The following is an extract from our latest report – Machine Learning: Post-trade’s white knight? The report explores the transformative potential of machine learning for post-trade data processing. However, we believe that the following best practices apply to machine learning regardless of how and where it’s utilised.
Machine learning can have a huge impact on post-trade processing. It’s important that this technology is harnessed correctly to unlock the full suite of benefits it can deliver.
At Duco, we believe any machine learning system needs to have the following features.
Machine learning needs to be embedded at the heart of the user experience. It’s the only way to ensure maximum performance and functionality for users. Software that treats machine learning as an add-on lacks the integration to make full use of the data it is processing. This results in less accurate models and fewer opportunities to automate tasks.
In the cloud
You want the most accurate predictions from your machine learning platform. A machine learning engine living on premise or in your data centres only has access to your data. One that lives in the cloud can train on much larger datasets. For example, Duco Alpha is able to train on non-sensitive data from all customers without compromising on segregation.
This means that predictions made by Alpha are based on the previous actions taken by thousands of users across the globe. Machine learning can offer accurate predictions and suggestions ‘out of the box’ if it has access to data on the cloud like this. An algorithm that can only train on siloed data is much more limited and will take time to train. Even when it is functional it will still be less accurate than a system trained on a much larger data set.
Human in the loop
Algorithms are great at spotting patterns and making predictions, but they can’t reason in the way humans can. We believe the role of machine learning is to support human operators by offering suggestions.
Machine learning can take away the complexity and repetitive nature of many tasks in post-trade processing, but it’s important that a human makes the final decision. This improves auditability and prevents errors propagating downstream without anyone noticing.
Continuous improvement is essential
Training a machine learning algorithm doesn’t just happen once. The models constantly need to be refined and improved as more and more data becomes available. This keeps them accurate, addresses biases in the models and ensures your processes are agile in the face of change.
We describe this as the Flywheel Effect, and it’s one of the fundamental principles we used when designing the Duco Alpha machine learning engine. Systems built on this principle are able to constantly improve and deliver a better customer experience the more they are used.
Discover how machine learning will empower firms to conquer the challenges of post-trade data. Download the report today.