- Legal Futures - https://www.legalfutures.co.uk -

Meet Ivy, our data-driven adviser for fee-earners

A guest post by Dr Mayowa Ayodele, data process & application scientist at DWF

DWF has embarked on a 30-month knowledge transfer partnership (KTP) with the University of Manchester, with the ultimate aim of transforming how the business uses data and harnessing that data to provide greater value to clients through efficiencies and analytics.

This is the fourth in a series of quarterly blogs on how the KTP is developing, which explores the achievements of the project to date. Read the first [1], second [2] and third [3] blogs.

Ayodele: Data limitations exposed

In the last blog, we shared the expectations of the stakeholders for the KTP, giving an analogy of having an ‘Alexa’ for case handlers. A year in and the KTP is now focusing on building a decision-support system for case handling within motor insurance called Ivy.

This is a data-driven adviser for fee-earners. Ivy will use knowledge from historic data to drive performance within the context of settling cases quicker and exploring strategies that lead to making higher damages savings for our clients.

The development process of Ivy has followed the data analytics stages: descriptive, diagnostic, predictive and prescriptive. While the predictive and prescriptive stages help to predict possible outcomes and also suggest optimal strategies, reliable results cannot be achieved without understanding the data well enough.

A lot of effort has therefore been channelled into the descriptive and diagnostic stages. Here, patterns in the data are explored for better understanding of the data features. Initially, patterns explored in the data were misleading because of incorrect and inconsistent data. While eliminating incorrect data by excluding outliers could be a valid approach, determining whether they are incorrect or correct but unusual is challenging.

To address this, the project has worked closely with the domain experts. This collaboration has helped to expose the implicit business rules and client-specific information. We have also been able to explicitly define some data logic, making it easier to eliminate incorrect data.

Furthermore, working closely with the domain experts has helped to identify data features that can potentially support Ivy. Following this lead, feature selection algorithms have then been applied to identify features that are more likely going to affect the outcome of a case.

The KTP has played its part in exposing some of the data limitations. Some of these have been addressed through training and others by improving the case management system so that it flags some of the common errors at the point of capture.

The next few months will be dedicated to creating a data-driven prototype for Ivy.