Automating KYC as a wealth manager: dossiers built with AI

Automating KYC does not mean automating the review. AI reads the documents, extracts the facts, flags contradictions and gaps, and drafts the source-of-wealth narrative, with every statement backed by a source. The risk rating, the PEP assessment and the decision about the client stay with humans. That division is exactly the point: compliance decides faster and better because it reviews a complete, referenced dossier instead of assembling one itself.

Example situation (fictional, but close to everyday reality): an entrepreneur with three companies in two cantons wants to move to your firm. On the table sit commercial register extracts, the sale agreement of a former shareholding, tax returns, bank statements and notes from two initial meetings, several dozen documents in all. Turning that into a reviewable dossier with a clean account of where the wealth comes from typically takes hours spread over several days. Most of that time is not review but search: who owned what when, where each amount originated, which document proves which statement.

How does the KYC review work today?

The flow is similar in most firms, regardless of size:

1. Identification: verify the ID; for companies, establish the beneficial owners.
2. Collecting documents: commercial register, proof of assets, tax documents, contracts.
3. Clarifying circumstances: occupation, origin of funds, purpose of the business relationship.
4. Writing the narrative: the evidence becomes the story of the wealth, the source-of-wealth dossier.
5. Assessing risk: set the risk profile, check special exposure such as politically exposed persons (PEP).
6. Deciding and monitoring: decide on acceptance, define criteria for ongoing monitoring.

The framework is set by the Swiss Anti-Money-Laundering Act (AMLA) the FINMA Anti-Money Laundering Ordinance (AMLO-FINMA) and the rulebook of your supervisory organisation. How the duties apply in a specific case belongs with your compliance specialist. This article deliberately looks only at the manual work: where time is lost, and which parts can be prepared without shifting responsibility.

The industry evidence confirms the pain: in the study "AI Use Cases and Crypto Assets in External Asset Management" by the Lucerne University of Applied Sciences and Finup from July 2026, onboarding, KYC and research rank among the biggest pain points of independent wealth managers, while AI-supported KYC and AML checks are at the same time among the most frequently reported use cases (HSLU/Finup, July 2026). The effort is well known, and the first firms are tackling it.

What does the AI take over in dossier preparation?

Five steps can be prepared reliably today:

1. Reading documents. The AI reads commercial register extracts, contracts, tax documents and bank statements, and recognises what each document is and what it says.
2. Extracting facts. Names, dates, amounts, ownership structures and addresses end up in a structured overview instead of twenty open PDFs.
3. Flagging contradictions and gaps. If a date of birth differs between two documents, a year is missing from the wealth history, or an amount does not match the stated sale proceeds, the spot is flagged. That is the most tedious search work in the entire process.
4. Drafting the narrative. From the facts comes a draft of the source-of-wealth text: where the wealth comes from, in what sequence, evidenced by which documents.
5. Backing every statement with a source. Every figure and every claim in the draft points to the document and the location it comes from. Review becomes spot-checking instead of a scavenger hunt.

The result is a dossier draft, not a decision. It reaches compliance in the state a dossier has today after hours of manual work, only within minutes.

Where does the human still decide?

The line is easy to draw: everything that is a judgement about the client stays with the human. Everything that is gathering and ordering facts can be prepared by AI. As a process with marked human checkpoints it looks like this:

Step	What the AI prepares	Human checkpoint
Collecting documents	Completeness list, reminders to the client	Confirm completeness
Reading documents	Structured facts with source location	Spot-check against the source
Finding contradictions	Flagged deviations and gaps	Judge whether a deviation matters
Writing the narrative	Draft with a source reference per statement	Review the text against the evidence and approve
Risk rating	(nothing)	Always with the human
PEP assessment	(nothing)	Always with the human
Acceptance decision	(nothing)	Always with the human

The prepared dossier process: AI gathers and orders, the human judges and decides.

This separation is not a brake but the condition for the result to hold up in an audit. Anyone who hands the rating or the decision to the machine gives up exactly the responsibility they must answer for towards their supervisory organisation.

How do large institutions set this up?

From collaborating with additiv I know how large institutions industrialise this preparation: in the self-service onboarding of a European neobank market leader for discretionary portfolio management, data capture, completeness checks and document preparation run in the background. The human reviews and decides at the end of a prepared process instead of carrying the process themselves.

These platforms are multi-year projects with dedicated teams. The logic behind them, however, can be adopted in weeks: capture documents cleanly once, structure the facts, make the decision easy for the human with a referenced dossier. The tools for this are ready today, without building a platform.

From my practice: In first conversations, KYC is almost always the first pain mentioned, and still I advise against starting there. KYC is the case with the highest data sensitivity and the most demanding data foundation. Firms that first automate protocols or research build within weeks the review-and-approve routine a KYC dossier later requires. Firms that start with KYC wait months for a first result and lose the team along the way.

What does a small firm need in place?

Three things must be in place before the first dossier runs through AI:

1. Clarified hosting. No project gets past one question: where does my data live? The good answer is simple today: a protected Swiss cloud, hosted for example on Microsoft Azure in the Zurich North region, cleanly separated from public models. What is allowed with client data in public chatbots and what is not is covered in the article on ChatGPT in the wealth manager's daily work.
2. Logging. Every AI action needs a timestamp and a rationale, and every client-relevant output a human approval. That audit trail is exactly what examiners will want to see later.
3. Documentation. Purpose, data, assumptions and limits of the application belong on one page. In its Guidance 08/2024, FINMA sets out what it looks at when AI is used: inventory, responsibility, data quality, testing, documentation, explainability and independent review (FINMA, December 2024). What that means concretely for a small firm is in the article on FINMA requirements for AI in financial SMEs.

If one of the three is missing, start there, not with the dossier. The order and the other six use cases are in the pillar article on AI for independent wealth managers.

Note: We are not a law firm. All statements on legal and regulatory topics, in particular on AMLA, AMLO-FINMA and the rulebooks of the supervisory organisations, are subject to change and do not replace legal advice. For specific questions, please consult a lawyer or your compliance specialist.

Conclusion: your next step

Take the next new-client dossier and track once how many hours flow into search and ordering and how many into actual judgement. In most firms, the search share is well above half, and exactly that part can be prepared. What the full onboarding process looks like and where AI saves time there is described in the article client onboarding at wealth managers. If you would like to see an environment where hosting, logging and approval are built in from the start: on AI for wealth managers we show what that looks like.

FAQ

Does AI replace the compliance specialist?

No. AI prepares the dossier: reading documents, ordering facts, flagging contradictions, drafting the narrative. Risk rating, PEP assessment and the decision about the client always stay with the human. Responsibility cannot be delegated, and that is exactly why preparation is the right place for AI.

Is an AI-prepared dossier compliant with AMLA?

AMLA, AMLO-FINMA and the rulebooks of the supervisory organisations require due diligence, not a specific technology. What matters is that a human reviews and decides, and that the path there is documented. How to classify your specific setup is a question for your compliance specialist or a lawyer.

From what size does dossier preparation pay off?

Less a question of size than of volume: if dossiers whose assembly takes hours come up regularly, preparation pays off. Firms with only a few new clients per year are better off starting with a simpler case such as protocols or research and taking KYC as the second or third step.

What does getting started cost?

That depends on the scope: a pilot including setup sits in the five-figure franc range, and the running costs are comparatively small. Concrete orders of magnitude are in the article What does AI cost for financial services firms?.

What data may the AI see in the process?

Only data in an environment with a clear contract, a verified location and logging, typically a protected Swiss cloud. Client documents do not belong in public chatbots. The rules are in the article on ChatGPT and client data.