AI in banking that gets every client file ready to review.
Onboarding files wait weeks for one missing proof of address, and compliance rereads every page by hand. We build agents that check and summarize the documents inside your own systems, so your people spend their time on judgment. Accepting a client or granting credit stays with them.
- KYC file preparation
- Advisor meeting notes
- Procedure answers with sources
AI in banking: the work it can take on today
AI in banking works best on the paperwork around each decision: reading identity and income documents, checking a file against your onboarding checklist, summarizing an advisor’s meeting, answering staff questions from your procedures with the source. It prepares and explains. Opening an account, lending or escalating a client stays with a named person.
Wealth management follows the same pattern. Before a review meeting, the agent gathers the client’s last statements, notes and stated objectives on one page. After it, a draft suitability statement waits for the advisor to correct and sign, and nothing reaches the client without that signature.
Where the data sits is settled first: on your servers or on dedicated hosting in Europe, under a written contract that names the locations, as DORA requires of a financial entity’s contracts with its ICT providers. Every document read and every answer given is logged for your audit trail.
Heard in the onboarding queue
- A file sits for days because one proof of address is out of date.
- Compliance reads the whole file again because nobody knows what was already checked.
- Advisors type their meeting notes in the evening, from memory.
- The procedure manual is so long that staff call compliance instead of opening it.
- Every regulator questionnaire starts with a hunt through shared drives.
Four files an agent can prepare for your teams
Each one runs on your documents and your core systems. Each one ends with a person who checks and decides.
01
Onboarding and KYC files
Identity documents, proof of address and company registry extracts are read, checked against your checklist and expiry rules, and the missing piece is requested from the client right away.
Document AI02
Advisor meetings, before and after
Before: one page on the client’s holdings, last contacts and stated goals. After: a structured note and a draft suitability statement for the advisor to correct.
AI agents03
Answers from your procedures
Staff ask how to handle a power of attorney or a client living abroad, and get the answer from your own procedures, with the paragraph it comes from.
Knowledge assistant04
Alert reviews on your own servers
For each transaction monitoring alert, a model running in your infrastructure gathers the account history and the documents on file into a draft for the analyst, who closes or escalates.
Private LLM
Who does what in a bank that uses AI
The split we apply on every banking project. The right-hand column names the text behind each line, all cited further down this page.
| Task | What the agent does | Who decides | Rule that applies |
|---|---|---|---|
| Customer due diligence | Reads the documents, checks them against your checklist, lists what is missing | A compliance officer | EU AMLR, art. 76(5): meaningful human intervention on accepting or refusing a client |
| Credit application | Gathers income documents and account history into the file | The credit analyst | EU AI Act, Annex III 5(b): credit scoring is high-risk |
| Transaction monitoring alert | Assembles the history and past alerts into a draft review | The AML analyst | Your internal AML procedures |
| Investment advice | Drafts the suitability statement from the meeting notes and client profile | The advisor, who corrects and signs | MiFID II, art. 25(6): a statement on suitability for the retail client |
| Staff questions | Answers from your manuals, with the source paragraph | The employee, who checks the source | EU AI Act, art. 50: people know they are talking to an AI |
| Hosting and contract | Runs on your servers or dedicated hosting in Europe | Your CIO and your DPO | DORA, art. 30: the contract names where data is processed |
Illustrative scenario
Inès opens Monday’s onboarding queue with every file already checked
A private bank, a small compliance team, and new client files that arrive incomplete. This scenario shows where the agent sits in Inès’s week. It describes no real client, and the live demo next to it runs on a fictional company’s procedures.
See the mechanism on the live knowledge assistant: it runs on a fictional distributor’s procedures and shows the passage behind every answer. On your project, your own procedures take their place.
- Over the weekend, the agent reads the documents sent with each new file.
- Each file gets a checklist: present, expired, inconsistent or missing.
- Clients with a missing document receive a request in the bank’s own wording.
- Inès starts with the complete files and decides on each one herself.
- Every check, request and decision stays in the bank’s audit log.
The texts we design for from the first workshop
- A person decides on the client.From July 10, 2027, the EU anti-money laundering regulation (2024/1624, art. 76(5)) allows decisions from automated processes or AI systems only if accepting, refusing or keeping a client gets meaningful human intervention, and the client can obtain an explanation and challenge it. We build to that standard today.
- Credit scoring is high-risk.The EU AI Act lists systems that evaluate the creditworthiness of individuals or set their credit score as high-risk (Annex III, point 5(b)), fraud detection excepted. We keep scoring out of scope unless its classification and duties are settled with you first. Dates are in our AI Act guide.
- DORA shapes the contract with us.DORA (Regulation (EU) 2022/2554) applies to banks, investment firms and payment institutions since January 17, 2025. Article 30 asks for a written contract that names where data is processed and stored, with advance notice before any change. Ask every AI provider, us included, for that contract before work starts.
- Advice comes with a suitability statement.When a firm gives investment advice, MiFID II (Directive 2014/65/EU, art. 25(6)) requires a statement on suitability in a durable medium, saying how the advice meets the retail client’s preferences, objectives and other characteristics. The agent drafts it from the meeting notes; the advisor corrects and signs it.
Financial services: your questions
Does the AI decide whether we accept a client?
No. It reads, checks and lists what is missing; a compliance officer accepts or refuses. The EU anti-money laundering regulation that applies from July 2027 asks for exactly that: meaningful human intervention on those decisions, and an explanation the client can obtain.
Where do client documents go?
To the hosting you choose: your own servers, or dedicated servers in Europe under a contract that names each location. The model provider receives nothing for training, and every document read is logged where your audit team can see it.
Can it work with our core banking system and CRM?
Yes, through the APIs or exports your systems already offer, read-only to start. When a system has no API, we work from its exports or build a connector. Writing back into the core system comes later, step by step, once your team trusts the output.
Is it useful for wealth managers and financial advisors?
Yes. Meeting preparation, notes, client letters and the draft suitability statement are the paperwork advisors finish in the evening. The agent prepares them from your notes and the client profile; the advisor corrects, signs and keeps the relationship.
Are you an ICT provider under DORA?
If you are a financial entity under DORA, yes. Your contract with us then has to carry the clauses of article 30, and the service goes into your register of information. The code and the accounts stay in your name, which keeps an exit plan simple.
How fast can a first agent run?
A first agent usually runs in four to eight weeks, often on one file type such as onboarding. Scope and price are fixed in writing before we start, and the free 30-minute assessment tells you where to begin.
Keep exploring
- All industries What AI does in each sector we serve
- AI use cases Every use case, by function and industry
- Live demo: knowledge assistant Ask it a question, check its source
- Private LLM Models hosted in Europe or on your servers
- EU AI Act compliance Risk levels, deployer duties and dates in force
- AI in insurance Claims and policy documents processed
- AI for accountants Client documents collected and pre-booked
Open Monday’s queue with every file checked.
Tell us which file slows your team down the most. You get an answer within one business day, then a free 30-minute call on your own documents and a straight view of where an agent helps and where it should stay out.