How McDonald's is building a data-driven retail experience

McDonald's have used screens to deliberately move the decision and purchase point away from the counter and onto digital surfaces - where they have complete control of how, when and what they show to customers. These digital surfaces become the perfect platform from which Dynamic Yield’s powerful personalisation engine can go to work. On-screen content can be personalised to the individual or changed based on what’s trending in the restaurant. The meals they offer become more relevant, more timely and ultimately more appealing.

McDonald's, once famed for their practice of mass-marketing, are now in the business of mass-personalization.

Starting with the Drive-thru

The first deployment of Dynamic Yield’s technology will be in the Drivethru. In the US, the majority of customers don’t leave their cars to order their McDonald's. However the Drive-thru journey itself has barely changed since its initial introduction, making it ripe for a digital shakeup.

It’s also a great candidate for this kind of experiment because it offers a very defined ‘one-car-at-a-time’ journey in which each stage can largely be controlled and the impact measured. So what will this new digital Drive-thru feel like for customers:

1. As you approach the Drive-thru you’re greeted with standalone screens displaying banners and promotions. This is largely to prime you before your menu choice. Promotions that were previously driven by marketing can be based on car size or number of people within the car.

2. You arrive at the main ordering station where the menu is now a large digital screen. The menu displayed is now completely personalised to that particular McDonald's. Algorithms crunch data such as weather, local traffic, nearby events and historical sales to curate the most appropriate menu and surface popular items.

3. As you talk through your choice, your order is built on the screen in front of you. Add-on recommendations are displayed in real-time alongside your order, such as a ‘tasty to go with…’ milkshake for a couple of extra dollars. Instant visual bundling of additional products outperforms typical up-selling techniques and can be deployed across surfaces such as the self-serve kiosks.

Over time, as the algorithm gets exposed to more customers and more training data - more powerful insights can start to be generated. For example, if someone orders two Happy Meals at 5pm, we might infer it’s a parent ordering for their kids, therefore highlight a coffee or snack for them and they might decide to treat themselves.

More broadly the algorithm could start to make the whole Drive-thru process more efficient based on demand. For example, if the Drivethru is moving slowly, the menu can dynamically switch to show items that are simpler to prepare, to help speed things up. Likewise, the display could highlight more complex sandwiches during a slower period.

You can start to see how predictive analytics plays a role in every aspect of the business, starting with insights that help the kitchen more accurately and efficiently prepare food based on real-time demand.

Additional opportunities exist all the way down the supply chain including the purchase of stock, where stock decisions can be made on a more granular and localised basis. For a high-volume, low-margin business such as McDonald's, anything that helps cut down waste makes a big difference.

Finally you can see an impact on R&D, as menus start to be creatively shaped by customer insight and even new burger recipes get invented by data. As McDonald's CEO Steve Easterbrook pointed out ‘We’ve never had an issue in this business with a lack of data. It’s drawing the insight and the intelligence out of it’

Personalisation at a ‘restaurant-level’ clearly has many operational benefits. But what are the implications of personalisation at a ‘customer-level’. If customers are willing to identify themselves (give access to their mobiles, location and preference data) McDonald's can be more useful to an individual customer. Implementing Dynamic Yield’s engine inside the ‘Click & collect’ app could pull up your favourite / last ordered meals or even make new bundles of products just for you.

Self-serve kiosks could use beacon technology to identify a specific customer as they approach and adjust the digital menu to suit that person’s preferences or persuade them to try something new.

1. Capture better data at the moment of purchase

• Today, static tills and loyalty cards give you intermittent and unreliable data.
• Bringing the purchase moment onto a digital interface gives you a live view of purchase habits : what individuals are browsing, purchasing and when.
• Contextual data points can also enrich this purchase data : time of day / year, store busyness, weather, social listening, even who they’re shopping with at the time.

Suggested actions:

• Create a valuable digital interface that customers use to make purchases
• Collect additional contextual data points to enrich purchase data

2. Analyse and extract patterns from the data

• Similar to the analysis of online analytics, patterns can be recognised through in-store data giving you insight into purchase behaviour.
• Using this evidence, teams can start to form hypotheses about what customers want and are most likely to engage with.
• As a way to prove or disprove these hypotheses, lightweight experiments should be run inside the store to validate assumptions.

Suggested actions:

• Analyse patterns within the data to draw insights about purchase behaviour
• Run lightweight in-store experiments to test your hypotheses

3. Surface recommendations to customers in-store

• Simplistic hypotheses can start to be fed into a recommendation engine as rules - i.e if a user does this, offer them this.
• Over time, this engine can start to compute more complicated and more personalised suggestions and offer them to customers directly through the digital interfaces.
• Collecting better data and understanding better insights will ultimately lead to more relevant recommendations and even new store layouts.

Suggested actions:

• Build a recommendation engine that can be fed simple rules
• Design what, where and how you digitally surface recommendations to the user