automated processes (Nikolay Klimenko/Shutterstock.com)

How AI helped California’s DMV adapt to COVID closures

Intelligent document processing has helped the California Department of Motor Vehicles overcome challenges associated with COVID-related field office closures.

“When the offices were closed, we established a virtual field office for our customers where they could request certain services that we could provide remotely,” said Ajay Gupta, chief digital transformation officer at DMV. For paperwork that customers must still submit as a physical document such as vehicle titles for title transfers, DMV set up a digital mailroom using intelligent document processing tools from ABBYY, a digital intelligence company.

The original documents get to DMV in two ways: Customers may submit most documentation via an upload feature supported by automated document classification and extraction from ABBYY, or DMV officials can scan them in. Either way, the documents go through an artificial intelligence-based process for extracting the data, classifying the documents and posting them to the department’s Salesforce-based virtual field office case management system where they interact with bots.  

“These bots are basically doing the work to talk to our legacy systems and process some transactions and complete the whole thing like a technician would have in a real field office,” Gupta said. “The technicians knew that a document has arrived in the [digital] mail and it’s complete and this is the document and here’s the data and next steps.”

DMV also uses intelligent document processing to accelerate Real ID applications. The Real ID Document Verification Project lets customers self-submit cellphone images of the original documents before going to a DMV field office. No physical documents are mailed to DMV.

The AI validates the documents and extracts the needed information. The department also created a human-in-the-loop process so that anything the AI cannot do humans can complete. “Customers would get a notification after they submit all the documents, saying, ‘Yes, your package is complete. You can go to the field office,’” Gupta said.

This process reduces the touchpoints for DMV workers who no longer have to handle the documents because customers have uploaded images of them that were then verified by AI. That also saves paper because employees don’t have to make photocopies and send them to the department’s scan unit, Gupta said.

With centralized, intelligent document processing and the human-in-the-loop approach, it now takes AI about 30 to 45 seconds to review each document. That’s down from two minutes when the effort started in August.

 “We’ve gone from 28 minutes per Real ID transaction to about 10 minutes,” he said. “AI-driven pre-screening is one of the reasons why we are able to do that. It also reduced customer visits because there are reduced returning customers due to incomplete documentation.”

 “The No. 1 benefit to us was elasticity,” Gupta said, especially with the requirement that all of California’s 34 million driver’s license holders transition to Real ID by October 2021.

DMV is using AI in other ways, too. For example, it introduced AI for IVR, or interactive voice response, in August 2020. “It understands your sentiment, it knows what you’re saying in multiple languages, it also tries to find the right answer without having to go through the hierarchical tree to get to the answer as well,” Gupta said. “Now we are looking at expanding that voice channel to add more transactions so people could pay some of their fees … and integrate the voice channel with our back-office channel as well, like the contact center.”

Additionally, the department is planning to roll out automated proctoring for knowledge tests that customers can take at home, further reducing the amount of time they have to spend with DMV workers. That’s one of the main goals of turning to AI, even before COVID, Gupta said.

He boiled the challenges to AI implementations into three categories: technology, people and process.

The tech challenge was that DMV had only a limited amount of data to train the AI, so it started with basic templates and built from there. Also, the quality of the data from self-service was inconsistent, which added to the challenge.

Training and reassigning workers was the people challenge. To address it, DMV observed workers to determine their strengths and increase back-office user experience.

In terms of process, DMV officials wanted to make sure that as they rolled out features, they were not replicating known problems to 176 field offices. To avoid that, they “started small, which was really good because we learned and then we expanded,” Gupta said. “We are still expanding.”

Additionally, the fluidity of the pandemic has made planning tough, he said. With the governor issuing closures and reopenings as the situation warranted, DMV had to be flexible.

But one thing on which Gupta is unbending is that AI has a permanent place in DMV operations.

“Obviously, per-unit transaction cost is lower with self-service channels and AI-based channels,” he said. “It creates more elasticity. It’s more nimble to react to things like COVID. We also had some riots three months ago, so this also gave us an avenue to continue to serve our customers safely.”

“None of this is going away,” he said. “The value proposition has been very clear.”

About the Author

Stephanie Kanowitz is a freelance writer based in northern Virginia.

Featured

  • Records management: Look beyond the NARA mandates

    Pandemic tests electronic records management

    Between the rush enable more virtual collaboration, stalled digitization of archived records and managing records that reside in datasets, records management executives are sorting through new challenges.

  • boy learning at home (Travelpixs/Shutterstock.com)

    Tucson’s community wireless bridges the digital divide

    The city built cell sites at government-owned facilities such as fire departments and libraries that were already connected to Tucson’s existing fiber backbone.

Stay Connected