How Collectors and Verizon use AI: Billion-dollar plans and 1,000 models

Plenty of arguments surround AI’s long-term worth, but it is hard to argue with billion-dollar results. C-suite leaders from Verizon and Collectors spoke recently at the OverdriveAI Summit, shedding light on how their companies currently use AI. TechUnited hosted the summit at Nokia Bell Labs in New Providence, NJ. Verizon and Collectors operate in very different digital and analog spaces, and use AI for different tasks, but they are seeing its impact on their operations. In the case of telecommunications provider Verizon, AI is used for predictions and efficiency. Meanwhile, Collectors use the technology to authenticate prized items. As different as their needs are, these tech leaders have growing plans to put AI to work within their enterprises.

Verizon’s three-prong AI strategy

Mano Mannoochahr, chief data, analytics and AI officer for Verizon, opened up his keynote with an outline of the strategy his company adopted to leverage AI. Before Mannoochahr broke down each element of the strategy, he gave a glimpse of where Verizon is headed with some of its efforts. “We want to bring AI closer and closer to the edge, because that’s where it gets consumed most of the time on the phones,” he said. Verizon has enabled AI resources for all its employees to enhance knowledge workers’ capabilities on strategic tasks, Mannoochahr said. That might come into play with significant reported changes underway at the company. Last week, it was reported that Verizon plans to lay off 15,000 employees to cut costs while dealing with rising expenses from customer retention. The reported layoffs come just weeks after Dan Schulman, formerly of PayPal, took the reins as CEO in October. Verizon has not yet responded to questions regarding the reported layoffs. AI is definitely being put to work by Verizon, which currently has more than 1,000 AI models in production that run various aspects of the business, Mannoochahr said. One of the models is used to predict why current customers call the company by analyzing their experiences on the network over the two weeks preceding the call. “It’s looking at hundreds of data points in real time,” he said.

Training AI to know customers

Citing himself as an example, Mannoochahr said that when he called Verizon customer service, the AI model presumed he was checking the status of an order he had placed for new phones. “As soon as you’re trying to make a touch point with us, we’ve got AI that’s already trying to predict why you might be calling us,” he said. After the initial assessment of presumed needs, the AI model tries to pair the customer with an agent who is suited to address their specific issues. “This is where we’ve got an AI model that’s looking at additional data and then trying to figure out who the best person to handle your call might be,” Mannoochahr said. Once connected with a representative, additional AI models may activate and listen to the conversation with customer service in real-time and then offer suggestions on how best to provide assistance, he said. This also means millions of calls to Verizon get digitized and can be consumed by its ad models for marketing needs, as well as used for other parts of the business, Mannoochahr said.

Not all about going big with AI

Verizon also trained a small language model to respond to questions about changes in billing, he said. The company also works with prescriptive AI that could predict if a customer might need additional resources to keep them from switching to another carrier. Verizon is also using generative AI to develop advisory tools that bring data, information, and knowledge together for decision-making, he said. At least for now, he said, consumers represent the largest users of AI as enterprises are just at the beginning of their journey with the technology, but that is expected to shift. “In the foreseeable future, we’re going to have humans and AI agents that will be working together in large companies as well as with consumers,” Mannoochahr said.

AI authentication and a house of cards

Dan Van Tran, CTO for Collectors, delivered a keynote discussing how AI and other changes contributed to his company’s growth from a $850 million valuation to $4.3 billion in valuation over 14 months as of 2022. Collectors has since gone private, and Van Tran said the company keeps its current valuation secret. Collectors authenticates and grades collectibles such as trading cards and autographs. Despite being a legacy business, Van Tran said AI assisted with its core operations. During his speech, Van Tran held up a trading card valued at more than $1 million in order to demonstrate the scrutiny collectibles can face. The company got its start in 1986 in collectible coins, diamond authentication, and stamp collecting. The company evolved into tradeable cards, and when the pandemic struck, the public poured money into collectibles as they remained at home. Van Tran said he joined in 2021, as the company sought to build up its technology to accommodate the surge in demand. “They had me come in to reimagine the technical foundation.” Part of that reimagining included the use of AI to identify counterfeit collectibles. For example, Van Tran displayed two similar-looking 1986 Fleer Michael Jordan rookie cards, which could be worth well in excess of $100,000 based on the condition and grade. One card was a fake. Professional graders of collectibles, he said, spend years training to understand the differences in an item’s condition and how to find counterfeits. It can be a simple tell, such as the absence of a pattern of dots on a trading card. Modern-day counterfeiters work with more advanced printing technology that can be harder to detect. Even with new training to help graders catch advanced counterfeits, the volume of the market has ballooned, making it difficult to keep up.

Using AI to keep pace with demand

“It’s a hard challenge because there are anywhere between 450 [billion] and 650 billion cards that have been manufactured over time,” he said. Graders must understand the nuances of every single one in order to determine if a card is a counterfeit or not, Van Tran said. The technology update at Collectors included the development of a custom machine with automation that scans cards and performs quality control. “Instead of taking about seven minutes per card, it takes about seven seconds,” he said. The complete replacement of humans in the review process will take time, he said. For now, Collectors developed AI and ML models that can give recommendations that are still reviewed by human experts. “Those responses help to train the model to be better for the next set of collectibles,” Van Tran said. AI also plays a role with customers who turn to Collectors to assess their items via a mobile app. The company released a function earlier this year that compares images submitted by users through the app to the database built up by Collectors. Van Tran said the feature initially saw less than 10% matched, accurate assessments. The team increased the match rate by first having an LLM look at the image and then describe it in text. A text search across databases and metadata boosted the match rate to 85%. Van Tran said Collectors also uses AI to help software engineers with coding, which is valuable as the team ships more features and more code endures longer. “It just went to show for us that you can’t just rely on the modern AI/ML technologies, that you actually had to just go back to your roots, go back to the fundamentals of good software engineering, good design,” Van Tran said.