Case Study
Simplifying documentation using machine learning
Challenge
Every day, millions around the world interact with dense information they struggle to understand, from adverts on public transport to their personal banking communications to their social media terms and conditions. This information can be really complex, and the general public cannot reasonably be expected to comprehend them.
All of this exists against the backdrop of regulatory requirements, including the requirement for some communications to be “clear, fair and not misleading.” In particular, banks and financial institutions have recently come under fire because consumers were signing up to mysterious terms and conditions without properly understanding what they were signing up to. This has led to growing interest by financial institutions to provide a greater duty of support and care to their customers.
The CEO of Amplifi approached MDRx with an idea of how to address this. Amplifi needed a full machine learning solution to address ongoing consumer risk, including creating a database with long-form legal text, developing a risk engine – including the design of machine learning algorithms, and producing audit trails – aligned with legal agreements and T&Cs.
Action
MDRx devised a sophisticated data science approach to identify problematic wording, terms and sentence structures. We ingested vast quantities of training data, comprising terms and conditions including those contributed by one of the UK’s largest telecomms companies and a world-leading bank.
We built a robust data pipeline, hosted in the client’s AWS and written primarily in Python, and contributed to the design of a responsive, attractive and easy-to-use front end for the application. The core machine learning engines were focused on:
- language assessment that learned contextual relationships between words in the text;
- an intelligibility score provided at document, paragraph, sentence and word level with real-time scoring updates and a heat map highlighting areas of particular complexity;
- a recommendation engine that showed where and how to simplify the text.
This information could then be abstracted into reporting and risk-management tools that could track improvement over time and across multiple document types, with customisable dashboards for good risk management, management information, compliance and regulatory reporting requirements with a clear audit trail.
We then prepared the client for long-term success, helping them to recruit an internal development capability that would hold the IP and preserve the value of the business in the long term.
Impact
- Simplified information: We developed a system that can have a profound impact on the long term health and wellbeing of consumers interacting with complex information every day.
- Powerful machine learning: We harnessed the latest technology, ensuring Amplifi exists right at the cutting edge of machine learning, data science and natural language processing using large language models.
- Innovation de-risked: We engineered the system with the overarching legal, regulatory and policy direction in mind, tapping into expertise from across The MDR Group.
- Long-term success: We set up Amplifi to thrive without us, helping them to design and recruit an in-house technology capability.
- Award-winning: Amplifi has rightly received significant recognition, and has been accepted into some of the UK’s most prestigious and competitive Sandbox projects.
