{"id":15364,"date":"2022-03-26T23:31:02","date_gmt":"2022-03-27T03:31:02","guid":{"rendered":"https:\/\/digital.hbs.edu\/platform-digit\/?post_type=hck-submission&p=15364"},"modified":"2022-03-28T12:13:47","modified_gmt":"2022-03-28T16:13:47","slug":"the-secret-surveillance-sauce-to-dominos-success","status":"publish","type":"hck-submission","link":"https:\/\/d3.harvard.edu\/platform-digit\/submission\/the-secret-surveillance-sauce-to-dominos-success\/","title":{"rendered":"The Secret Surveillance Sauce to Domino Pizza’s Success"},"content":{"rendered":"\n\n\n
In 2019, Domino\u2019s ran a Super Bowl commercial that had hungry viewers intrigued:<\/p>\n\n\n\t\t
\n\t\t\t \n\t\t\t\t<\/div>\n\t\t\t<\/figure>\n\t\n\t\t\t\t\t\t\t\n\t\t\t\t\tThe 12-week promotional campaign invited anyone to take a picture of any pizza (Domino\u2019s or not) via the Domino\u2019s app. For every picture the AI-enabled app identified as \u201cpizza,\u201d customers earned ten points (with a limit of one pizza scan per week). Once customers reached 60 points, they were able to redeem them for one free medium two-topping pizza from Domino\u2019s.<\/figcaption>\t\t\t\t<\/figcaption>\n\t\t\t\t\t<\/div>\n\t\t\n\n\n
As pizza-lovers rejoiced, Domino\u2019s and its strategic partners, Nvidia and Dragontail Systems, hoped to couple AI and the campaign to tackle customers’ vexing complaints [1].<\/p>\n\n\n\n
Using a deep learning neural network that classified pizza images, a model was built on an NVIDIA DGX and trained using more than 5,000 images of pizzas. Automation helped label the unique dataset, and the system was even able to recognize unexpected images of plastic, dog toy pizzas [2]. <\/p>\n\n\n\n
Two years later, the DOM Pizza Checker was born.<\/p>\n\n\n\t\t
\n\t\t\t \n\t\t\t\t<\/div>\n\t\t\t<\/figure>\n\t\n\t\t\t\t\t\t\t\n\t\t\t\t\tThe DOM Pizza Checker uses advanced machine learning, artificial intelligence and sensor technology to identify pizza type, even topping distribution and correct toppings. The AI-enabled tool sits in the kitchen, checking pizzas when they come out of the oven. It scans for pizza type, correct toppings, topping distribution, and aesthetic appeal. It’s main focus is to review the quality of the pizza before delivery, reducing the \u201cmy pizza doesn\u2019t look like it should\u201d complaint rate.
The camera captures the image of pizzas placed on the cutting bench and analyses the pizza to make sure it meets Domino\u2019s quality standards. If it does, the system matches the pizza to the appropriate order, displaying it on the screen with a corresponding quality score. If the pizza has an issue, the checker will alert the manager with a visual pop-up and a sound and display the problem detected. The manager can then decide either to agree with the computer, in which case the system will automatically re-log the order, or override it, whereby the system will keep a log but move on to the next pizza. A thumbnail of the picture is then sent to Domino\u2019s tracker, allowing customers to check their pizza, targeting a growing desire for transparency.<\/p>\n\n\n\t\t
\n\t\t\t \n\t\t\t\t<\/div>\n\t\t\t<\/figure>\n\t\n\t\t\t\t\t\t\t\n\t\t\t\t\tThe DOM Pizza Checker is deployed in 850 stores across Australia and New Zealand and now expanding into the U.S.<\/sub><\/figcaption>\t\t\t\t<\/figcaption>\n\t\t\t\t\t<\/div>\n\t\t\n\n\n
The system sends all logged inputs to Domino\u2019s headquarters. This shared data allows the head office to check products sold under its brand name, score each store, determine how to allocate training resources effectively, and ultimately reduce the need for auditing.<\/p>\n\n\n\n
The solution has also found unique value in its ability to quickly add additional features. For example, in response to the increased customer attention to health and safety during COVID, a cleanliness-checking feature was added to the camera ensuring the cutting bench was cleaned as often as the company\u2019s standards require.<\/p>\n\n\n\n
Domino\u2019s claims to have witnessed a 14% to 15% improvement in product quality in stores that installed the AI tool. While the cameras\u2019 adoption was not entirely smooth and skepticism of the technology causing initial doubts in some franchises, the overall project was received quite well and franchisees also added that it created a climate of positive competition, motivating employees to improve the quality of their pizzas [3]. <\/p>\n\n\n\n
AI: The New Hot Topping<\/strong><\/h2>\n\n\n\n
In 2020, Domino\u2019s AI success pushed them to make a substantial investment in migrating from CPUs to GPUs in order to improve the accuracy rate of predicting when a customer\u2019s pizza would be ready.<\/p>\n\n\n\n
With CPUs, it took three days for Domino’s to train an algorithm to predict when orders would be ready. Using GPUs, the company was able to train the same algorithm in less than an hour (and continuously retrains it using more data every day). These enhanced algorithms improved delivery time accuracy from the previous 75% to 95% accuracy rate by leveraging a load-time model that factors in labor variables such as how many managers and employees are working, the number and complexity of orders in the pipeline, operational factors, and even current traffic conditions to improve the order-ready accuracy rate [4][5]. <\/p>\n\n\n\n
![\"\"](\"https:\/\/d3.harvard.edu\/platform-digit\/wp-content\/uploads\/sites\/2\/2022\/03\/Screen-Shot-2022-03-26-at-9.50.01-PM-1024x574.png\")
<\/a>![\"\"](\"https:\/\/d3.harvard.edu\/platform-digit\/wp-content\/uploads\/sites\/2\/2022\/03\/Screen-Shot-2022-03-26-at-9.50.21-PM-1024x870.png\")
<\/a>![\"\"](\"https:\/\/d3.harvard.edu\/platform-digit\/wp-content\/uploads\/sites\/2\/2022\/03\/Screen-Shot-2022-03-26-at-10.06.54-PM-1024x516.png\")
<\/a>![\"\"](\"https:\/\/d3.harvard.edu\/platform-digit\/wp-content\/uploads\/sites\/2\/2022\/03\/Screen-Shot-2022-03-26-at-9.50.01-PM.png\")
<\/a>