Skip to content

The way we work

At the request of banks, TMNL brings transaction data from business customers from different banks together and makes meaningful connections between this data. TMNL is creating smart models to detect these potentially unusual transactions. These models are used effectively and responsibly, while excluding risks — such as the risk of discrimination — in the process.

The links that these models make provide new insights into potential money laundering and the financing of terrorism. Small multidisciplinary teams, consisting of AML experts, data scientists, data engineers, and machine learning engineers, are creating these models and working together using a fixed approach.


  • AML experts from TMNL, banks and public parties discuss money laundering risks and patterns that TMNL could analyse.
  • TMNL coordinates its proposal for possible models with national priorities in the fight against money laundering, addressing money laundering in a way that banks in isolation cannot.
  • Based on inputs from the wider stakeholder field, TMNL proposes developing a specific model (e.g. on underground banking) as part of its model roadmap.
  • On approval by the banks, the actual building starts.

Build & run

  • AML experts define the new model’s objectives and principles in accordance with the wishes of the banks.
  • With data scientists, data engineers and machine learning engineers they design the new model, which is then tested in multiple iterations.
  • The data scientists and machine learning engineers write the technical code behind the model and make sure it interacts as expected with the selected TMNL data.
  • Samples are tested to validate and finetune the model, technical tests validate the code’s correctness.
  • Risks around data (quality), privacy, unintentional bias and ethical considerations are discussed internally (also with the TMNL Risk team), weighed, mitigated and documented.
  • On approval by the TMNL Model Board and User Council (in which the banks are represented), a first test batch of alerts is shared with the banks for review.

Follow up

  • Specialised alert reviewers in the banks receive specific training and guidance on the new model.
  • Reviewers can access the alert and guidance in the TMNL alert review portal, where they can also document their feedback. Access to client and transaction information is obtained through the bank’s system.
  • Data specialists at the bank link the alerts to the bank’s own internal datasets to provide the reviewers with the required information.
  • Reviewers investigate the alert in the context of the client profile, and contact the client for further information if needed.
  • Feedback on alert level is sent to TMNL, without disclosing information on the client or the final judgment on the alert.


  • TMNL evaluates the feedback that the alert reviewers provide through the portal.
  • The model’s performance (effectiveness) is analysed quantitatively and qualitatively.
  • TMNL investigates how the model can be improved and made more effective.
  • TMNL investigates how the model can be improved and made more effective.
  • Following feedback and approval by the TMNL Model Board and User Council, TMNL can either develop an enhanced version of the model or abandon the model.
Build and test Follow Evaluate