Dafin and WIZE: AI for Wealth Managers That Never Sees Client Data

Dafin and WIZE have entered a strategic partnership. AskWIZE, the AI assistant inside the WIZE wealth and asset management platform, is powered by Dafin.

For most firms, the AI conversation stops at one question from compliance: where does our client data go? With AskWIZE the answer is: nowhere. The language model receives the question. It never receives the data.

Two Swiss teams, one shared conviction

WIZE is the all-in-one wealth and asset management platform behind more than 130 clients across 28 countries, with over 3,000 users and CHF 120 billion in assets represented on the platform.

Dafin builds enterprise AI for financial institutions. It was founded by practitioners with more than twenty years inside wealth management, banking and financial software, and it is designed to run inside a regulated firm rather than around it.

The two teams share one conviction: in this industry, AI only creates value if your compliance team can sign off on it.

The pilot stalled. Shadow AI did not.

Most firms have run an AI pilot by now, and many of those pilots are still waiting on a security review. In the meantime the questions do not stop, so the answers get found elsewhere.

• A relationship manager pastes a client holding list into a public chatbot to summarise it.

• The question, the client name and the positions leave the perimeter in a single copy-paste.

• None of it appears in your logs, your audit trail or your retention policy.

That is the real cost of waiting. The firm takes on the risk of AI adoption without getting any of the value.

How AskWIZE works: the model sees the question, never the data

AskWIZE rests on a strict separation between reasoning and data.

• The language model only receives the question, never the client data.

• The answer is computed inside your perimeter, directly on your WIZE data.

• Every figure comes from WIZE and can be traced back to its source.

• If the data is not there, the assistant says so instead of guessing.

In practice, the model reads the question and returns a method: which systems to query, which calculation to run, and in which order. The Dafin engine executes that plan inside the perimeter, against your WIZE data. The numbers are computed, not generated, which is why the same question always returns the same answer.

What that looks like at the desk

Ask: summarise the Mueller portfolio YTD versus MSCI ACWI and flag the top and bottom contributors.

The assistant returns the portfolio at +6.4% against +5.1% for the index, an excess return of +1.3 points, with Swiss equities the highest contributor at +0.9 points and emerging-market bonds the lowest at -0.4 points. Above the answer sits the line that matters to a reviewer: calculated from two sources, three steps, reasoning trace logged.

And where the data stops, so does the assistant. A private-markets valuation that is not in the source data is flagged as missing rather than estimated. Figures here are illustrative.

One assistant, every desk

The same assistant serves the whole firm, because every desk is asking a version of the same question about the same data.

• Relationship managers: which of my clients are most exposed to this move?

• Risk and compliance: which clients are due for MiFID suitability review?

• CIO: which mandates are misaligned with the growth model?

• Trading: what is the desk net JPY exposure against limits?

Why traceability beats fluency

A figure nobody can trace is a figure nobody can use. Ask a generic AI tool the same portfolio question twice and you may get +7.2%, then +9.4%. Neither number can be defended in a client meeting, let alone an audit.

Because AskWIZE calculates rather than generates, every answer carries its sources and its reasoning trace. That is the difference between an interesting demo and something a risk committee will approve.

Built for Swiss standards, useful anywhere

Both companies are Swiss, and the confidentiality standards of Swiss wealth management shaped the architecture: your data stays inside your own perimeter, no client data is sent to model providers, and every answer can be reconstructed months later. Firms working under MiFID II, FINMA or GDPR obligations tend to arrive at the same requirement list, which is why the design travels well beyond Switzerland.

What it means if you already run WIZE

Nothing gets replaced. The assistant works on the WIZE data you already maintain, under your existing permissions, so a user only ever sees what they were already entitled to see. There is no data migration, no eighteen-month replacement project, and no new copy of your client data sitting in a third-party system.

If you run other systems alongside WIZE, Dafin connects to those as well: the platform side adds modules only where you have a gap. The security architecture is documented on our security page.

See it on your own data

The fastest way to judge any of this is to ask the assistant a question you already know the answer to. Book a 20-minute demo with our founding team, or start with the product overview.

Two Swiss teams. One shared conviction.