Rethinking UX for Traditionally Unpleasant Experiences like Tax-Filing: How Intuit Did It

Humans can make educated inferences from unclear visuals; a machine cannot. This was a hurdle Intuit sought to overcome when it decided to revamp the necessary evil of tax-filing into a simpler, automated customer experience.

Ease customer pain points for experiences that are traditionally unpleasant

Intuit cultivated a cult-like following with TurboTax by allowing customers to file taxes from their smartphones. Simply take a photo of your W-2 and other applicable forms and the machine learning software will “read” the image and populate your tax form with the corresponding information.

“We took a step back to try and understand from a mobile perspective how we could leverage the smartphone to make our customers’ lives easier,” Desiree Gosby, VP of identity and profile at Intuit, said at the O’Reilly AI Conference in New York City. “When it comes down to your finances and taxes, you’re still dealing with a lot of paper that requires lot of manual entry to get that data imported into our system.”

Know the limitations of the technology and adjust the user experience accordingly

Data entry is cumbersome enough on a desktop interface – be it filling out a contact form or a survey – but it doesn’t work for a mobile interface, especially when numerous steps are involved. When Intuit proudly rolled out the first iteration of its image-capturing feature on the mobile app, the company didn’t realize just how difficult it is to take a photo of a tax form in the right conditions for a machine to understand.  

People were submitting photos of folded or crumpled W-2s in poor lighting taken on mattresses, cluttered kitchen countertops, textured carpets and even closed toilet lids. As a result, the AI would often pick out multiple documents in the background, or be unable to read it altogether because of poor contrast and brightness.

“Needless to say, our accuracy rates were not great when we first started releasing this feature; it was not a great customer experience,” said Gosby, “and customers had no problem letting us know.”

After fielding innumerable ventings from frustrated customers on social media, the team started to understand the limitations of the technology and incorporate assists into the user interface to guide customers on how to take a usable image. Intuit identified three pain points it needed to address to fix the CX gap:

1. Define how to get a good image
2. Teach the computer to recognize the different types of forms and their respective layouts
3. Make sense of the data and apply it in the right context

Designers added edge detection, contrast, brightness and focus adjustments to the image-capturing feature.

Once they’d accumulated a corpus of hundreds of millions of documents they could teach the AI even better techniques like foreground detection and automatic image cropping, and also “giving the customer a better experience by predicting upfront if this isn’t going to be an image that we can actually process...and directing the user to a different method of correcting the W-2.”

Today, the technology is trained to differentiate between W-2s, 1099-MISCs, 1098-Ts, and other less common types of tax forms and understands their different layouts. The company knows it’s doing a better job because it’s customers say so.

“Just like customers let us know they were not happy with us when we first started, they let us know when we were getting it right,” said Gosby.