Services

Artificial intelligence consulting

Independent AI consulting, from discovery workshop to production. Together we choose the use cases worth pursuing and the right model, including open-source models hosted in Switzerland.

Artificial intelligence consulting

Most AI projects do not fail because of the model, but because the use case was poorly chosen or nobody had set the rules. Our work therefore starts with your processes: where time goes, which decisions repeat, what data is involved. The choice of model comes afterwards, and it is not tied to any vendor.

Why independent consulting

An OpenAI model on Azure, a model from Anthropic or Mistral, or an open-source model such as Llama, Qwen or Gemma each have different strengths, different costs and different consequences for your data. An integrator tied to a single vendor will offer you its platform. We compare the options on your case, with your documents, and we also tell you when AI is not the right answer.

Sorting files and requests

A model reads each incoming file and sorts it according to your rules: those that can follow a simple route and those, fewer in number, that need a human eye. Low-value claims, supplier invoices, customer requests, compliance alerts: the principle is the same, and the rules are yours.

Document extraction and checking

Invoices, contracts, forms, identity documents, reports: the model extracts the useful information, checks that a file is complete and flags inconsistencies. Your teams review the discrepancies instead of re-entering everything.

Internal assistants on your documents

An assistant that answers from your procedures, contracts and files, citing its sources. It can run in Microsoft 365, in your business tools or on infrastructure hosted in Switzerland, depending on how sensitive the content is.

Integration into your tools

Useful AI lives where your teams work: email, ERP, business software, document management. We handle the integration, access rights, logging and monitoring, and we document everything for your IT team.

Our method, from workshop to production

Six steps, each with a deliverable and a clear decision: continue, adjust or stop.

  1. 1. Discovery workshop

    A working session with your business and IT teams. We review your processes, volumes and constraints: professional secrecy, data protection, tools already in place.

  2. 2. Use case selection

    We rank the ideas by value, feasibility and risk, and we keep one or two. Cases where an error would cost too much, or where the rule cannot be written down, are set aside.

  3. 3. Choice of model and hosting

    We test several models, proprietary and open source, on a sample of your documents, and we compare quality, cost and consequences for your data. You receive a reasoned recommendation, with its limits.

  4. 4. Measured pilot

    The pilot runs on a limited scope, often alongside the existing process. The indicators are set before the start and measured on your own figures.

  5. 5. Going into production

    Integration into your tools, access rights, logging, supervision, team training and a manual fallback procedure in case of an outage.

  6. 6. Governance

    Who decides the rules, who changes them, who checks the errors, how often the model is reassessed. We document these points so that the system stays under control over time.

The technologies we work with

Proprietary models: OpenAI (including through Azure OpenAI), Anthropic, Mistral. Open-source models: Llama, Mistral, Qwen, Gemma, deployed on your infrastructure or in a Swiss cloud. Microsoft ecosystem: Microsoft 365, Copilot, Power Platform and Azure, when your teams already work there. We have no exclusivity with any vendor: the list changes as models change, and the choice is always made on your case.

Benefits for your business

Productivity gains

Automate repetitive tasks and allow your teams to focus on high-value activities.

Cost reduction

Lower your operational costs by optimizing your processes and reducing manual errors.

Better decision making

Base your decisions on in-depth data analysis rather than intuition alone.

Competitive advantage

Adopt the technologies that will make the difference against your competitors in the years to come.

Models and hosting: how we choose

There is no best model in general. There is a model suited to a case, to certain data and to a budget. Here is how we reason.

When to choose what

A managed service from a major vendor suits varied, low-sensitivity tasks, small volumes and teams without operations skills. An open-source model hosted in Switzerland becomes interesting when the data is covered by professional or medical secrecy, when the task is repetitive and well defined, when volumes make pay-per-use billing expensive, or when you do not want to depend on a single supplier.

Open-source models hosted in Switzerland

Open models such as Llama, Mistral, Qwen or Gemma can be installed on your own servers or in a data centre in Switzerland. For targeted tasks such as sorting, extraction or summarising, they often give sufficient results, provided they are tested on your documents. Their licence varies from one model to another: we check it before any use.

What "hosted in Switzerland" means in practice

For each project we set four points in writing. Where inference runs: on machines located in Switzerland, either yours or those of a Swiss host. Where the data is: documents, indexes and results stay on that same infrastructure. Who has access: a named list of people, with limited rights. What is logged: decisions and accesses, without keeping more content than necessary. Hosting in Switzerland does not replace your own legal analysis; it simplifies it.

Microsoft, OpenAI, Anthropic or Mistral, when it is the right choice

If your teams live in Microsoft 365 and the content is already there, staying in that environment is often the simplest option. For complex drafting or reasoning tasks, the large proprietary models keep an edge. We then review with you the processing region, the contractual terms and data retention.

What we do not promise

A model sometimes makes mistakes. We do not sell certification, turnkey compliance or full automation. We design systems where the rules are yours, where doubtful cases go to a person and where every decision can be explained after the fact.

Reference guide

AI Agency in Geneva: What Services to Expect and How to Compare Offers

Discover the services offered by an ai agency geneva and learn how to decipher, compare, and evaluate offers to maximize benefits for your SME.

Client projects and typical scenarios

Anonymised client projects and representative scenarios, described without names or quantified results.

Major Swiss medical institution: minor liability claims

Client project. For the legal department of a major Swiss medical institution, a model hosted in Switzerland classifies minor liability claims (lost items, damaged items, dental injuries), checks that files are complete, prepares acknowledgements and a draft reply. A lawyer validates every reply.

Read article

International school: security review of an AI platform

Client project. For an international school in Switzerland, a review of the code, Azure architecture, integrations and AI layer of an in-house platform holding student data: automated analysis, manual review of sensitive areas (authentication, authorisation, secrets, AI), then a report and a workshop with leadership.

Open source hosted in Switzerland or a US subscription

Decision guide. Data, cost, quality, maintenance, skills: the five criteria for choosing between an open-source model hosted in Switzerland and a US AI service, with a comparison table and the limits of each option.

Read article

Small insurance claims

Illustrative scenario. AI sorting separates the claims that can be settled directly from those that need a claims handler. The thresholds remain the insurer's and no refusal is automated.

Read article

Invoices, expense reports and VAT at a fiduciary firm

Illustrative scenario. Each document receives a proposed entry or a reasoned anomaly flag. The accountant validates conforming documents in batches and concentrates on the discrepancies.

Read article

KYC files and AMLA alerts

Illustrative scenario. The model checks that files are complete and ranks alerts by priority with a sourced summary. Closing alerts and any compliance decision remain human.

Read article

Mail and email automation

Illustrative scenario. An AI system classifies each incoming letter and email by type, extracts the key information, routes it to the right file and flags urgent items. Staff can then find any communication through semantic search, and a person validates the uncertain cases.

Read article

Automatic deadline detection

Illustrative scenario. A SharePoint application analyses documents to spot legal and contractual deadlines, then creates calendar events in Teams with progressive reminders. The detected deadlines remain to be confirmed by the person handling the file.

Read article

Intelligent invoice payment reminders

Illustrative scenario. Reminders for unpaid invoices are drafted by AI, with a tone and content adapted to how late the payment is and to the history of the relationship. Each message can be reviewed before it is sent.

Read article

Automatic legal document summarization

Illustrative scenario. A summarisation tool produces two levels of structured summary for reports and contracts, so that readers can quickly decide whether a full reading is needed. The summary does not replace reading the decisive passages.

Read article

Knowledge base with semantic search

Illustrative scenario. A knowledge base with semantic search understands the intent behind questions, and a model generates concise answers that cite their sources. It helps new team members in particular to find information.

Read article

Meeting transcription and action extraction

Illustrative scenario. AI transcription of meetings produces a structured summary and creates a Planner task for each agreed action. A participant reviews the summary before it is shared.

Read article

Typical situations

Fictional, simplified examples of situations we can handle. They do not describe specific client projects or guaranteed results.

Situation: A claims or compliance team receives many files, most of which follow the same path, yet everything is read by hand.

What we do: Workshop to write your triage rules, a pilot with human validation on test files, then production with a decision log; model hosted in Switzerland where the data requires it.

Situation: An SME wants an assistant that answers from its procedures and contracts without those documents leaving Switzerland.

What we do: Cost comparison between a US subscription and an open-source model hosted in Switzerland, model choice, connection to the documents with the existing access rights, quality tests before opening it to the teams.

Situation: Management has received several “AI agency” proposals and cannot tell which are serious.

What we do: Review of the proposals, use cases ranked by value and risk, a step-by-step roadmap with stop criteria.

Let's discuss your project

Contact us to discover how AI can transform your business.