Drive-Thru Technology: From Digital Menus to Artificial Intelligence
Why drive-thrus need AI:
Figure 1 – Three Charts to Show Why QSRs Need AI in their Drive-Thrus
Drive-Thrus Represent Huge Potential for AI
According to Merriam-Webster, artificial intelligence (AI) is a branch of computer science that simulates intelligent behavior in computers, allowing machines to perform complex tasks previously reserved for humans. Conversational AI is a type of artificial intelligence that uses text and speech to automate communication and build personal customer connections on a large scale. The market for AI in drive-thrus is growing as more restaurant owners and operators look for ways to compete in an increasingly crowded space.
There are 660,755 locations of quick-service restaurants (QSRs) and full-service restaurants (FSRs) in the US, and total sales are expected to grow from $825 billion in 2018 to $863 billion by the end of 2019. Off-premises sales, which include delivery, takeout, drive-thru, and catering, represented 38% of total sales in 2018 and are projected to grow by 5.6% in 2019.
QSRs make up 346,105 of this total and had 2018 sales of $256 billion, 70% of which came from drive-thru alone. By the end of 2019, QSRs are expected to grow by 4.9%, which is slower than the 5.1% growth the previous year but still faster than the 3.4% growth of FSRs for the same time period.
As off-premises ordering increases, drive-thru restaurants will need to find ways to maintain food quality, improve order accuracy and increase speed of service. Quick service and fast casual chains that do not have a robust technology deployment strategy run the risk of damaging their reputations. As more QSR operators balance dine-in orders with the multitude of ways consumers can now make a take-out order, they risk overlooking drive-thrus as a way to bridge the gap. Some chains, like Chipotle, are rethinking drive-thrus as mobile order pickup stations, while others, like McDonalds, are implementing conversational artificial intelligence to increase check averages.
Challenges with Drive-Thrus
According to Vinay Shukla and Rahul Aggarwal, founders of artificial intelligence startup ConverseNow, the order-to-service time is too long for drive-thru customers, leading to more lines and an inconsistent experience. With their company, Vinay and Rahul aim to improve the drive-thru experience from both inside and outside the window.
From the outside-window perspective, the number of cars at the drive-thru are increasing and menus are becoming more complex, resulting in inaccurate order taking and longer wait times. QSR Magazine’s 2019 Drive-Thru Performance Study shows that order accuracy dropped to an average of 84.4%. As Figure 1 Chart 3 shows on page 1, it was consistently over 88% in the three years prior. Looking at Figure 1 Chart 2, wait times in 2019 were 11.8% higher than the average for 2016 to 2018.
From the inside-window perspective, Vinay and Rahul say that employees’ multitasking leads to increased fatigue and incorrect order entry, reducing the chance they will remember to upsell customers as they struggle to keep up with an influx of mobile, store and drive-thru orders.
These major challenges amount to a $26 billion loss in the US alone. Also, the demanding work environment within fast food restaurants is resulting in more workers quitting than ever before. Turnover is currently 130%, which increases hiring and training costs to over $2,800 for a 3-month period.
Speed of service is the most critical metric when it comes to the drive-thru. The average order time was 9% higher for 2019 over the previous year, with an average wait time of 255.34 seconds, but there is a huge variation across the industry. Dunkin’ was the fastest in 2019, taking 216.75 seconds to complete an order, and Chick-fil-A was the slowest at 322.98 seconds. Surprisingly, Chick-fil-A was rated as the friendliest drive-thru, showing that many customers do not mind waiting in line for their favorite food, but they do mind waiting if the overall service quality, including the way drive-thru employees treat them, is poor. To improve their speed of service, Chick-fil-A employees are now using tablets to take orders outside while customers wait in line at the drive-thru. Digital employees powered by AI will further improve the order-taking process, since every QSR can now reduce the wait time for ordering and deliver better customer experience.
How Will AI Work with Existing Technology?
Consumers have been able to order from a multitude of brands since food aggregators like Grubhub entered the scene, representing a major leap forward for the restaurant industry. As consumer loyalty shifted away from individual brands, however, restaurants shared more of their profits with third-party service providers.
AI allows restaurants to retake control of the customer experience and earn customers’ loyalty by automating the ordering process through traditional channels like phone, drive-thru, mobile, and online rather than outsourcing the process to an app. Moreover, advancements in conversational AI and speech recognition technology have made it possible to process complex, messy and unconstrained language used by customers while placing orders. Consumers of the future will prefer voice-based ordering rather than using touchscreens or waiting in line to speak with a live person. As the voice assistant ecosystem matures, new voice channels are emerging that restaurants will use to promote their brands and take control of the overall customer experience.
The ConverseNow Voice AI platform automates restaurant orders on high-volume voice channels, such as phone calls, drive-thru stations, self-service kiosks, and voice-assisted mobile chat. By integrating with Brink POS, ConverseNow’s software can process all requisite details—like menu items, modifiers, past ordering history, and prices—from the point of sale terminal and push all orders directly into the POS system after processing. This seamless integration ensures that all orders are captured and priced accurately based on the availability of items and supported modifiers.
ConverseNow improves the customer experience and increases operational efficiency by re-imagining the ordering process. As an off-premises restaurant partner and intelligent food ordering expert, the company provides a consistent ordering experience across all restaurant voice channels.
According to ConverseNow, manual ordering and POS system entry cause restaurants to miss out on the latest innovations many of their competitors are implementing. For restaurants, ConverseNow’s integrated voice AI platform improves top-line performance by increasing upsells, eliminating the chance of forgetting to recommend an add-on. The integrated platform also improves bottom-line performance by reducing the percentage of inaccurate orders.
Conversational Technologies Are Going Mainstream
In order to meet growing customer expectations, some of the major QSR chains have already started adopting conversational technologies and training their “digital employees” to handle complex menu combinations, modifiers and conversational user experience (CUX) expectations of their consumers in order to be ready for large scale deployments. CUX focuses on the quality of a conversation, including the helpfulness and personalization of a computer’s responses to human speech. It involves the tone of computer responses, which can be a measure of sarcasm, wit, or some other qualitative assessment of a conversation.
Conversational technologies include chat, automatic speech recognition (ASR) and natural language processing (NLP). For chat, two-way communication can occur between humans and chatbots that fit into one of three categories. Rule-based chatbots allow customers to click on buttons to answer questions, which the software processes to determine the best answers. Intellectually independent chatbots reply to certain words or phrases they are programmed to recognize with predetermined answers, becoming more adept at identifying word combinations through machine learning. AI-powered chatbots can respond to random questions, improving their abilities through machine learning, artificial intelligence and natural language processing.
ASR involves software that converts human speech into a series of words, while NLP software converts human speech into data that can be interpreted to deliver a human-like response. Natural language understanding (NLU) is a part of this process, and it is the way in which a voice-based assistant uses deduction to decipher the meaning of an expression. It is the reason Amazon’s Alexa knows enough to provide the same answer to different questions like “How is the weather today?” and “What is it like outside?”
Stages of Conversational AI
For a look at the state of artificial intelligence today, consider the example of Tesla when it was a fledgling company. The first Tesla simply had auto braking and lane assist features. As the company learned from customer interactions on the field, it added more autonomous functionalities, eventually creating the Tesla Full Self-Driving computer (FSD).
In a similar manner, “digital employees” for restaurants must be trained in a live environment with real conversations before they are able to handle orders with full autonomy.
ConverseNow’s suite of solutions retrofits into restaurants’ existing drive-thru hardware without forcing them to upgrade their hardware for using AI. It allows restaurants to take an organic approach to automation while ensuring a seamless ordering experience for their customers. Its technology acts as a conduit between employees and customers, focusing on innovating drive-thrus in three ways: EmployeeAssist, OrderAssist and Ordering Intelligence.
Employee Assist TM – AI as a human assist
Employees take orders while AI learns, resulting in increased order accuracy, employee efficiency, upsell callouts and reduced training costs.
EmployeeAssist learns from live conversations between employees and customers and is released with the POS system to improve employees’ performance from inside the drive-thru window, focusing on order accuracy and providing contextual cues for operators to recommend items that have a high probability of getting accepted. When customers recite their orders, EmployeeAssist AI listens into the audio stream and prevents order entry errors by providing visual cues for drive-thru workers. Employees still take the orders, but EmployeeAssist prompts them to make personalized menu recommendations based on its digital order memory, reducing training costs and boosting efficiency.
This solution also does real-time sentiment analysis to observe the mood of both customers and operators, which can help operators to make a corrective course of action if needed. For instance, if customers deliver a curt response, the employee can be more straight-to-the-point in order to avoid upsetting them further. This technology is also future proof, training itself to get better at working with order confirmation and dynamic digital menuboards.
Order Assist TM – Humans as an AI assist
Voice AI (the “digital employee”) takes orders while human employees handle escalations. Autonomous voice AI ordering software is trained on extensive menu and food data to provide an Amazon-like personalized ordering experience and upsell consistently.
OrderAssist can be deployed on existing drive-thru hardware to drive a fully autonomous ordering experience while involving live agents to handle any exceptions, leading to more qualified callouts and conversions. A digital voice greets customers and responds to their queries to take the order, while employees handle customer complaints and any other scenario with the potential to escalate. Machine autonomy at this level allows for dynamic suggestive selling with 100% callout, leading to more sales at a higher ticket price. This technology is essentially a digital employee because it can have conversations with customers, further reducing service time, increasing order accuracy and making employee turnover a thing of the past.
Ordering Intelligence increases operational efficiency and improves the bottom line of your restaurant business. It captures data from the same high-volume voice channels your FSR or QSR is already using and converts it into five different types of abstracted machine intelligence:
Menu metadata intelligence is based on data from orders captured via AI. It provides valuable information to restaurants so they can optimize their menus based on what is selling and any relevant modifiers the system supports.
Order intelligence uses data captured from multiple sources, such as dish combinations and modifiers, the number of orders, order value, and customer preferences in the context of the time of day, weather, traffic conditions, and other factors.
Upsell intelligence considers the offer rate, acceptance rate, context, and customer preferences to offer appropriate upsell recommendations.
Process intelligence takes what the AI system has learned from individuals’ ordering behavior and their response to AI ordering flows in order to optimize the customer experience.
Customer experience intelligence analyzes sentiment associated with the context of an order.
How AI Will Continue to Improve
Artificial intelligence works with your existing drive-thru technology by constantly listening, learning and improving to convert the ordering process into a continuously evolving algorithm.
If you’ve ever heard the phrase “you are what you eat,” then you are familiar with Consumer Food Identity (CFI) – a concept relating to the habits and preferences of individual consumers when it comes to food. To build its data foundation, AI constructs CFI profiles that will allow it to evolve from EmployeeAssist to OrderAssist through frequently sampling, learning and updating the characteristics of food orders and associating them with different consumers.
Backend testing involves simulating the end-to-end ordering process in a controlled environment. Before being able to implement EmployeeAssist, the AI system needs to become well-versed in your menu modifiers and types of upsell recommendations. After configuration, training, deployment, and field trials, the AI analyzes key performance indicators (KPIs) to guide its data analysis and training efforts.
Using OrderAssist, the eventual goal is to achieve a fully autonomous voice ordering experience so AI can close orders directly by engaging with customers in a human-like manner. Vinay and Rahul also believe that such technology needs to be introduced organically, where it can be deployed to work along with humans and trained before it starts interacting with end-users.
How An AI-Led Drive-Thru Experience Looks
By suggesting relevant products based on what other consumers have ordered, AI can drastically increase overall check averages. ConverseNow gives one example of a theoretical consumer ordering a medium pepperoni and sausage pizza and 20-ounce Coke. Here is how the dialogue plays out to increase check size:
A digital voice coming through the drive-thru speaker says, “Adding one order…” and then repeats the order to the consumer to ensure accuracy.
Digital Voice: “Would you like me to add anything else?”
Consumer: “No, thank you.”
Digital Voice: Recites the order total and says “Okay, people go for parmesan bread bites and cinnamon bread twists with this. You want to include that?”
Consumer: “Yes, please.”
Digital Voice: “Got it. I added one order of cinnamon bread twists to your order. What size would you like for parmesan bread bites?”
Consumer: “16 piece”
Digital Voice: “Got it. I added one order of 16 piece parmesan bread bites to your order” and then says the total price of the meal.
Medium hand-made pan pizza with pepperoni and Italian sausage
20 oz bottle of Coke
Order after dynamic suggestive selling
Medium hand-made pan pizza with pepperoni and Italian sausage
20 oz bottle of Coke
16 piece parmesan bread bites
Cinnamon bread twists
In this scenario, the sale price increased by 84% – even when the consumer initially did not want any further menu items.
Apprente’s Drive-Thru Voice Automation Rollout
In June 2017, Itamar Arel and Moshe Looks started Apprente to improve machine learning. By January 2018, the company secured $4.75 million in funding and started testing its technology in a few McDonald’s locations. McDonald’s acquired Apprente in September 2019 and made Arel the vice president of McD Technology Labs and Looks the senior director. By headquartering its new development group in Mountain View, California, McDonald’s is signaling to the world that it is serious about developing innovations in artificial intelligence.
McDonald’s knew it had to respond to the changing world of the drive-thru, especially after its speed of service went from 188.83 seconds in 2012 to 273.29 seconds in 2018. While other QSR brands have 6 or more automobiles in their drive-thrus less than 5% of the time, this occurs 11.9% of the time for McDonald’s, making it even more necessary for the QSR to innovate.
Technical Challenges of AI
According to a patent filed by Apprente, artificial intelligence systems use deep gradient descent and back propagation to minimize errors. If an AI system observes a different response than it had in its training, it can run the new input through the network of inputs and outputs it established during training. This, however, requires many hours of investment to build a training dataset large enough to deliver meaningful results – underscoring ConverseNow’s commitment to gradual learning.
Different Applications of AI Make Their Way to Fast- Casual and Quick Service Chains
The rapid growth of digital ordering has encouraged many restaurants to deploy their own variations of drive-thrus to complement their other technological developments.
Digital sales have been growing at a brisk pace for Chipotle, increasing by 100.7% and 99.1%, respectively, in Q1 and Q2 2019. In early 2018, a year before Chipotle announced the construction of Chipotlanes, the company’s own version of a drive-thru that functions as a mobile order pickup lane, it started testing artificial intelligence technology for phone ordering systems in 10 locations and expanded to 1800 stores by the middle of 2019.
Unlike traditional phone orders, this conversational AI allows customers to pay ahead and suggests toppings they may be forgetting, learning the differences in the way people order to become more intuitive over time. Although Chipotlanes focus on using the Chipotle app to order, AI-based phone ordering technology has a lot of potential for improving drive-thru speed of service because it can correct errors before consumers even pull up to a Chipotlane. Nevertheless, Chipotle is already seeing impressive order fulfillment times thanks to the ability of its app to accept customization and payments as well. According to Chipotle’s CEO Brian Niccol, consumers only have to wait 12 seconds to get their order from a Chipotlane—which is almost 20 times faster than even the quickest QSR drive-thrus.
To maintain this speed of service, Chipotle made improvements inside its restaurants so employees have separate areas to focus on mobile orders for Chipotlanes. For instance, Chipotle added mobile pick-up shelves and “second makelines” to accommodate the 87.9% year over year increase in digital sales it saw in the third quarter of 2019. Adding a new workflow to their back of house operations helps to separate in-store from off-premises orders so an influx of mobile orders will not hold up customers waiting in the traditional line. Employees can see customizations from a large screen as they complete the assembly-line style process for digital orders, which grew from 11.2% to 18.3% of sales from Q3 2018 to Q3 2019.
Domino’s is also experimenting with drive-thrus, albeit on a smaller scale. A year after the pizza chain started renovating its stores in 2012, it built its first Pizza Theater restaurant in Louisiana, which has a drive-thru for pickup orders. Although Domino’s does not mention its plan to expand drive-thrus to more restaurants on its website, some franchise owners are expressing interest in the concept. One Domino’s location in Buffalo, New York, built a drive-thru without speakers or order screens for picking up phone and online orders in 2018, and other locations in Illinois and Pennsylvania also have their own drive-thrus.
These restaurants have the potential to test how the drive-thru will affect the way consumers use Domino’s numerous other technologies, given that the company has been consistently innovating its ordering process for over a decade. Since starting its Pizza Tracker in 2008, Domino’s market share doubled to 18%. In June 2014, Domino’s partnered with Nuance Communications to launch a voice-based ordering assistant called DOM. Customers using Domino’s iPhone or Android app can use DOM to submit an Easy Order or create a completely new order, and it suggests add-ons and helps them find coupons. DOM analyzes phone numbers to determine if an incoming call is a new order or a request for a status update. In October 2018, Domino’s started using Twilio to take a more proactive approach to voice ordering, using Amazon Alexa and text-based push notifications to update customers on their orders. Using Twilio’s Pay technology, customers can enter their credit card numbers through the phone to avoid saying this private information aloud. Twilio’s Autopilot technology allows Domino’s to send geo-targeted text messages once customers complete an order. One user in the Sydney test market, for instance, received a text during the summer asking if it was too hot to cook along with information on various deals.
Domino’s is also using AI, machine learning and sensors to improve the quality of its products with the DOM Pizza Checker. After launching in Australia and New Zealand in mid-2019, DOM improved consumers’ quality ratings by 15%. According to Domino’s, the DOM Pizza Checker is a smart scanner located above the cutting station that grades pizzas based on different criteria, including pizza type, topping accuracy and how evenly the cheese is distributed, and takes a picture for customers to see on the Pizza Tracker Page as they wait for their order. Pictures of these pizzas are stored in a database that will eventually teach DOM how a pizza is supposed to look. Domino’s has been working with Dragontail Systems to develop the technology for the past 2 years. If the wrong topping ends up on a pizza, DOM will automatically alert customers that their pizza is being remade.
Many other QSR brands are leveraging AI in order to gain market share. DD Perks, Dunkin’s rewards program for mobile app users, tracks the type of menu items customers order, as well as the frequency and time of each purchase. This allows the doughnut and coffee chain to send coupons to customers that ordered a certain item frequently in the past and then stopped suddenly. Combined with Alexa and Google Home integrations, this technology helps to spur repeat purchases.
By April 2019, McDonald’s had acquired Dynamic Yield for $300 million, representing a leap forward for AI in the drive-thru with digital menuboards that personalize suggestions based on the time of day and the type of items in the order.
Chatbots and AI
Voice ordering through the Domino’s app is just one of 11 different options on Domino’s Anyware platform. Customers can also order through Slack using chat-based AI, and even collaborate on making a pizza if they include “@Domino’s” in their messages. The same goes for Facebook Messenger, where typing “Hello” into a new message to the company allows customers to start a new order or track their current one. Other “Anyware” technologies like text, Google Home, Alexa, Twitter, and Zero Clicks focus on making it as easy as possible for customers to place their Easy Orders.
TacoBot, an AI-based chat technology integrated with Slack, has been in beta testing at Taco Bell since 2016. It responds to conversational text like “Can I get a soft taco?” by clarifying the order and asking if customers want to see a list of add-ons. It even suggests popular add-ons and delivers witty responses to funny statements. To place the order, all customers have to do is type “checkout.”
Challenges with Fast-Casual Restaurants and Drive-Thrus
As more fast-casual restaurants build drive-thrus in response to the off-premises ordering trend, the market potential for conversational AI technologies will increase. Since fast-casual restaurants tend to have more complex menus at a higher selling price than traditional QSRs, building a drive-thru has its own unique set of challenges that AI can address.
Pie Five, a maker of handcrafted pizzas, is building drive-thrus for its new stores to target large families on their way home from work. The company plans to keep speed of service under 360 seconds, which is still 37 seconds longer than the slowest drive-thru in the 2019 QSR Magazine Drive-Thru Performance Study.
For the drive-thru, Pie Five simplified its outdoor menuboard, removing some signature pizzas in the hope that consumers will either choose from a smaller list or remember their previous in-store orders. The fast-casual chain’s static menus make personalizing orders and maintaining a relatively fast speed of service even more difficult, but digital menuboards with AI-based suggestions could address these problems in the future. With digital displays, preserving menu real estate would no longer be necessary. If a drive-thru customer wants a signature pizza typically reserved for in-store orders, the menu screen can change to display each ingredient and then transition back to the main screen for the next customer. To increase the speed of service, AI can distinguish between customers who have ordered this item in the past and those who are trying it for the first time. For repeat customers, a conversational AI system could ask if they want to apply past customizations to the current order to simplify the decision-making process. Any prior customization would then appear on the menu screen, giving customers the ability to verbally edit their orders. For new customers, a friendly AI-based voice could work in conjunction with a digital menuboard and say something like “This order has all of the ingredients shown on the screen. Would you like to change anything?” and even suggest popular ways to customize the order.
When Chipotle launched its own drive-thru concept, Chipotlanes, with a plan to build 60 by the end of 2019, it addressed the problems shown in the Pie Five example by only allowing orders through its mobile app. This establishes the drive-thru as a mobile order pickup lane without the menuboards and speakers found in most drive-thrus.