How We Harnessed Deep Learning for a Better Customer and Partner Experience
At BMO Financial Group (BMO), Jon Wang, Head of the Digital Core Technology group, and his team have found a way to use deep learning technology to automate the capture of customer data. It reduces the amount of time required to process car loans requests, ultimately leading to a smoother and quicker customer and partner experience.
Customer data is the lifeblood of the financial industry. But for something so important, there is very little standardization in the way that information is collected. It seems that no two organizations use the same basic information forms. Even in this digital age, a large part of the world also relies on paper to capture customer data. The challenge is how to capture that data quickly without the burden of keying it in manually.
Bo Wu – the Lead Technology Officer for the Digital Core Digitalization team at BMO – explains the problem using the example of shopping for a car. Once a relatively relaxed experience — when dealers lots were full— purchasing a car has become as competitive as buying a home. Buyers can’t be certain that the car they look at in the morning will still be on the lot after lunch.
This puts pressure on everyone, including the bank financing the purchase. “We used to have days to process applications for car loans,” says Wu, “but customers need to move quickly to get the car they want, and we need to be able to facilitate this in hours.” The Digital Core Technology team had to find a way to automate the capture of customer data, eliminating the time-consuming process of manually keying it in.
Digital highway to smart scanning
For BMO, the answer is deep learning, which is modeled on how the human brain processes information. Wu explains further that “although deep learning is far from matching human cognition, it can be trained to do many tedious and inefficient manual tasks, including automated document processing.”
The digital highway starts with scanning documents at the branch or the car dealership using a variety of electronic content management solutions. Progressing down the highway, however, requires a cloud-based, automated way to turn electronic documents or e-documents into processable data.
Using the cloud-based Digital Core technology, Wang’s team first trained the deep learning neural net to extract information from known, structured data forms such as a government T4 slip or internal bank forms. Since structured forms are consistent, this was the simpler problem to solve.
Next on the list was unstructured documents. By taking hundreds of examples of a similar kind of document, such as an agreement to purchase a vehicle, the software eventually learned to identify the important information – with much help from the trainers, who correct it when it makes errors. The neural net must also be able to extract information and recognize what the document is not. Sometimes different documents can contain much of the same information. For instance, a car warranty and a car purchase agreement could have as much as 70 percent of the same information. Often a single word is the key to recognizing the document, such as the word “new” or “used”.
Already reaping benefits for customers and partners
According to Wu, the program is now able to process over 1,500 in-scope documents per day, meaning that it has increased processing times by a multiple of 20. It also reduced errors that need to be corrected along the way. Added up, it’s a significant cost savings. More important than bottom line savings, from a customer perspective, is the speed with which applications such as car loans can now be processed. This makes the car loan experience much more appealing for customers and demonstrates the strides that BMO has made in offering value-added digitally enabled services.
The new automated process mitigates operational risk by reducing manual keying errors,and the built-in encryption provides greater security for the customer data. From the dealer’s perspective, it is a better experience – which makes BMO a more attractive partner for dealerships and manufacturers.
Leveraging the digital core
The great advantage of the BMO Digital Core as an AI platform is that it can be leveraged by many other automatable operations in the bank. The processing capabilities that underlie the neural net are not specialized to the car loan approval function.
“The most exciting part of this project,” says Wang, “is how broadly we can deploy the technology. Throughout the project, the Digital Core Technology team has constantly been thinking of new things we can train it to do. Document management is just the beginning.”