---
title: "Fixed-Price AI Pilot for Wealth Managers: Which Vendors? - S-PRO"
url: https://s-pro.io/blog/vendors-fixed-price-ai-pilot-wealth-manager
lang: en
description: "Which vendors offer a fixed-price AI pilot for a wealth manager? What fixed means, what FINMA expects, and how to compare vendor types before you sign."
updated: 2026-10-28T07:51:27+00:00
source: S-PRO (https://s-pro.io)
---

[Back to insights](https://s-pro.io/blog)

Article

# Fixed‑Price AI Pilot for Wealth Managers: Which Vendors?

Author

[Igor Izraylevych](https://s-pro.io/author/sproadmin)

CEO and Co-founder of S-PRO

Published

October 27, 2026

Time to read

13 min

Table of content

- Key takeaways
- Table of Contents
- Who sells fixed-price AI pilots to wealth managers?
- What makes an AI pilot fixed-scope?
- What should acceptance criteria say?
- How do you compare vendors on a pilot?
- Fixed pilot, custom proof of concept or broader implementation?
- Perspective: what we think a good pilot looks like
- How to run the selection in four steps
- Where S-PRO fits
- FAQ
- How S-PRO helps
- Sources

Share

A pilot is fixed only if scope, data and acceptance criteria are agreed in writing before work starts. A price on a proposal does not make it fixed. A vendor who can’t name the use case, the data it will touch and the test that decides success is quoting a guess, not a fixed pilot. For Swiss wealth managers, the same written terms also line up with what FINMA Guidance 08/2024 (18 December 2024) expects from AI governance: materiality, tests and monitoring, documentation and independent review.

Last updated: 28 October 2026

S-PRO

s-pro.io

Move Your AI Pilot Toward Production

S-PRO builds AI solutions for wealth managers, including AI pilots to production designed for financial services requirements.

[Explore S-PRO’s AI services](https://s-pro.io/)

## Key takeaways

- “Fixed” describes the agreement, not the invoice: use case, data, tests, documentation, independent review and exit decision are written down before work begins.
- Vendors that sell pilots fall into a few types, and each defines “fixed” differently. Compare the type first, then the individual supplier.
- FINMA Guidance 08/2024 gives you a ready checklist for the contract: materiality, tests and monitoring, documentation, independent review.
- A pilot that passes its tests is still not a production system. Ask in advance what changes between the two.
- Price is a conversation to have after scope is fixed. Ask each vendor for its pilot terms directly; we publish none on this page.

## Table of Contents

- What makes an AI pilot fixed-scope?
- What should acceptance criteria say?
- How do you compare vendors on a pilot?
- How to run the selection in four steps
- Where S-PRO fits
- FAQ
- How S-PRO helps
- Sources

## Who sells fixed-price AI pilots to wealth managers?

No single list answers this, because wealth managers buy pilots from different kinds of suppliers, and the term “fixed-price” rarely means the same thing twice. The pages that rank for this question mostly review consumer-facing AI advisors and robo-advisor apps. They don’t help a Swiss independent wealth manager or private bank choose a delivery partner for an internal pilot. So we compare by vendor type.

**Product vendors with a packaged pilot.** Software suppliers with a ready product for advisor notes, research summaries or client reporting often offer a short trial on a standard configuration. The scope is fixed because the product is. The trade-off: your workflow adapts to the product, and integration with your portfolio management system, CRM and custodian feeds is usually limited in a pilot.

**Large consultancies.** They tend to run structured discovery and proof-of-concept phases. Scope can be tightly written, but the pilot is often a stage in a larger programme, and the contract may assume the follow-on work.

**General AI model providers and cloud platforms.** They supply models and tooling, and you or an integrator build the use case. Their commercial terms cover usage of the platform, not the outcome of your pilot. Hosting location, subprocessors and data retention need to be checked against your own confidentiality duties.

**Specialist engineering firms for regulated finance.** These firms build custom agents and copilots around your existing systems. Scope is shaped to your data and workflow. Whether the commercial form is fixed or time-and-materials depends on how well the use case is defined, which is why the written scope matters more than the label.

**Boutique AI agencies.** Quick to start and often inexpensive, but check whether they can document the pilot to the standard a supervised institution needs, and whether they will support an independent review.

In each case, ask what the vendor means by fixed. Fixed price, fixed scope and fixed duration are three different promises. A supplier can honestly offer one without the others.

## What makes an AI pilot fixed-scope?

A fixed-scope pilot has six elements that exist in writing before the first sprint. If any is missing, the vendor and you will fill the gap later, usually at your cost and with an argument.

The starting point is the use case. Research by HSLU and Finup, [reported by investrends in August 2026](https://investrends.ch/aktuell/news/studie-so-nutzen-schweizer-vermogensverwalter-ki/), looked at 20 Swiss independent wealth managers actively using AI. Investment research was the most common use case, followed by KYC and AML due diligence and client communication. Onboarding and custodian-bank data reconciliation were comparatively weak. That is a useful reality check: the tasks with clear inputs and outputs are where firms start, while tasks that depend on messy data feeds are harder to pilot.

A good use case for a pilot is narrow enough to describe in one sentence, has a human who owns the outcome today, and produces output you can compare with the current process. “An assistant that prepares a client-meeting brief from the CRM and the latest portfolio report, reviewed by the relationship manager before use” is a pilot. “AI for client advisory” is a direction.

### What the FINMA guidance adds

FINMA’s [Guidance 08/2024 on governance and risk management when using artificial intelligence](https://www.finma.ch/en/~/media/finma/dokumente/dokumentencenter/myfinma/4dokumentation/finma-aufsichtsmitteilungen/20241218-finma-aufsichtsmitteilung-08-2024.pdf) was published on 18 December 2024. It is guidance, not a statute, but it shows what the supervisor looks at. We anchor pilot terms to four of its themes: how material the application is, including whether it touches retail or institutional clients; tests and ongoing monitoring; documentation; and independent review. A pilot contract that covers these four is easier to take into production, and easier to defend if you are asked about it.

## What should acceptance criteria say?

Acceptance criteria are the part of a pilot contract that decides whether it is fixed in practice. They answer one question: what result lets you say yes, no or not yet, and who decides?

The table below sets out the six elements, what to put in writing for each, and the FINMA theme it connects to.

| Pilot element | What to fix in writing | FINMA hook |
| --- | --- | --- |
| Use case | One task, the users, the clients or client segments affected, what the output is used for, and what the pilot explicitly excludes | Materiality, including whether retail or institutional clients are affected |
| Data | Which systems and documents are in scope, who approves access, where data is hosted and processed, retention and deletion at the end | Materiality and documentation of data used |
| Tests | A representative test set agreed by both sides, the baseline from today’s process, the measures used, the pass threshold, and how errors are classified | Tests and monitoring |
| Documentation | Deliverables: architecture, data flows, model and prompt description, known limitations, audit trail design, human approval steps | Documentation |
| Independent review | Who reviews the results (a function that did not build it), what access they get, and what happens with their findings | Independent review |
| Exit decision | Who decides to stop, extend or move to production, on which evidence, and what you receive if you stop (code, documentation, data returned or deleted) | Governance and accountability for the decision |

A few practical notes on the criteria themselves.

**Write thresholds, not adjectives.** “Accurate” is not a criterion. A named test set, a defined measure and a number agreed before the pilot starts is. If the vendor will not agree a threshold, the pilot cannot fail, which means it cannot prove anything.

**Define what a wrong answer is.** In wealth management, a summary that omits a restriction in a client mandate is a different error from a clumsy sentence. Agree categories and which ones fail the pilot regardless of average performance.

**Keep a human decision in the loop.** State where a person reviews and approves output, and record that in the audit trail. This also gives the independent reviewer something concrete to check.

**Use real, representative material.** A demo on clean sample files is not evidence of deployability. Build the test set from the documents and data your teams handle, with confidentiality terms agreed first.

## How do you compare vendors on a pilot?

Once each candidate has agreed to the same six elements, compare them on the same questions. Using one test pack and one scorecard for every vendor keeps the comparison honest.

1. **Scope clarity.** Can the vendor restate your use case, exclusions and acceptance criteria in its own words, without selling a broader project?
2. **Data and hosting.** Where is data processed and stored, which subprocessors are involved, and what protects client-identifying information? For Swiss firms, ask how the setup supports Swiss data hosting expectations and the revised Federal Act on Data Protection (nFADP).
3. **Human oversight and audit trail.** Can you see who approved what, and when? Can the log be exported for your reviewer?
4. **Integration.** Which systems does the pilot connect to, and which does it mock? A pilot connected to nothing says little about production.
5. **Operating ownership.** Who runs the system after the pilot: your team, the vendor, or a third party? What skills does that need?
6. **Total cost and exit.** What does production cost compared with the pilot, and can you take the documentation, prompts and code with you if you stop?
7. **Regulatory documentation.** Will the vendor supply the material your compliance and risk functions need, including for the EU AI Act if you serve EU clients?

If you want a structured view of how pilots turn into production costs at a bank, we have written separately about AI agent pilot-to-production cost in a bank. It is the question to ask before you sign the pilot, not after.

### Where the EU AI Act fits

Swiss wealth managers that serve clients in the EU, or that use AI inside EU-regulated processes, should know the timing of the EU AI Act. Regulation (EU) 2026/1744 (the AI Omnibus) has been in force since 27 July 2026. Under the [Commission’s AI Act page](https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai), stand-alone high-risk (Annex III) obligations apply from 2 December 2027, and high-risk AI embedded in products (Annex I) applies from 2 August 2028. The general application date of 2 August 2026 is not the date for high-risk obligations. Whether your pilot touches Annex III depends on the use case, so check Annex III against the use case with your legal team rather than assume.

The practical point for a pilot: ask the vendor to produce documentation that would serve both FINMA’s expectations and the EU AI Act, so you do not rewrite it later.

## Fixed pilot, custom proof of concept or broader implementation?

These three are often confused, and the difference is mostly about who carries the uncertainty.

**A fixed-scope pilot** carries the uncertainty on the vendor side for a narrow, well-defined task. It suits use cases where inputs, outputs and tests are already clear, such as summarizing documents or preparing meeting briefs.

**A custom proof of concept** suits a question you cannot yet define, such as whether your custodian data is clean enough for automated reconciliation. Scope expands as you learn, so the commercial form is usually time-based. That is reasonable. It is not a failure to be fixed.

**A broader implementation** is a delivery project: integration, security, operating model, training and monitoring. It is not what a pilot proves, even a successful one.

The mistake to avoid is treating a successful proof of concept as evidence that production will work. It shows the idea is feasible on selected material. It says nothing about integration effort, run costs or how staff will use it.

## Perspective: what we think a good pilot looks like

We are a Zurich-based software development company working with banks, wealth managers and fintechs, and we build AI agents under the expectations of the EU AI Act, FINMA, DORA and GDPR/nFADP. Our view is that the label “fixed-price” is the least interesting part of a pilot. What matters is whether the pilot was designed to be decided on. That means a narrow use case, an agreed test pack, a human approval step, documentation a reviewer can read and an exit decision that someone owns.

We also think pilots should be designed backwards from production. If the data feed, the access rights or the audit trail cannot exist in production, the pilot proves less than it appears to. A wealth manager is better served by a smaller pilot with real constraints than a polished demo.

We do not publish fixed-price terms for pilots on this page and we do not claim to sell them. Commercial terms depend on the use case and the agreed scope, and we discuss them once the scope is on paper.

## How to run the selection in four steps

1. **Pick one use case.** Use your own priorities. Research, KYC and AML due diligence and client communication are where Swiss independent wealth managers [most often start](https://investrends.ch/aktuell/news/studie-so-nutzen-schweizer-vermogensverwalter-ki/), according to the HSLU/Finup study.
2. **Assemble the test pack.** Representative documents, a baseline from today’s process and a list of failure categories. Remove or protect client-identifying data under agreed terms.
3. **Send the same brief to three or four vendors of different types.** Ask each to return scope, acceptance criteria, data handling, documentation list and exit terms in writing. Only then ask about price.
4. **Score on the same scorecard.** Use the seven questions above. Involve compliance and risk from the start, and name the independent reviewer before work begins.

A vendor who resists writing things down at this stage will resist later, when it matters more.

## Where S-PRO fits

S-PRO is a Certified partner in the Claude Partner Network, and our team includes Claude-certified engineers and architects. For wealth and asset managers, we work on AI agents for research, document processing, onboarding and reporting, and on taking pilots to production with audit trails and human-in-the-loop controls. You can read more on our page about AI for wealth and asset managers.

## FAQ

### Does a fixed-price AI pilot exist as a standard product?

Some product vendors offer a packaged pilot on a standard configuration. For custom agents built around your systems, fixed terms depend on how clearly the scope, data and acceptance criteria are written. If a vendor offers a fixed price before those exist, treat the price as unreliable.

### What should a wealth manager ask for in writing before a pilot starts?

Ask for the use case and exclusions, the data in scope and its hosting, the test set and pass thresholds, the documentation deliverables, the independent reviewer and the exit decision rule. These six map directly onto the table above and onto the themes of FINMA Guidance 08/2024.

### Is a successful pilot enough to go live?

No. A pilot shows the approach works on selected material. Production adds integration, security review, monitoring, operating ownership and run costs. Ask the vendor up front what changes between the two, and have your compliance and risk functions confirm.

### Do the EU AI Act dates matter for a Swiss wealth manager?

They can, if you serve EU clients or your AI use falls within the Act’s scope. Stand-alone high-risk (Annex III) obligations apply from 2 December 2027, and high-risk AI embedded in products (Annex I) from 2 August 2028. Check Annex III against the use case with your legal team.

## How S-PRO helps

S-PRO is a software development company headquartered in Zurich, at Löwenstrasse 45, 8001 Zürich, with R&D centers in Zurich and Łódź. We were founded in 2014, have 250+ engineers and are ISO 27001 and ISO 27701 certified. We help banks, wealth managers and fintechs define AI pilots with written scope and acceptance criteria, then take them to production with audit trails, human-in-the-loop controls and documentation for the EU AI Act, FINMA, DORA and GDPR/nFADP. To discuss a use case, start with our overview of AI pilot to production for banks.

## Sources

- [Studie: So nutzen Schweizer Vermögensverwalter KI, investrends](https://investrends.ch/aktuell/news/studie-so-nutzen-schweizer-vermogensverwalter-ki/)
- [FINMA Guidance 08/2024 on governance and risk management when using AI](https://www.finma.ch/en/~/media/finma/dokumente/dokumentencenter/myfinma/4dokumentation/finma-aufsichtsmitteilungen/20241218-finma-aufsichtsmitteilung-08-2024.pdf)
- [European Commission, AI Act regulatory framework](https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai)

Subscribe

Get technical insights, and industry news

Thank you!

Get ready! You will receive handpicked content right to your inbox.

---

Link copied to clipboard

[Get in touch](https://s-pro.io/get-in-touch)