At a bank, payment institution or fintech, a great deal of time goes into compliance work: identifying every new customer, checking transactions for money laundering and fraud, and keeping customer files up to date. Much of that time is often spent on false alarms. In this sector, AI is therefore not primarily used for marketing copy, but to make that control work faster and more precise. This guide shows which tools are in our index, what to look for and what they cost. The full overview is on AI for banks and fintech.
Where AI helps in this field
Onboarding: identification and verification
During onboarding you check the identity document, run a liveness check and screen the customer. Amsterdam-based Fourthline combines document and biometric checks with AML screening and fraud detection, and can also remediate existing customer files. Germany's IDnow checks more than 3,800 document types and focuses on detecting deepfakes and synthetic identities. Veriff from Estonia handles document checks, passive liveness and Know Your Business. Norway's Signicat is an eID hub that lets customers log in with, among others, iDIN, and has electronic signing built in. If you only need to process documents, Klippa can extract data from identity documents and detect document fraud. All tools that process documents are listed on AI for documents at banks.
Transaction monitoring and fewer false alerts
This is where machine learning delivers the most. Discai, developed within the KBC Group, combines adaptive ML models with a rules and simulation engine. Hawk and Napier AI offer transaction monitoring, sanctions and payment screening, and continuous customer risk scoring. Both say they make their alerts explainable. Napier has more than 100 built-in AML typologies and runs in the cloud or on-premise.
Fraud and scams in real time
Feedzai focuses on real-time fraud detection across all channels, scam prevention (APP fraud) and account takeover. It uses behavioural and device intelligence for this. Hawk also detects money mules and scams.
Alert investigation and case files through AI agents
The newest step is AI agents that prepare an alert before the analyst looks at it. The Luci agent from Lucinity gathers the evidence, analyses the case and writes the case report. Hawk has an AML Investigative Agent for this. Fenergo deploys KYRA agents for KYC tasks within a client lifecycle management platform.
What to look out for
Sensitive data and outsourcing. You process identity documents, biometrics and transaction data. Choose a provider with EU data location, ISO 27001 and a data processing agreement, and ask whether your client data is used to train the provider's models. That last point really matters: according to the privacy statement of Veriff, data is also used to train its own algorithms. IDnow runs on its own servers in Germany, Signicat has data centres in Norway and the EU, and with Fenergo you choose the region yourself, including the EU. For Discai, Feedzai and Napier the data location is not stated on the site, so ask for it. If you outsource critical processes, DORA and EBA rules apply. Put exit and audit rights into the contract.
Explainability and human decision-making. For transaction monitoring, the regulator expects you to explain why an alert does or does not trigger, to have a person decide on a report to the FIU, and to test the model for unequal treatment of groups of clients. An AI agent may prepare a case, but it may not decide itself.
EU AI Act and new AML rules. Creditworthiness assessment of natural persons is a high-risk application under the EU AI Act. From 10 July 2027 the European AMLR largely replaces the Dutch Wwft. So check whether your provider is prepared for this. We track regulators' decisions on the page AI, GDPR and supervision.
General chatbots. Do not paste client or transaction data into a free chatbot. The business versions of ChatGPT and Claude do not train on your input by default. Microsoft 365 Copilot falls under the EU Data Boundary, but the Claude models within Copilot fall outside it. Mistral stores data in the EU by default. DeepSeek stores data in China and is therefore unsuitable.
Three starting scenarios
1. Fintech or payment institution that wants to onboard quickly.
Compare Fourthline, IDnow and Veriff with a test set of real (anonymised) cases. Look at the percentage of clients who drop out, the processing time and how often a fraud attempt is detected. If you serve Dutch clients using iDIN, include Signicat as well. Signicat is the only one of these four for which a free trial is listed in our data.
2. Bank with too many false alerts.
Run Discai, Hawk or Napier AI on historical transactions whose outcome you already know, alongside your current rules. Compare the number of false alerts and missed cases, and for each alert ask for an explanation you can show to the regulator.
3. Compliance team that gets stuck on documentation work.
Test Lucinity or the KYRA agents of Fenergo on closed cases. Have analysts place the AI's file next to their own. Luci is built on Microsoft Azure OpenAI, so ask about the region and the data retention arrangements.
Common mistakes
- Looking only at detection scores. A model that finds more but cannot be explained becomes a problem for supervision and audit.
- Not excluding training use. Agree contractually whether client data may be used to improve the provider's models.
- Forgetting exit arrangements. Under DORA you must be able to replace a critical provider. Make sure you can get your data and logs out.
- Letting the AI agent decide. A report to the FIU or refusing a client remains a human decision.
- Not testing for unequal treatment. Periodically check whether certain groups of clients are affected more often without a valid reason.
Cost overview
| Tool | Price |
|---|---|
| Klippa | €25 free credit; thereafter tiered pricing + custom |
| Veriff | On request (self-serve and enterprise) |
| Signicat | On request; free trial |
| Fourthline, IDnow | On request |
| Hawk, Napier AI, Discai, Feedzai | On request (B2B) |
| Fenergo, Lucinity | On request (B2B) |
| General assistant (business) | ChatGPT Business from $20 per seat, Claude Team from $20 per seat, Copilot Business $21 per seat, Mistral Team from $24.99 per seat |
Almost all professional platforms in this sector work with bespoke pricing, which usually depends on the number of verifications or transactions. So request a price based on your own volumes, and ask about implementation and integration costs too.
See the full overview of AI tools for banks and fintech, or the related sectors insurers, financial advice and debt collection. You'll find explanations of terms such as AML, KYC and data processing agreement in explanations and terms.
