en

Marlette Funding, Best Egg Loans: Fraud Detection with ML

Marlette Funding, Best Egg Loans were able to improve their fraud detection capacities by 10 percent by switching to a machine learning based model.

For most modern businesses, talking about data science, machine learning, and – increasingly – AI is exciting. It’s the future, and it means cutting-edge development and change. For financial services, these terms tend to evoke less enthusiasm and more fear (especially when the term “black box” comes up). And rightfully so – to be sure, the industry’s relative hesitation in embracing these technologies is thanks in part to a generally stricter and higher-stakes regulatory environment.

There’s no denying that it requires financial service companies to be prudent. Yet despite these challenges, there are some institutions rising above the rest, working on the cutting-edge of the data science world in creative applications that improve business results and make the customer experience better.

Unfortunately, machine learning is not really an accepted or mainstream practice in the financial service industry for compliance reasons. That’s partly because whenever someone talks about ML, everyone’s mind goes to underwriting – but not us. We found several other opportunities to be much more attractive.
– Evgeny Pogorelov
Director of Decision Science at Marlette Funding

Get the guide to Fraud Detection in Banking

Marlette Funding, Best Egg Loans is using machine learning (ML) to transform business processes across the organization in revolutionary ways. To make sure they produce a best-in-class fraud detection model for their first foray into ML (and best-in-class data projects in general when working with other parts of the business), the six person team at Marlette Funding:

  1. Considers return on investment (ROI). Before taking on a data project, the team considers first and foremost the potential business impact of the project. In the case of fraud detection, they calculated that if the model were to catch even one instance of fraud, they could save a personal loan lender an average of $15,000. But they also considered indirect benefits, like the fact that a more sophisticated model would speed the process of getting a loan for customers by minimizing the number of cases that are not fraud.
  2. Gathers all available data. The key to an innovative data science project is to put as much data in to create the model. In the case of fraud detection project, that means creating a massive dataset to work with using not only internal data, but externally available datasets from credit bureaus, fraud detection vendors, and more.
  3. Tests/ benchmarks against current strategy. It is necessary to compare developed models with the current solution because if the performance is not better than the one of the existing system it will cause more unnecessary work in monitoring.
  4. Deploys to production. Once tested and benchmarked, they are put in production, where they can actually have a real impact on the business. The fraud detection model at Marlette Funding is currently deployed and generating cost savings for the lenders.

Data analysts at the core of the organizational structure

Most of the business units at Marlette Funding have their own analysts who look at data and for opportunities for more advanced analytics. From there, they can approach the central data team to collaborate on projects together. This allows the technical skills of the data team to be enhanced by the business knowledge of the analysts and other experts in business units for more optimal project results.

Here’s how you can create your Data Team.

Pros: 

  • Extremely tight correlation between data projects and business value
  • Different skillsets on the data team allow them to work on a wide range of projects
  • Small, agile team means they can move quickly on projects

Cons:

  • Small data teams, if they lack an easy way to deploy and manage models in production, can have trouble scaling.

The data science team doesn’t have a business function on its own – it serves the entire business. So the data team worked with the fraud operations team, for example, who can provide the relevant data and knows the overall fraud strategy.
– Evgeny Pogorelov
Director of Decision Science at Marlette Funding

How Marlette Funding, Best Egg Loans Uses Dataiku:

  • Deployment to production (one-click deployment)
  • Data blending, manipulation, & feature engineering
  • Machine learning model creation

The way we see it, we’ve already done all the traditional modeling and looked at the traditional data. If we want to be the best in class (the best in marketing, the best in fraud detection, the best in customer service, the best in pricing, etc.), we need to go beyond the traditional tools.”
– Evgeny Pogorelov
Director of Decision Science at Marlette Funding

Rabobank: Ethical Enterprise AI – A Guideline or Compass?

In this EGG talk, Martin Leijen shares how Rabobank determines and protects their privacy and ethical standards, as well as how financial institutions can we effectively maintain a firm commitment to moral and ethical standards while at the same time encouraging a strong drive to optimize business opportunities and profitability.

Read more

Deep Learning for Better Option Pricing

An in-depth walkthrough of a use case for deep learning (GANs) in the finance industry.

learn more

AI in Banking

Most banks fundamentally have a lot of the pieces in place already for AI; this white paper walks through challenges, use cases, and initiatives to start now.

get the white paper

Infographic: Top Banking AI Use Cases

A quick walk-through of the best path forward for banks getting started in AI.

learn more

Fraud Detection Use Case

Get a demo of an end-to-end use case for putting a machine learning-based fraud detection system in production.

watch now