You Asked, We Answered: 7 Top Questions on Automating Mortgage Processing

The banking and financial sector is particularly ripe for digital transformation: Many complex and highly manual workflows can be improved and streamlined through AI-powered RPA. One such common task is mortgage processing. In a recent webinar, WorkFusion Machine Learning Engineer Arnesh Sahay discussed how various aspects of mortgage handling can be automated, and why the combination of machine learning and robotics is crucial to doing so. At the end of the presentation, participants offered some very interesting and insightful questions, with many focusing on WorkFusion’s machine learning capabilities. Read on for Arnesh’s answers:

What are the core capabilities of WorkFusion SPA and what functions do they cover?

  • WorkFusion’s RPA provides bots for any type of application or data.
  • WorkFusion’s Everyday AI allows you to easily apply machine learning to processes that contain unstructured data.
  • Optical Character Recognition (OCR) is used to digitize image-based data in any language and can be added to any process.
  • Our built-in Process AutoML quickly trains on your unique data sets, in your secured environment, operationalized in your processes — without the efforts of a data science team or costly third-party integrations.
  • WorkFusion’s Workflow capabilities allow users to configure end-to-end processes and automate the routing of work to the right bot or person at the right time.
  • WorkFusion’s Operational Analytics enables operations owners to predict costs, quality, capacity, and productivity by gathering insights across bots, manual workers, and process data.· WorkFusion’s Operational Analytics enables operations owners to predict costs, quality, capacity, and productivity by gathering insights across bots, manual workers, and process data.

How is an ML model initially trained?

WorkFusion’s Process AutoML capability runs thousands of experiments in the background while your team completes their day-to-day tasks within WorkSpace. By capturing insights from the generated training data, the ML model learns how to perform those same tasks and gradually completes more work on its own as it gains more confidence in its responses. We have created an animated video that visualizes this process. Watch it below.

How long does it take to train an ML model?

WorkFusion’s Process AutoML has optimized the machine learning process to reduce the number of examples that are needed to start deriving value from models. For example, our information extraction models, which collect data from documents like insurance claims, mortgage deeds, or identification documents, are often trained with an initial set of 500 documents, not the hundreds of thousands that are typically required. WorkFusion’s R&D team is also working on integrating approaches like Differential Privacy and Transfer Learning to further optimize the model training process.

How is the quality of training data identified?

When you consider the quality of training data it’s important to not just collect data upfront, but to also take the ongoing maintenance of the model into account. This is simplified with WorkFusion SPA. When collecting the training sets themselves, features like peer review and data adjudication are applied for initial and ongoing training data collection. When executing the model, only answers with high confidence are accepted, with others sent for manual completion. Furthermore, for these high confidence answers, an AutoQC approach can be applied to monitor outputs and further adjust training set quality.

How does the ML model know where the APN number is in the document?

WorkFusion’s SPA ML models can learn by picking up on a variety of features in the underlying data, such as words or symbols around a key data field or the position of a field within a document. If a certain feature is heavily present on documents within the training set, then a higher ‘weighting’ will be associated with it, which means that the model can be more confident in its prediction. Hence, extraction of key data fields is solved through the model’s understanding of the data across a variety of layouts, not by relying on template-based approaches.

What are the differences between WorkFusion SPA and RPA Express*?

RPA Express* is our free product offering available for download on our website. With RPA Express*, business users or developers alike can create RPA bot tasks and manual tasks to automate end to end business processes. WorkFusion SPA introduces Process AutoML and operational analytics features such as a centralized Control Tower, in addition to the features offered within RPA Express*.

What are some measures WorkFusion takes to maintain confidentiality within ML models?

WorkFusion’s customers own their data since most models are trained on premise and the model never leaves the customer’s secure environment. However, WorkFusion has introduced Differential Privacy and Transfer Learning to share models with customers’ permission to achieve the joint benefits of larger, secure data sets.

*RPA Express is no longer included in WorkFusion’s product offering.

Also published on Medium.

Get our blogs delivered to you.
Arnesh Sahay
Arnesh Sahay

Machine Learning Engineer, WorkFusion

Share this article