Claude for banks, wealth and asset managers
Claude implementation partner for financial services
Claude Certified Services Partner(Anthropic)S-PRO is a Zurich-based software company that builds production AI agents on Claude (Anthropic) for banks, private banks, wealth managers and asset managers. S-PRO is a Certified partner in the Claude Partner Network, and our team includes Claude-certified engineers and architects. We take one use case from a fixed-scope pilot to a governed production agent, with evaluations, audit trails, human-in-the-loop approval and a deployment designed for FINMA, nFADP, GDPR and EU AI Act requirements, with the compliance decision staying with your institution.
01
Why financial institutions choose S-PRO for Claude
- Financial services focus. Since 2014 we have built software for banks, wealth managers and fintechs, including Swiss-regulated institutions such as AMINA Bank.
- Production, not demos. Multi-step agents with evaluations on every agent, human sign-off and a full audit trail.
- Your data stays under your control. Claude runs through the Anthropic API, Amazon Bedrock or Google Cloud (Gemini Enterprise Agent Platform, formerly Vertex AI); data residency options (route, region setting, pseudonymisation, logging location), isolation and access boundaries are agreed per project.
- Certified security. ISO 27001 and ISO 27701.
- Model-agnostic where it matters. If a use case needs Azure OpenAI or an on-premise model, the same governance layer applies.
- Claude Partner Network. S-PRO is a Certified partner in the Claude Partner Network, and our team includes Claude-certified engineers and architects.
02
What we build on Claude
Swipe to compare →
| Agent | What it does | Typical users |
|---|---|---|
| KYC & onboarding agent | Researches the client, drafts the storyline, checks source-of-wealth gaps, routes the file to a human approver | Onboarding, compliance |
| Periodic review agent | Answers review questions per client with guardrails and sign-off | Compliance, RMs |
| Adverse media & sanctions screening | Triages hits and gives a verdict with a supporting quote, routed to the RM | Financial crime |
| Credit memo agent | Extracts documents, calculates ratios, drafts the memo against policy | Credit |
| RM copilot | Pre-call briefings, call-report drafts, portfolio questions in natural language | Private banking, wealth |
| Investment research agent | Collects filings, news and market data and drafts research notes | Asset managers |
| Document processing | Extracts and classifies data from contracts, statements and forms | Operations |
| Client support agent | Drafts replies from account history and policies for a human to approve | Client service |
Not sure where your bank stands on AI agents? Take the 3-minute readiness check.
Start the readiness check03
How a Claude project with S-PRO works
- Free discovery workshop. We sort your use cases into what works today, what is easy to configure and what needs more maturity.
- Architecture. We map your systems (core banking, CRM, KYC, document repository) and design the agent, data and governance layers.
- Design thinking sprints and AI prototyping. 2–4 weeks to validate hypotheses with users and agree an effective scope.
- Solution pilot. The full solution on a real process within up to 2–3 months depending on complexity, with clear go/no-go criteria.
- Production. Evaluations, monitoring, role-based access, audit logging and model-risk documentation.
- Scale. New agents reuse the same governed data foundation.
Reference architecture
Interface (Microsoft 365 Copilot, Teams, web, your apps or Claude itself) → agent layer (Claude-based agents with tools) → governed data layer (one versioned client file, entity resolution, PII access control) → systems of record (core banking, CRM, KYC, documents, screening data) → governance across every layer (observability, auditability, testing, role-based access).
Guide
How does S-PRO work: our own agents platform plus your solution?
S-PRO has its own agentic platform, S-PRO Agents, which is the core of our products and client solutions. Around it we build the solution for your bank and deliver it as a service. In a Claude project, the work goes into your processes, your data and your approval points.
| What | From S-PRO | Built for your bank |
|---|---|---|
| Platform | S-PRO Agents, the base of our own products: RegRadar (regulatory change monitoring), Microsoft Copilot Booster (an agents hub that connects to Microsoft Copilot) and AI SDLC automation | The solution around it for your workflow, for example KYC review, support drafting or policy answers, delivered as a service |
| Control | Human approval points designed into each agent | Your approvers, your policies, your data classification |
| People | Claude-certified engineers and architects; forward deployed Claude engineers on site with your team | Your process owner and compliance contact |
S-PRO is a Certified partner in the Claude Partner Network. Our engineers are officially certified. The model route (Anthropic API, Amazon Bedrock, Google Cloud Vertex AI) is covered in the route table below. For the full stack and delivery questions see AI agent development for banks and AI process automation.
04
Which deployment routes exist for Claude in a bank?
A bank can run Claude through three routes: Anthropic's own API, Amazon Bedrock and Google Cloud Vertex AI. Each provider documents Claude access on its platform. The right route depends on where the bank already runs its workloads, which models and regions that route offers, and what the bank's security and legal teams accept.
- Anthropic API. Direct, first-party access. Anthropic documents its data-residency controls; the options change, so they are checked against the current documentation at the start of each project.
- Amazon Bedrock. Claude inside the bank's AWS environment, with its access control and logging. AWS sets model availability and cross-Region inference profiles per model.
- Google Cloud Vertex AI. Claude inside the bank's Google Cloud environment. Google sets model availability and processing locations per model and endpoint.
Residency options are agreed per project: route, region or endpoint setting and logging location. FINMA Guidance 08/2024 on governance and risk management when using AI expects institutions to identify, limit and control AI risks, and names growing dependence on third parties such as model and cloud providers among them; the compliance decision stays with the institution. Where each route processes data and what a Swiss bank should check are covered in Can a Swiss bank use Claude? Data residency, FINMA and client data.
Guide
Which route should a bank use to run Claude?
Choose the route by where the bank's data-processing contract, cloud footprint and approval process already sit, not by the model. The routes most banks consider are the Anthropic API, Amazon Bedrock and Google Cloud (Vertex AI); Microsoft Foundry is a further option for banks that standardise on Azure. For a regulated bank, the route decides who is the data processor, which regions are available and how access and logging are controlled, so settle it before the first build.
| Route | Fits when | Ask before choosing |
|---|---|---|
| Anthropic API | The bank can contract Anthropic directly and wants the most direct access | Data-processing terms, retention settings, available regions, subprocessors |
| Amazon Bedrock | The bank already runs on AWS and wants Claude inside its existing AWS governance | Global or regional endpoints (Bedrock documents regional endpoints as guaranteed data routing), available models, identity and access setup, logging; published example: the neobroker project uses Anthropic models via Amazon Bedrock (version, region and hosting not published) |
| Google Cloud Vertex AI | The bank already runs on Google Cloud | The same questions; Google documents global, multi-region and regional endpoints, so ask which one applies and check regional availability per model with the provider |
S-PRO's view. The route can be changed later if the agent talks to the model through one interface layer. For Swiss data-residency questions see Claude for Swiss banks: data residency.
Guide
What should be in place before a Claude agent goes live at a bank?
| Check | What to ask | What good evidence looks like |
|---|---|---|
| One process, one owner | Which process, and who owns the result? | A written scope and a named business owner |
| Evaluation | How is quality tested before go-live? | An evaluation set and logged test runs, not a demo |
| Data classification | Which client data may reach the model, and where is it processed? | A data classification and a documented route (see the route table above) |
| Human approval | Who approves what reaches a client or changes a record? | A named approver per action; the agent prepares, a person decides |
| Controlled access | How does the agent reach core and document systems? | Controlled interfaces (APIs, MCP), read-only first, write actions added one at a time |
| Logging and audit | Can every answer be traced to its sources? | Logged inputs, sources and approvals |
| Rollback and handover | What if quality drops? Who runs it after go-live? | Rollback path, monitoring, client ownership of the result |
Use this list for any Claude project, including ours. S-PRO is a Certified partner in the Claude Partner Network, our team includes Claude-certified engineers and architects, and our engineers are officially certified. See also security and compliance for financial institutions and taking an AI agent from pilot to production.
Which published projects use Claude?
A leading German neobroker
The Zendesk plugin uses Anthropic models via Amazon Bedrock for two tasks, classifying tickets and drafting replies; the operator checks and sends.
Read the case study
For the wider map of Claude partners in Switzerland see Claude implementation partners in Switzerland.
Models, version, region and hosting are not published for this client.
05
Compliance by design
- EU AI Act: AI systems are classified by risk (Regulation (EU) 2024/1689). We classify each agent during design and build technical documentation, logging and human oversight in either case.
- FINMA and nFADP: outsourcing and data-protection requirements are agreed contractually per project. Responsibility for compliance stays with the supervised institution; S-PRO does not give regulatory advice.
- DORA: for clients that are EU financial entities, DORA (Regulation (EU) 2022/2554) requires ICT third-party risk management and specific contractual provisions with ICT providers such as S-PRO.
Results, already in production
Proof from production
AI support agent for a leading German neobroker
The agent drafts each reply from account history, order status and policies, and a support agent approves it. A person signs off on every AI draft before it is sent.
Policy answers in 30 seconds for a Swiss bank
Employees of the Swiss subsidiary of an international bank search internal policies and procedures through a role-based AI assistant that shows the source of every answer.
Earlybird
An AI platform for one of Europe's most established venture capital firms: data aggregation and analysis, automated due diligence and risk prediction for investment decisions.
View caseFAQ
Questions buyers ask us
Is S-PRO an Anthropic partner?
S-PRO is a Certified partner in the Claude Partner Network, and our team includes Claude-certified engineers and architects.
Can a Swiss bank use Claude with data residency requirements?
Yes, with the design agreed per project. We deploy Claude through the Anthropic API, Amazon Bedrock or Google Cloud Vertex AI. Residency options (route, region or endpoint setting, pseudonymisation of client data, logging location) are agreed per project with your security and legal teams, who decide what meets FINMA and nFADP expectations for the institution.
How long does it take to get a Claude agent into production?
We start with design thinking sprints and AI prototyping (2–4 weeks) to agree an effective scope. Building the full solution and running the pilot then takes up to 2–3 months. The production timeline depends on integrations and internal approvals.
How much does a Claude implementation cost?
A design thinking pilot (workshops, prototype and user testing) starts from CHF 20k. The scope and budget of a full solution pilot depend on complexity. Production budgets depend on complexity. Production scope and price are agreed per project; we prepare the estimate after discovery, once scope, data and controls are clear.
Do you only work with Claude?
No. We are model-agnostic and also build on Azure OpenAI and on-premise models, under the same governance layer. We recommend Claude for long-document, reasoning-heavy and agentic workflows.
Has S-PRO used Anthropic's models through Amazon Bedrock?
Yes. For a leading German neobroker we built a Zendesk app on Anthropic's models through Amazon Bedrock. It classifies each ticket and drafts a reply that an operator checks, edits and sends. Reported results: in 85% of cases the agent's draft reply was not wrong, 90% correct ticket classification and more than 40% faster ticket handling.
Does a person approve what Claude writes to customers?
Yes. In both support deployments we describe on this page, a person reviews every AI draft before it is sent. The agent prepares the answer; the operator remains accountable.
Where is S-PRO based?
S-PRO is headquartered in Zurich, Switzerland (Löwenstrasse 45, 8001 Zürich), with R&D centers in Zurich and Łódź and business representatives in the US and the Netherlands.
How do you implement Claude in a bank, step by step?
Pick one process with a named owner and an approval point. Classify the data and choose the route (for example Anthropic API, Amazon Bedrock or Google Cloud Vertex AI). Build with controlled interfaces, read-only first. Test on an evaluation set before go-live, run a pilot, then pass production gates with monitoring and a rollback path, and hand the result over to the bank. In S-PRO's view, the route and the approval point are the two decisions that are expensive to change later.
Does the choice between the Anthropic API, Amazon Bedrock and Google Cloud Vertex AI change what we can build?
The use cases are the same; what changes is the contract, available regions, access control and logging. Choose the route your security and procurement teams already know. The neobroker project runs on Anthropic models via Amazon Bedrock.
What should a bank have in place before a Claude agent goes live?
One owner, an evaluation set, a data classification, a named approver for every outward or record-changing action, controlled read-only access first, logging of sources and approvals, and a rollback and handover path. The table above lets you check any partner against it, including S-PRO.
What does "partnering with S-PRO" for Claude agents mean in practice?
You work with one team that has its own agentic platform, S-PRO Agents, the core of its products and client solutions, and builds the solution for your institution around it, delivered as a service. The work follows the six steps above, from discovery workshop to production.
What are you building?
Contact
You will talk to engineers who build for banks, not to a sales script. Share your ideas, findings and questions with us.
- Security and privacy firstISO 27001 and ISO 27701 certified.
- Claude Partner NetworkS-PRO is a Certified partner in the Claude Partner Network. Our team includes Claude-certified engineers and architects.
- Proven in regulated financeBanks, wealth managers and fintechs in Switzerland, the EU and the US.
Who you’ll talk to
Michael BarskyiHead of Banking & Fintech
Dustin SchulzeSenior Business AnalystClaude Certified Services Partner(Anthropic)Contact us
Send us your request
Share your ideas, findings and questions with us.