Tech Talk: Why Does it Matter? SmartSort with Grace Wu

Through Intuit’s innovative technology, we are able to provide the best possible products for our customers. Behind that technology are the incredible employees who strive every day to bring superior experiences to customers in order to help them prosper. In our series, “Why Does it Matter,” Intuit employees discuss the technology behind innovations they helped

Through Intuit’s innovative technology, we are able to provide the best possible products for our customers. Behind that technology are the incredible employees who strive every day to bring superior experiences to customers in order to help them prosper.

In our series, “Why Does it Matter,” Intuit employees discuss the technology behind innovations they helped create at Intuit.

Grace Wu, data scientist at Intuit, gives us a closer look at the QuickBooks Self-Employed (QBSE) feature, SmartSort, and how her team brought this technology to customers.

+++++

SmartSort: Why Does it Matter?

SmartSort saves customers time by automatically sorting their transactions between “business” and “personal” related categories. Identifying those business-related expenses is critical for maximizing one’s tax deductions.

Before SmartSort, self-employed individuals and small business owners had to manually sort their own transactions, which can amount to hundreds of dollars missed in tax deductions. This is a time-consuming, inefficient, and even intimidating task. With many of the other pressures associated with running a small business, we found that many of our customers put off this task, making it even more difficult to remember why an expense was made.

To automate this work for our users, a team of data scientists, data analysts, data engineers, designers, and a product manager developed a machine learning model to predict the likelihood that a new transaction is business or personal. The model learns from previously reviewed transactions, including information such as whether it was marked as business or personal, its description, and when it occurred. The model can be retrained based on user feedback, allowing us to continually improve its performance.  

We released SmartSort to real users in an A/B test. We used their feedback to tweak the product before bringing it to market for all of our new users. For example, although the algorithm was working and users liked the experience, we found that most users would rather see only our most confident predictions, even if it meant that we weren’t able to predict on more of their transactions.  

In January 2017, we launched SmartSort for QBSE customers. New users can see the algorithm in action when they connect their bank for the first time. After the product was built, we let machine learning continue to power the technology. It is clear that customers are impressed with this feature, as they are more likely to subscribe to our product when they see this automation.

We’re so thrilled that customers appreciate the SmartSort technology. We continue to work on the product, with the goal of applying this technology to any transaction for any given user. Automated categorization saves our users time and allows them to focus on actually running their business.