How to gain an advantage with machine learning

Data scientists are turning to AI and machine learning to help them manage and make sense of all the unstructured or semi-structured data that organizations are frequently drowning in.

In a ghost broking scam, fraudsters purchase legal insurance policies with fake or stolen information before selling them to unsuspecting customers. It is one of the most difficult types of fraud for insurance firms to detect. It was time-consuming to monitor for insurance giant Covéa, which sends over 2 million quotes per day.

Covéa can now scan insurance policies with artificial intelligence thanks to machine learning (ML). A set of machine learning models scans policies 24 hours a day, analysing millions of individual bits of data. The model can predict whether or not a policy is fraudulent with a high degree of accuracy. “Previously, the financial crime team would review policies and make a decision based on ‘gut feel,’” explains Tom Clay, Covéa’s principal data scientist. “We created a model that mimics how their gut instinct functioned but can scan more data faster.” If the ML detects a policy, it is forwarded to the financial crime team to be manually reviewed.”

49% of CIOs have already implemented or plan to implement AI and machine learning technology in the next 12 months.

In a decision-making process, the human brain can only consider a few variables. Machine learning, on the other hand, can handle hundreds or thousands of variables at a considerably faster rate than humans. Machine learning could assist organisations in solving a variety of challenges as data volumes double every 18 months. Traffic applications that suggest you the best route home are a basic example of how this works. The software examines aspects such as weather, historical traffic reports, time of day, and roadworks before determining the best route.

Organizations have an abundance of data, but it is often underutilised.

The concept of developing a predictive model based on past data appears straightforward. According to Case, very few organisations have a single data source that may serve as “ground truth data.” Organizations contain data in a variety of formats, silos, and locations, making it difficult to gain a whole picture. Furthermore, implementing, scaling, and managing models throughout an organisation may be a time-consuming and costly process.

According to Clay, when Covéa launched its AI initiative, only 2% of its customer data was regularly labelled. He believes that it takes nearly two years to properly identify all of the data and train ML models. “When it came to constructing the ML models, we had to beg the business to be patient with us.” “We employed a cloud-based MLaaS service to help us reliably design, test, and deploy models, which resulted in an 11% ROI in six months,” he explains.

The good news is that the rise of off-the-shelf machine learning apps and ML as a Service (MLaaS) platforms has made it easier to get started with machine learning. Large solution providers can assist in the construction of ML models and undertake the heavy lifting of data analysis, relieving the burden on in-house IT staff.

MLaaS is a new collection of services that can help businesses get started with machine learning more quickly. They can provide processing power for AI tasks as well as assist development teams in developing, building, testing, and deploying models at scale. According to Transparency Market Research, spending on MLaaS is estimated to reach almost $20 billion (£16.7 billion) by 2025.

That comes as no surprise to Case, who predicts a surge in demand for enterprise AI services over the next five years. “By implementing AI technologies, organisations may better understand their customers, allowing them to develop and target products more efficiently.” They can automate tedious or repetitive tasks, saving money and freeing up workers for more vital tasks. And in many situations, it enables organisations to do things that were previously impossible,” he argues.