMEs, deliberate AI use rose from 22% to 34%, and 57% of AI users reported time savings. Those results show adoption, but they do not establish which kind of consulting support an individual company needs.

34%
Swiss SMEs deliberately using AI
Up from 22% the year before, 2025
89%
Employees using AI at work
EY Switzerland, 604 respondents, 2026
57%
AI-using SMEs reporting time savings
AXA SME survey, 2025
31%
Still in pilot or proof of concept
EY Switzerland, 2026

EY's 2026 Swiss survey, with 604 respondents, found 89% using AI at work and 31% reporting a pilot or proof-of-concept stage in their company. The AXA and EY samples differ, so their percentages should not be treated as directly comparable measures of company readiness.

Check experience beyond the prototype

An AI application that reads internal data or updates a CRM needs an access model, logging, tests and a recovery plan. Ask a prospective partner to explain those parts of a comparable deployment. The answer should cover what happens when a model, data source or integration changes.

EY's survey identified data quality and silos as the most frequently reported obstacle, followed by security and privacy concerns and a shortage of skilled people. Ask who on the proposed team will address those dependencies and what they need from your staff.

Evidence from Swiss financial institutions

In its 2025 survey of around 400 supervised institutions, FINMA found roughly half using AI or developing applications, with another 25% planning use within three years. Data quality, data protection and explainability were prominent concerns.

FINMA's survey findings provide useful context for financial firms. They also illustrate why sector experience matters when selecting the project team.

Nine questions worth asking

Use the same questions with each shortlisted provider. Ask for examples and name the person responsible for each part of the proposed engagement.

Do they start with the business case?

Ask which process should improve, who uses it and how the improvement will be measured. A recommendation should explain the expected value and the alternatives considered.

How will they assess your data?

Ask how sources will be connected, permissions enforced, quality checked and access logged. The assessment should identify missing data and preparation work before development is priced.

Can they build what they propose?

Identify who will write the pipelines, APIs, interface and deployment configuration. If another team will implement the proposal, check how responsibilities and acceptance criteria will be handed over.

Can they integrate with what you already run?

Describe the documents, APIs, CRM, ERP and support tools involved. Ask for relevant integration experience and how the provider will test against the access and network constraints of your environment.

Can they explain where your data goes?

Request a data-flow diagram covering processing locations, subprocessors, storage, logging and retention. Ask how access rights are enforced and which responsibilities remain with your organisation.

How is success measured?

Agree the metric before the work starts: handling time, automation rate, answer quality, error rate, cost per case, or a business KPI. Then agree what result would justify stopping.

How do they evaluate model quality?

Ask for representative evaluation cases, acceptance thresholds and results by failure category. Check how model and retrieval changes are tested alongside ordinary software tests before release.

What happens after launch?

Agree on monitoring, model versioning, rollback, incident handling and documentation. Name the operational owner and define support hours and escalation times in the contract.

What would they not solve with AI?

Ask for an example where the team recommended ordinary software, a process change or stopping a pilot. The answer should explain the evidence and the trade-off.

Assess Swiss data protection during discovery

The Federal Data Protection and Information Commissioner states that the Swiss Federal Act on Data Protection applies to AI-supported processing of personal data. Transparency and, for processing likely to present a high risk, a data protection impact assessment need to be considered when planning the system.

As of September 2026, Switzerland has no general AI act; a consultation draft is expected by the end of 2026. Swiss organisations should also assess whether their activities fall within the EU AI Act's scope. Article 50 transparency obligations apply from 2 August 2026, while parts of the high-risk regime have later application dates.

Document the data the system needs, where it is stored, who receives it and which users may access each record. Define the actions an agent may take and the approvals required. Resolve legal interpretation with the relevant privacy and legal specialists, and reflect the resulting requirements in the architecture.

Compare strategy, model and implementation specialists

Strategy only

Useful for assessing priorities and developing a roadmap when an implementation team is available.

  • Workshops, use case maps, business cases
  • Ask who builds it, and get a name
  • Works when you have a strong internal engineering team

Models only

Useful for specialised model and data work. Confirm who owns the surrounding application and its integrations.

  • Data science, fine-tuning, evaluation
  • Ask about identity, permissions and deployment
  • Works when integration is already solved

End to end

Covers assessment, engineering and operation. Check the named team, subcontractors and responsibilities at each stage.

  • Use case selection through to running the system
  • Software, cloud and DevOps alongside model work
  • Ask for a case where they recommended stopping

How we approach it at Alpine Edge

Alpine Edge combines AI implementation, software engineering, cloud infrastructure and DevOps. That lets us work on the model alongside authentication, company data, internal APIs and deployment. We agree on scope and ownership during the assessment, including what your team will operate afterwards.

  1. 01
    Audit

    Understand before proposing

    • Use cases scored on value, feasibility and data availability
    • Data sensitivity, storage locations and who may access what
    • Existing infrastructure, identity and integration points
    • Compliance requirements, including FADP obligations
  2. 02
    Build

    Implement and integrate

    • Self-hosted or private LLMs where data cannot leave
    • Retrieval over your own documents, with permissions preserved
    • MCP servers and agents connected to approved tools and systems
    • PII detection and redaction, encryption, role-based access, audit logging
  3. 03
    Run

    Operate and maintain

    • Monitoring, logging and cost tracking
    • Model versioning and rollback
    • Iteration against the metric agreed at the start
    • Knowledge transfer, so the capability stays with your team

For a private AI deployment, the handover includes the surrounding application as well as the model: data access, retrieval, integrations, permissions, monitoring and operating instructions. Those components need clear owners and maintenance arrangements.

Choose a partner whose proposal explains who will build the system, how you will judge it and who will keep it running.

AIConsulenzaSvizzera

Domande, risposte

The deliverables depend on the contract. Discovery may produce prioritised use cases, a data assessment and an architecture. Implementation should add source code, integrations, evaluation results and deployment configuration. Operational support may include monitoring, runbooks and training. Specify which outputs you will receive and own.

A data scientist concentrates on data analysis, statistical methods and building or evaluating models. An AI consultant works wider: business goals, use cases, technology selection, data, project steering, governance and delivery. On larger projects both roles work together, usually alongside data engineers, software engineers, cloud or DevOps specialists and domain experts.

You do not need a finished data platform or an AI strategy already in place. Making those gaps visible is part of the job. What helps is a real business problem, an internal sponsor, and people from both the business and IT who can make decisions and explain how the affected process actually works.

There is no universal answer. External cloud and AI platforms can speed up development and scaling. Private or self-hosted AI makes sense when data control, infrastructure or integration requirements are strict. The decision depends on the data categories involved, the business process, performance, cost, governance and your regulatory context.

The Federal Act on Data Protection applies when an AI project processes personal data. Assess transparency, data access, retention, international transfers and any need for a data protection impact assessment. As of September 2026, Switzerland has no general AI act, with a consultation draft expected by year-end. Confirm the requirements for your use case with qualified privacy or legal specialists.

Request a priced first phase with named deliverables, assumptions and acceptance criteria. Compare the total work required, including data preparation, integration, evaluation and support, as well as day rates. Use the first phase to decide whether further development is justified; there is no single useful price for every AI consulting engagement.

Stai lavorando a qualcosa di simile?

Alpine Edge costruisce e gestisce questo tipo di sistema per clienti in tutta Europa e MENA. Raccontaci cosa stai cercando di risolvere e ti diremo come lo affronteremmo.

Parla con un ingegnere

Leggi dopo

Tutti gli articoli