Swiss2S
Google Cloud specialists · Switzerland

AI, data and cloud, in production.

Swiss2S is a senior team of experts that designs, builds and runs AI agents, data platforms and cloud environments on Google Cloud, with the AI models of your choice. With safeguards your security and finance teams can read for themselves.

  • Google Cloud
  • BigQuery
  • Vertex AI
  • Claude
  • GPT
  • Gemini
  • Mistral
  • MCP
  • Terraform
  • Python
agent · accounting
  1. 1
    You ask“Where do we stand on this quarter's VAT?”
    Question
  2. 2
    The connector queriesAccounting · journal and VAT returns
    Read
  3. 3
    The safeguard appliesRead-only · nothing is changed
    Allowed
  4. 4
    The answer comes backDeclared revenue CHF 248’600 · 3 entries to check
    Sourced

AI budget this month38%

Every answer is sourced

Illustrative examples, fictional figures

50business screens delivered in a single console
10AI modules built into one platform
32MCP servers in production
1’000+sets of minutes produced by agents
References
  • UC Berkeley School of Public Health
  • The Forum for Collaborative Research
  • Paris International Liver Meeting
  • Paris MASH Meeting
  • JHC Meeting
  • CHEP Meeting
  • A European energy leader
  • A global telecom operator

01 · Services

Three ways to work together.

For mid-sized and large companies that want a working system, not another study.

Flagship offer

Make your systems talk

An MCP connector links Claude, ChatGPT or Gemini to your applications: ERP, accounting, CRM, business databases. Your teams ask in plain language and get a sourced answer. Anything that writes goes through an approval.

  • A first connector in service on one of your tools
  • Works with Claude, ChatGPT, Gemini and Le Chat
  • Read by default, writing under approval, every call logged
  • Scheduled agents: triage, summaries, drafts

Data platforms in your cloud

From collection to dashboard: a documented data model, versioned business rules and forecasts that can be explained. On BigQuery, or on the warehouse you already run.

  • Data collection and warehouse
  • Dashboards and business applications
  • Forecasting, anomalies, marketing mix
  • Analyses combining business data and open data

Your cloud under control

On Google Cloud, a foundation that security and finance both accept: least-privilege access, infrastructure as code, costs under control.

  • Architecture, access rights, network, Terraform
  • Budget alerts and kill switch
  • Hosting in Switzerland or in the EU, with the Swiss FADP and the GDPR in mind

02 · Safeguards

What makes AI acceptable in a company.

Four rules applied to every project. They can be checked in the code, not in a charter.

Approval

The model proposes, a rule authorises.

Anything that commits the company (writing, sending, paying) goes through a preview approved by a person, or through a rule written without AI.

Traceability

Every figure has a source.

A fact points to the document or the moment it comes from. What has no source is not written.

Budget

A cap, an alert, a kill switch.

The cost of each agent is logged. An overrun triggers an alert, then a reversible shutdown.

Sovereignty

Your data stays with you.

Everything is deployed in your own cloud project, in the region you choose, under your access rights. You keep the code.

03 · Work

Five examples of systems designed and built by our team.

Each one answers a concrete need and carries its own safeguard.

Events

Congress platform

The full management of an international medical congress, with AI in every module.

In service forParis International Liver Meeting · Paris MASH Meeting

The need
A small team handles registration, payments, speakers, sponsors, programme and budget for several congresses, without multiplying tools.
What runs
One console shared by several congresses. AI reviews, checks and prepares the team's work, and any assistant (Claude, ChatGPT, Gemini) queries the database through a connector.
The safeguard
AI proposes, the organiser approves: no correction applies on its own, and each one can be undone. The connector is read-only. Every question and every call is logged.
10 AI modules
  • Badge name audit
  • Fee audit
  • Programme check
  • Invoice reading
  • Travel ticket reading
  • Dates extracted from emails
  • Requests received by email
  • Marketing studio
  • Console help assistant
  • Connector for assistants
Fee audit screen: the AI looked up a registrant on the student fee, found he runs a private practice, cited its sources and drafted an email; the organiser decides. Marketing studio screen: speaker visual generated from the programme and a LinkedIn post drafted by Gemini, ready for review.
Registrants map (animated reproduction), fee audit and marketing studio · demonstration data, fictional names
  • Gemini
  • MCP
  • App Engine
  • Stripe
Finance

Accounting assistant on bexio

The books of a Swiss company, queried and fed by AI.

The need
Keep accounts in francs and euros without retyping, and get an answer without opening ten screens.
What runs
A connector gives your AI assistant (Claude, ChatGPT, Gemini) read access to all of bexio: chart of accounts, journal, balances, VAT, invoices, documents.
The safeguard
Only two write actions, always preceded by a preview to approve. Nothing can be changed or deleted.
  • MCP
  • Cloud Run
  • Firestore
  • Secret Manager
Knowledge

Company memory

Meetings and client exchanges turned into a knowledge base.

The need
Find what was said, decided or promised, months later, without rereading notes.
What runs
Scheduled agents turn each meeting into minutes, then keep summaries up to date by project, client and person. More than a thousand sets of minutes so far.
The safeguard
Every fact points to its source. Two model providers with automatic failover, and a capped monthly budget.
  • Claude
  • Gemini · Vertex AI
  • Cloud Run Jobs
  • BigQuery
Marketing

Media cockpit

Steering advertising spend from your own warehouse.

The need
Know where the budget goes, what it returns and where to move it, without relying only on the agency's figures.
What runs
One product, configured for each client: BigQuery warehouse, web application, marketing mix model, 90-day forecast and anomaly detection.
The safeguard
Data stays in the client's cloud project. Business rules are versioned: every change is dated, reviewed and reversible.
Overview screen of the media cockpit: spend, conversions, cost per acquisition, daily spend by ad platform and the list of checks in alert. Allocation screen of the media cockpit: return on investment per channel with credibility interval, and the response curve of one channel.
Overview and allocation · demonstration data, fictional brand
  • BigQuery
  • Meridian
  • Cloud Run
  • Terraform
Healthcare

Congress reports

From the room's video feed to a sourced report, designed to deliver within the hour.

The need
Publish the report of a scientific session during the congress, with no mistake on a figure or a molecule.
What runs
Feed capture, transcription, correction of technical vocabulary from the slides, then writing in three passes with review by a second agent.
The safeguard
A check written without AI verifies that every figure in the text exists in a cited fact. No bibliographic reference is invented.
  • Gemini
  • Firestore
  • PubMed · Crossref

04 · Method

From base camp to the summit, in four steps.

Each step ends with something you can use, on a date agreed in writing. We keep it. You decide what comes next at the end of each one.

1

Base camp

The diagnosis, as a short scoping: your use cases, your data, your security constraints. You leave with a prioritised plan and a clear decision.

2

Prototype

A working prototype on your own data, in your own cloud project. Your teams test it on real questions.

3

Go live

Security file, access rights, deployment, monitoring. We go through the IT, security and procurement reviews with you.

4

Run and hand over

We operate the system, document it and train your teams. The code and the data stay yours, and the handover is planned.

05 · Expertise

A core expertise: marketing data for large companies.

Swiss2S supports marketing and digital departments on their customer data programmes: choosing the tools, getting them accepted by IT, security, procurement and legal, then putting them into service.

The advice is independent of vendors and agencies.

  • CDP and data clean room tenders: specifications, scoring, vendor hearings
  • Server-side collection and conversion tracking
  • Advertising exclusion of existing customers
  • Marketing data warehouse and audit of advertising accounts
  • Measurement: attribution and marketing mix
Energy A European energy leader

CDP and data clean room tenders, marketing data warehouse on Google Cloud, server-side collection, media performance management.

Telecom A global telecom operator

Big data and audience analytics for television, recommendation engine, marketing for the launch of its mobile bank, certified audience measurement tools.

Sectors

  • Energy
  • Telecom
  • Banking
  • Insurance
  • Media
  • Retail
  • Industry
  • Healthcare
  • Public sector
  • Events

06 · Team

Three leads, and a team sized to your project.

Every project is led by three senior leads: architecture, data and delivery. Around them, the team grows with the specialists the project calls for. The people who scope your project are the ones who build it.

Lead · Google Cloud and AI

Cloud and AI architecture

Designs the architectures and writes the code that goes into production. Also carries the case to IT, security, procurement and legal, until they sign off.

Lead · Data

Data analysis

Combines your business data with open data to reveal what your dashboards do not show. Writes Python, deploys in the cloud and adds the AI components that are needed. Models and forecasts that can be explained to an executive committee.

Lead · Delivery

Project management

Holds the scope, the schedule and the budget. Decisions on record, regular progress reviews, and a go-live prepared from day one.

Swiss2S stands for Strategy to Systems, the Swiss way. We start from your strategy and stay until the system is in service, with the precision, reliability and discretion expected from Switzerland.

This format has a direct consequence for you: decisions made in days rather than in committees, no intermediary between your need and the people who deliver it, and one goal: a result in service.

Company
Swiss2S GmbH, Basel, Switzerland
Certification
Google Cloud Certified, Professional Cloud Architect
Languages
French, English

08 · Your questions

The questions your IT, security and finance teams will ask.

Here are our answers, before the first meeting.

01What happens if a key person is unavailable?

The project does not rely on word of mouth: versioned code, infrastructure described as code, documented decisions, standard Google Cloud services. Another engineer can take over from the repository.

02Who owns what is built?

You do. The code is delivered in your repository and the system runs in your cloud project, under your accounts. Your teams or another provider can take it over.

03Where is our data, and who can access it?

In your own cloud account, in the region you choose: Switzerland or the European Union. Access is named and limited to what is needed.

04How do you get through our security review?

We are used to the security and architecture reviews of large groups: architecture file, data flow analysis, least-privilege access, logging. We prepare the file and defend it with you.

05What if the AI gets it wrong?

It proposes, it does not decide. Anything that writes, sends or pays goes through an approval or a rule written without AI. Every answer cites its source, and every call is logged.

06What does it cost to run?

Each agent has a capped budget and its consumption is tracked. You see the cost per use, and an overrun triggers an alert, then a shutdown.

07Are you tied to one vendor?

For infrastructure, by choice: we are Google Cloud specialists, certified, with a region in Zurich. For AI models, no: Anthropic, OpenAI, Google or Mistral, or open-source models you host yourself. And our connectors use MCP, an open standard: they reach your systems wherever they run, including on AWS or Azure.

09 · Contact

Let's talk about your project.

Describe your need in a few lines, or call us. You get a reply within two working days.

What happens next

  1. A first conversation to understand your need
  2. A base camp proposal: scope, duration, price
  3. Kick-off, on the agreed date