Check it Out
awards
WiWo Award TRUSTEQ
BrandEins Berater 2026 Award
TRUSTEQ kununu awards
TRUSTEQ
CooperativeExcellence

Traditional technologies with new glow

Machine Learning: A New Look

At first glance, traditional machine learning (ML) technologies often seem unspectacular compared to generative artificial intelligence (AI) or large language models (LLMs). Yet it is precisely these technologies that perform crucial tasks behind the scenes in many companies: they identify risks, forecast trends and deliver highly accurate, data-driven predictions. And they have been doing so for a long time: as early as the 1980s, financial firms began using early forms of ML technologies to analyse big data. Even though the focus is currently very much on generative AI, proven ML technologies continue to deliver measurable added value.

It is therefore worth paying particular attention to these established methods. ML technologies refer to proven machine learning methods that systematically analyse data, identify patterns, make predictions and support data-driven decision-making. Typical examples include random forests, gradient boosting, support vector machines and deep neural networks, which have been used successfully for years to analyse structured data. 


Particularly in data-intensive sectors such as the financial sector, the application of ML technologies is already generating concrete, measurable added value in this way. Furthermore, the targeted combination of traditional ML technologies with modern AI developments opens up additional potential for data-driven innovation.

New approaches to traditional machine learning technologies in the financial sector

Traditional ML technologies have long been established in the financial sector and form the basis of many data-driven processes. New technological approaches are now expanding the scope of application for traditional models, particularly when dealing with sensitive data or in highly regulated environments. Four examples:

1. Federated Learning

Using data without sharing

The handling of sensitive data is particularly relevant for highly regulated sectors such as insurance and banking. Traditional ML technologies quickly reach their limits in this context, as data is often not permitted to be consolidated centrally.

The handling of sensitive data is particularly relevant for highly regulated sectors such as insurance and banking. Traditional ML technologies quickly reach their limits in this context, as data is often not permitted to be consolidated centrally. Federated learning offers a new approach to this: models are trained locally, directly where the data is generated. Instead of transferring data, only model parameters are exchanged between the parties involved. In practice, this means: Several insurance companies or financial institutions can jointly develop more powerful models without disclosing sensitive customer data. This results in clear advantages:

Sources

Use of distributed data sources without breaching data protection requirements

Quality

Better model quality thanks to a larger, more diverse dataset

Opportunities

New opportunities for cross-organisational use cases, e.g. in fraud detection

2. Temporal Graph Networks

Understand the dynamics


Many relevant issues in the financial services sector are not static but evolve over time. Traditional models and simple graph-based approaches – that is, models that represent relationships between entities such as accounts, customers or transactions as a network – often fail to adequately account for this dynamism. Who transfers how much to whom, how often and since when? Fraud, money laundering or credit risks often only become apparent when models track transaction patterns over a longer period of time.
This is precisely what ‘temporal graph networks’ do. They extend graph-based models to include a temporal dimension. This allows not only relationships between entities to be analysed, but also their development. 
This enables, for example:

  • Tracking evolving fraud networks
  • Early detection of changes in customer behaviour
  • Analysis of complex event chains across multiple time points

    • These models come closer to real, constantly changing processes and deliver more meaningful results.

3. Decision Intelligence

From knowledge to action

A key step in the development of machine learning (ML) involves shifting the focus from pure predictions to concrete recommendations for action. Decision Intelligence combines ML with optimisation methods and domain-specific decision rules. Optimisation methods calculate the best course of action within given constraints – such as budget or capacity; domain-specific decision rules set regulatory or internal corporate boundaries that no model is permitted to override. The aim is not merely to predict what might happen, but to actively determine which decision is the best under given conditions. 

In practice, this means: Models not only assist with risk assessment but also provide concrete recommendations, for example on optimal pricing, the selection of appropriate measures or the prioritisation of cases.

The advantages:

This transforms ML technologies from mere analytical tools into an active component of the decision-making process.

4. Self-Supervised Learning: Get more out of data


A large proportion of the data available in the insurance and financial sectors is unstructured or unlabelled and is therefore only used to a limited extent in traditional ML projects.
Self-supervised learning (SSL) addresses precisely this problem. Models learn structures and relationships in large datasets independently, without the need for extensive manual annotation. An example: An insurance company has millions of archived claims reports in free-text format – but only a fraction of these have been manually categorised. A self-supervised model can learn independently from the unlabelled texts which phrases indicate fraud, duplicate claims or unusual claim histories – without a human having to flag each case beforehand.  
SSL models enable:

  • The use of large volumes of previously unused data
  • Reduced effort for data preparation and labelling
  • Better generalisation of models to new, unknown cases


This creates significant additional value, particularly in data-rich organisations, as existing data sets can be utilised much more efficiently.

Conclusion

Traditional ML technologies have long been established in the financial sector and deliver reliable, measurable results.
The most exciting developments are emerging at the intersection of tried-and-tested and new approaches: temporal graph networks, federated learning, self-supervised learning and decision intelligence are expanding the possibilities for using data in a more targeted way. 
Those who combine these technologies in a targeted manner and continuously develop them further can not only optimise processes but also generate genuine competitive advantages and harness the full potential of AI.

At TRUSTEQ, we help your business effectively combine traditional ML technologies with modern AI technologies. This enables you to harness the potential of AI in a targeted manner – from intelligent assistance systems and automated analytics to robust, data-driven decision-making models. 
As your trusted advisor, we support you throughout the entire process: from identifying suitable ML use cases, through piloting and implementation, to ensuring ongoing regulatory compliance – including the requirements of the EU AI Act. With this holistic approach, we embed innovative AI solutions in your business in a practical, secure and sustainable manner.

Sophia Hofbauer

AI Developer @TRUSTEQ