Becoming the Amazon of Drive-Thru: A Quick Look at Segmenting Drive-Thru Customers Through AI and Machine Learning

When Jeff Bezos founded Amazon in July 1994, there were just 10,000 websites on the internet. They were only discoverable through rudimentary search engines that turned up little, if any, information relevant to the locally minded behaviors that companies have since sought to track, from ordering food from a nearby restaurant to searching for the closest drive-thru. Just a month later, the LA Times was calling Pizza Hut’s PizzaNet online pizza ordering portal a “Geek Chic” way to get food, reflecting how far away from the mainstream such a concept was at the time.

Today, Amazon is a master at using data to guide its business operations. Amazon Pinpoint, a technology under the Amazon Web Services (AWS) umbrella, uses companies’ real-time and historical data to create customer segments and campaign messages triggered by personal behaviors, such as adding a certain item to the shopping cart. People from Amazon’s inception year would have found it hard to imagine how segmenting customers and targeting certain behaviors could help improve speed of service at drive-thrus, but with 90,500 people a month now searching for a “drive thru near me”, more restaurants are building drive-thrus to focus on consumers who are picking up food they already ordered and paid for through an app.

By purchasing Dynamic Yield in 2019, McDonald’s is working on personalizing the ordering experience through collecting data on the weather, time of day, trending menu items, and restaurant traffic. This marks an important change over the past; when deciding which information to capture, restaurants now need to think beyond demographics to compete. After collecting a large amount of customer data, Dynamic Yield uses multi-armed contextual bandits instead of A/B testing in order to determine which menu items to recommend when customers pull up to a McDonald’s drive-thru. This new approach allows McDonald’s to test multiple combinations of suggestions on its digital menuboards and maximize the types of offerings that customers actually choose. For example, if several drive-thru customers who arrive for lunch during the summer do not order the McFlurry suggested on the order screen, Dynamic Yield’s software will automatically scrap this suggestion for customers who fit this criteria and recommend other menu items until it makes a successful upsell.

Through using AI to learn, revise, test, and re-learn, Dynamic Yield creates customer segments in real-time, allowing McDonald’s to change its menuboard messaging to fit countless variations of customers until the software identifies the right categories to target. Viant, an advertising technology company, published a report that separates QSR customers into 5 categories. These are not one-size-fits-all, so make sure you back up each category you generate by sales and ordering data that can help you improve the drive-thru experience for each segment. Viant, for instance, uses the following criteria, among other metrics:

  • Breakfast Buyers – 15% of visits occur during breakfast hours
  • Lunchtime Loyalists – 90% of visits occur from 12pm to 1pm
  • Primetime Patrons – 95% of visits occur during dinner hours
  • Weekenders – 35% of visits occur on Saturday or Sunday
  • Devoted Diners – make 300% more visits in a 110-day period than the second most loyal group

Remember to make the segments relevant to your target market, meaning the names and criteria for each may vary from one QSR concept to another. Viant’s model is from a study of 1,889,441 customers who visited a 1,000-store national sandwich chain a total of 4,880,438 times. The company used a machine learning algorithm called k-means clustering, which is a method of randomly selecting customers from the sample dataset, averaging their visits for each hour of operation to create clusters, and grouping the rest of the customers based on how closely their data matches each cluster. The process repeats until the average of each cluster, which in this case would be the average number of visits, stays the same.

Especially if your restaurant concept does not have its own machine learning engine like McDonald’s has with Dynamic Yield, you can outsource this part of the data collection and analysis process and use the insights to improve the drive-thru experience for the different customer segments that result. For example, if you owned the sandwich shop that is a part of Viant’s model, you could highlight Coca-Cola as a popular soft drink option on the digital menuboard during lunch hours, since Viant found that Lunchtime Loyalists are more likely to opt for this beverage than other groups. With the vast majority of lunch visits occuring from 12pm to 1pm for this example, you could also send an employee outside to collect and confirm orders with a tablet POS system to speed up drive-thru service during this peak hour.

With machine learning techniques in your toolbox, you’ll be able to identify and target your drive-thru customers more efficiently. McDonald’s has already gotten a big head start, but the basic tools to make data collection more intuitive are available from a number of sources. For more articles on drive-thru technology, check out our posts on improving drive-thru efficiency and how artificial intelligence is transforming the drive-thru.

Brink Cloud POS Software: Schedule your Brink Demo