Follow Datanami:
January 17, 2020

AI… for Pizza?!

If you order a pizza from Domino’s, you might be getting it with a free side of AI. The pizza giant is determined to make big data and AI cornerstones of a company that has typically focused more on pizza stones. Now, Nvidia is highlighting how Domino’s is leveraging the power of data to deliver valuable insights in addition to pizza.

Zack Fragoso, a data science and AI manager at Domino’s, explained how the company had grown its data science team exponentially – a move “driven by the impact [the team] had on translating analytics insights into action items for the business team.”

Domino’s made its first public foray into AI with “Points for Pie,” a Super Bowl ad stunt that allowed customers to send a smartphone picture of (any) pizza to Domino’s, earning points that could be used for free pizza. “No one was sure how to recognize purchases and award points,” Fragoso said. “The data science team said this is a great AI application, so we built a model that classified pizza images. The response was overwhelmingly positive. We got a lot of press and massive redemptions, so people were using it.”

Nvidia’s DGX-1

Domino’s is also starting to operationalize AI for a more common use case: pizza delivery. Fragoso and his team have tackled the model that predicts when an order will be delivered by assessing the number of employees working, the orders in the pipeline and current traffic conditions, improving its accuracy from 75% to 95%. 

Domino’s has been training these models on an Nvidia DGX system with eight V100 GPUs. “Domino’s does a very good job cataloging data in the stores, but until recently we lacked the hardware to build such a large model,” Fragoso said. “Once we had our DGX server, we could train an even more complicated model in less than an hour. That let us iterate very quickly, adding new data and improving the model, which is now in production in a version 3.0.”

Next, Domino’s is planning to use AI – and a set of Nvidia Turing T4 GPUs – to improve its real-time prediction capabilities, such as using computer vision to improve the pizza ordering process. “Model latency is extremely important, so we are building out an inference stack using T4s to host our AI models in production. We’ve already seen pretty extreme improvements with latency down from 50 milliseconds to sub-10ms,” Fragoso said.