Executive Summary
To provide an exceptional dining experience, cutting drive-thru wait times is a priority for Americana Restaurants, which operates KFC outlets across the United Arab Emirates (UAE). At the same time, the company wanted to increase the average order value and make it easy to support customers who speak Arabic, even though most staff are English-speaking.
The answer was to implement Sodaclick’s multilingual Conversational Voice AI order taker to allow customers to place their order at the speaker post. Rather than using scripted replies and questions, Sodaclick’s solution uses a large language model that understands context beyond the menu. It can, for example, recommend in-stock products when they ask for a competitor’s menu item such as a Pepsi when a customer asks for a Coca-Cola. Upsell recommendations are incorporated into the intuitive AI’s responses.
The software runs on AI PCs based on Intel® Core™ Ultra processors. These devices have a CPU, integrated GPU, and NPU (neural processing unit) on the processor die to efficiently handle AI workloads locally. Most customer requests are processed on the AI PC, a small form-factor device that easily fits within space-constrained retail environments. More complex queries go to the cloud, where they are processed using Intel® Xeon® Scalable processors.
Americana Restaurants has increased the average order value by 8.5 AED (about $2.20) using Sodaclick’s Conversational Voice solution running on AI PCs.
Challenge: Delivering Consistently Great Service at the Drive-Thru
Whether it’s for a quick coffee on the way to work or a late-night chicken burger after a movie, quick service restaurants are a part of daily life.
Restaurant operators across the globe face several challenges when it comes to delivering satisfying customer experiences. Wait times can be long, with an industry average of 5 minutes 29 seconds. When food delivery is slow, customers may be less inclined to return, especially if the delay means the food has gone cold. As many as 11 percent of orders are inaccurate with contributing factors including mishearing the order and language barriers.
The sector generally struggles to attract and retain staff, with some restaurants having to reduce opening hours or operating with smaller teams. As a result, service can be inconsistent, and restaurants miss out on upsell opportunities because staff don’t have time to interact fully with the customer.
Americana Restaurants is one of the largest restaurant operators in the Middle East, North Africa (MENA), and Kazakhstan region. For more than 50 years, Americana Restaurants has run outlets under global brands such as KFC, Pizza Hut, Hardee’s, Krispy Kreme, and TGI Fridays. The company operates 2,630 restaurants across 12 countries and has over 38,000 employees.
In the UAE, Americana Restaurants wanted to improve its drive-thru service at its KFC outlets. The goal was to cut wait times and enable more consistent upselling to increase average order value.
It was also important to enable multilingual customer service so that customers could enjoy the same high-quality service in their preferred language. While most of the restaurant staff speak English, many do not speak Arabic. Americana Restaurants wanted to easily serve customers who prefer to speak Arabic or English, including the multiple dialects within the languages.
Solution: Accelerating Orders with AI
Americana Restaurants implemented Sodaclick’s Conversational Voice AI automated order taker at the drive thru.
Using the technology is simple: Discuss your order at the speaker post. A customer could ask, for example, for a drink recommendation on a hot day or for something spicy from the menu just as they would do speaking to a member of staff. The solution integrates upsell suggestions into spoken responses and can handle customizations, such as a customer asking for no lettuce in a burger.
Because it’s integrated with the back-office systems, the software has the latest information on pricing and in-stock items. The customer’s order is shown on screen for confirmation before the order is finalized.
Interactions are dynamic and unscripted, with AI generating personalized responses for each customer. The goal is to make the AI feel as human-like as possible. The solution is based on generative AI and has been given context beyond the menu so it can help customers as much as possible. The automated conversational voice AI order taker can understand if customers attempt to order a rival chain’s menu item, turning it into an opportunity to recommend KFC’s alternative. The AI order taker can answer other helpful queries, outside of placing the order such as information on operating hours. Sodaclick, in collaboration with the restaurant operator, implements guardrails to ensure that the solution only answers the kinds of queries the operator wants it to. Any queries outside of those are answered with a request to get back to the ordering process, so that customers (and those queuing behind them) don’t waste time.
The restaurant operator has enabled four of the 96 languages Sodaclick supports, so customers can use their preferred language while interacting with the AI. The solution automatically detects language preferences and dialects without needing to prompt the AI to change language for seamless interactions.
The software runs on AI PCs based on Intel Core Ultra processors. These devices have a CPU, integrated GPU, and NPU (neural processing unit) on the processor die to efficiently handle AI workloads locally.
Sodaclick’s software uses the CPU and the integrated Intel® Arc™ GPU to process most queries on the device, reducing network bandwidth and the dependency on the cloud. The natural language processing of the customer’s speech is carried out on the CPU, and the integrated GPU is used for generative AI, creating the AI response. Using this architecture, there’s no need for a bulky, power-hungry discrete GPU. The small form factor of the AI PC helps retailers to introduce AI where there is limited physical space, without compromising performance.
While most customer requests are resolved locally, highly complex queries are sent to the cloud to be analyzed using Intel Xeon Scalable processors. Customers, for example, who seek additional nutrition information or wish to use multiple languages within a transaction, would have their query sent to the cloud for processing. These queries are encrypted and anonymized to protect customer privacy.
On the restaurant-side terminal, staff see the order details, including the customer’s full conversation with the AI software, so that they can prepare the order. Because the AI takes care of ordering, staff are free to focus on cooking and packing, and helping any customers who have additional service requirements.
Sodaclick says that its solution achieves high order accuracy, while helping to cut the customer’s wait.
Intel Helps to Improve Performance
Intel helped Sodaclick with software optimization, enhancing voice recognition accuracy and efficiency, even during peak times. The cooperation helped increase processing speeds and create a solution that scales across various order types and locations. The Intel® architecture enables language processing and order management to be carried out simultaneously on the device.
The Intel® Distribution of OpenVINO™ toolkit accelerated AI inferencing performance on Intel® processors.
“Using the Intel Core Ultra architecture allows us to create a solution that is fast, has enhanced security, and is scalable around the world,” said Ibrahim Jan, CEO of Sodaclick. “The key benefit of using Intel technology is that it enables fast processing speeds for smaller requests and supports real-time customer interaction. With Intel Core Ultra processors, we can run multilingual AI solutions locally.”
Technical Components of Solution
- Intel Core Ultra processor. Intel Core Ultra processors are high-efficiency processors built to deliver next-gen AI experiences in sleek and slim mobile form factors. These processors are the foundation of the AI PC, with a high-throughput GPU, low-power NPU, and fast-response CPU.
- Intel Xeon Scalable processor. Intel Xeon Scalable processors handle complex queries sent to the cloud. While the on-device AI can process any one language during an order, cloud processing can handle up to four languages during the same transaction, so the solution can support groups of people ordering together who have different first languages.
- Sodaclick Conversational Voice AI. The conversational voice AI enables customers to place their orders using natural language, handling any queries they have. The conversation includes upsell suggestions and can cover customer queries outside the ordering process, such as where facilities are located.
- Sephona. Sephona is Sodaclick’s cloud-based AI dashboard. It enables device management, including remote desktop and reboots, and AI analytics to enhance customer interactions, streamline operations, and improve decision-making. The AI analytics software processes customer interactions and external data, such as customer feedback and market trends, to help restaurant operators optimize their offerings.
Results
Introducing AI at the order point has enabled Americana Restaurants to serve customers quickly while increasing the average order size.
Sodaclick found that the Intel Core Ultra processors enable a fast time to first token, reducing wait times and improving customer satisfaction.
The AI software adds tailored upsell suggestions to 86 percent of orders, with acceptance rates reaching 75 percent in some periods. As a result, the average order value has increased by 8.5 AED (about $2.20).
“Sodaclick is all about innovation, using the latest cutting-edge technology,” said Jan. “We always test the latest AI models and look to incorporate them into our deployments because AI is becoming more intelligent as the weeks go by.”
Sodaclick has a close partnership with Intel across its business. “We can support businesses from small franchisees that have one to five stores, all the way up to the large ones that have thousands of stores, and they’re all powered exclusively by Intel technology,” said Jan. “Using Intel Core Ultra architecture to run a hybrid model helps us to ensure that everything’s fast, has enhanced security, and is scalable across the world.”