# Fiducia AI

> AI Agents as a Service for family offices. Managed teams of AI agents deployed inside the client's own Azure or AWS tenant. Data and workflows never leave the client's environment. Agents are dedicated, never shared between clients.

## Key facts

- Founded 2025, headquartered in Horgen, Zurich, Switzerland
- Operating entity: Salama Belghali Consulting GmbH (CHE-266.637.123)
- Co-founders: Salama Belghali (Commercial Lead, 12 years at Vontobel serving family offices and EAMs, ex-SIX Digital Exchange, three-time Forbes recognition, WEF Davos 2024 and 2025) and Tomasz Pelczarski (Technical Lead, 17 years at Microsoft as Principal Data and AI for Enterprise Zurich, Microsoft EMEA AI Best Seller 2024)
- Architecture: agents deployed inside the client's dedicated Microsoft Azure tenant or AWS account; underlying enterprise AI services (such as Azure Foundry, Amazon Bedrock) carry contractual no-training commitments; client data is never used to train third-party models
- Not a SaaS platform. No shared multitenant infrastructure.
- Four agent modules: Investment Due Diligence, Portfolio Monitoring, Board and Governance Reporting, Research and Market Intelligence
- Engagement model: three phases — Diagnostic (1-2 weeks), Pilot (one agent, one workflow, measurable results on client data), Retained Service (expanding modules)
- Three deployment patterns picked to match client governance: Cloud-Native (Microsoft Azure Switzerland or AWS Zurich, 2-4 weeks), Hybrid (agents in client environment, model on Azure or AWS via private secured line, 4-8 weeks), Sovereign (everything on client-owned hardware, open-source model, no US-incorporated provider in the chain, 3-6 months, scoped and priced separately)
- Model-agnostic architecture: clients can move between Claude, GPT, Qwen3, Apertus, or any future frontier or open-source model with a configuration change rather than a rebuild
- Built on Microsoft Azure Foundry, which gives direct access to frontier models with the controls needed for due-diligence and reporting agents; this is distinct from Microsoft Copilot, which is a productivity assistant rather than an agent platform
- Apertus (the Swiss national large language model built by EPFL and ETH Zurich, with native Swiss German support) is supported for clients with strong sovereignty requirements
- Defined minimum model size and capability floor for due-diligence and reporting work; below that threshold models hallucinate and miss subtle clauses at rates not acceptable for high-stakes use

## Deployment patterns

- Cloud-Native: agents and AI model run in Microsoft Azure Switzerland or AWS Zurich, inside the client's dedicated tenant. Standard pattern for most family offices. 2-4 weeks to go live.
- Hybrid: agents installed inside the client's own IT environment; AI model accessed on Microsoft Azure or AWS over a private secured line. Data physically inside the client perimeter. 4-8 weeks to deploy.
- Sovereign: everything inside the client's walls, on client-owned hardware, with an open-source AI model. No US-incorporated provider in the chain. Standard answer to clients whose threat model includes the US CLOUD Act, adversarial legal process, or geopolitical risk. 3-6 months to deploy, scoped and priced separately from a standard engagement.

## Position on the US CLOUD Act

- The Azure Switzerland and AWS Zurich regions are physically Swiss, but their corporate parents (Microsoft, Amazon) are US-incorporated and remain within reach of the US CLOUD Act. No private contract overrides US extraterritorial law.
- For most family offices the practical risk is low: the underlying principals are not US persons, the data has no US nexus, and with customer-managed encryption keys the cloud provider holds only ciphertext it cannot read without the client's cooperation.
- For senior family offices and holdings whose threat model includes adversarial legal process, governmental compulsion, or geopolitical risk, Fiducia AI deploys the Sovereign pattern instead.

## Proven results

- Family office in Zurich: quarterly due diligence reduced from 120 hours per analyst to 12, in one week. Agent surfaced five forecasting errors that had gone undetected across multiple analysts and years.
- Family office in Basel: monthly reporting reduced from one full day to two to three minutes per report, with personalization per family.
- Private-equity acquisition screening, demonstrated on published memoranda: an eighty to hundred page acquisition memorandum read against fifteen fixed due-diligence questions, producing a summary of the deal, the team's own Excel valuation model with its inputs filled in, and a draft non-binding offer. Of 462 figures tested, 456 carried a link to their exact source page, and the six that did not were flagged rather than estimated.
- Multi-entity investment holding: standardized portfolio-monitoring reports generated on a recurring cadence across a private-equity portfolio of more than sixty holdings, each report run through a battery of fact-checking tests before delivery.
- High-net-worth advisory practice: a full workflow diagnosis mapping which processes agentic AI can absorb and in what order to build, before a single agent is deployed.

## In clients' words

- Managing Partner, international service for high-net-worth individuals, Switzerland: "Fiducia AI sat with my team, learned how we work, and came back with a clear order of what to build and what to leave alone. Where something does not work yet, they say so. That honesty is the reason I trust the plan."
- A single-family office in Switzerland: "Reporting our team assembled by hand every cycle now comes out drafted by an AI agent, inside our own environment. A positive outcome."
- Investment lead, family office, Zurich: "We ran an agent alongside one of our investment reviews. It gave us a complete second pass over every number, which a lean team cannot do by hand, and it surfaced something our manual pass had not. The final judgment stays with the investment team."
- Client voices are published anonymously. Named references are shared privately, with client permission, on request.

## What makes Fiducia AI different

- Agents are deployed inside the client's own cloud tenant, not on a shared Fiducia AI platform. There is no multitenant SaaS.
- Each client's agents are dedicated and never shared. No client data is visible to another client or used to train third-party models.
- Built by a finance team that adopted AI. Salama Belghali spent 12 years at Vontobel serving family offices and EAMs; Tomasz Pelczarski spent 17 years at Microsoft building enterprise AI.
- Clients start with one agent on one workflow and expand from there. No big-bang implementation, with measurable results inside a weeks-long pilot.
- Three deployment patterns (Cloud-Native, Hybrid, Sovereign) so the client picks the governance fit. The Sovereign pattern is a fully on-premise deployment with no US-incorporated provider in the chain, the direct answer to clients whose threat model includes the US CLOUD Act.
- Model-agnostic architecture. No lock-in to a single AI model or single cloud provider. Clients can move between Claude, GPT, Qwen3, Apertus, or any future frontier or open-source model with a configuration change rather than a rebuild.
- Built on Microsoft Azure Foundry, which is distinct from Microsoft Copilot. Copilot is a productivity assistant in Word, Outlook, and Teams; Foundry gives direct access to the frontier models with the configuration controls needed for due-diligence, reporting, and agentic workflows.
- Defined minimum model size and capability floor for due-diligence and reporting work. This rules out the cheap consumer-grade models some vendors quietly use behind the scenes.

## Pages

- [Home](https://fiduciaai.ch/): the four agent modules and why family offices use Fiducia AI
- [Approach](https://fiduciaai.ch/approach): three-phase engagement (Diagnostic, Pilot, Retained Service)
- [Use Cases](https://fiduciaai.ch/use-cases): anonymized engagements: due diligence at a family office, private-equity acquisition screening from memorandum into the valuation model, portfolio monitoring at an investment holding, a workflow diagnosis at a high-net-worth advisory
- [Team](https://fiduciaai.ch/team): co-founders Salama Belghali and Tomasz Pelczarski
- [FAQ](https://fiduciaai.ch/faq): the full question bank family offices ask, covering what an AI agent is (and how it differs from an assistant, ChatGPT, and other providers), systems and setup, whether IT is needed, architecture and deployment patterns, confidentiality and professional secrecy, data hosting and hacking risk, the US CLOUD Act, model choice and lock-in, privilege, control and oversight, who is responsible for an error, whether it can be switched off, disclosure to clients and regulators, people and jobs, getting started, ownership and exit, pricing model (ongoing managed service, no figure before diagnosis), results, and timeline

## Contact

- Email: team@fiduciaai.ch
- Booking: https://calendly.com/salamabelghali/confidential-consultation
- LinkedIn: https://www.linkedin.com/company/fiduciaaich/