How implementing an AI restaurant manager boosted service speed, cut waste, and drove sales
PreciTaste embedded a suite of sensors in a QSR’s kitchen, lobby, and drive-thru. By measuring the number of customers entering the store and drive-thru, PreciTaste’s AI Restaurant Manager predicted the number of prepared burgers, chicken, and fries required to meet demand and sent cooking instructions to the kitchen staff via touchscreen monitors mounted near the cook stations. Doing so saw large improvements in labor efficiency, significant reductions in customer wait time, and completely eliminated product stock-outs.
The customer was a large QSR, with already highly process-optimized locations worldwide, serving burgers, fries, chicken and breakfast foods. This case study will focus on a single restaurant serving 2,500 customers daily in a 24/7 environment.
Like most QSRs, the customer struggled with slow service times in the drive-thru and lobby, especially during times of peak demand. For this location, the most frequently cited complaint on Google reviews was the amount of time guests spent waiting for their food. In order to compensate for this, the kitchen crew often overproduced burgers and fries during rush periods but saw significant waste after demand dropped off.
This location experienced a high rate of employee turnover and, consequently, management spent a lot of time and energy training new employees. In turn, the high level of inexperienced employees led to frequent production mistakes. New fry cooks struggled with accurately gauging the correct amount of fries to produce to meet demand, leading to stockouts (no available fries) or waste from overproduced fries.
PreciTaste deployed a server and sensor suite in the kitchen in a matter of hours overnight without impacting kitchen operations. Once installed, all training for the AI agents took place within the store without an external connection. The team utilized existing security cameras inside the restaurant lobby and outside the store to rapidly establish customer detection without needing to install any additional sensors.
In order to better predict the demand at this location, the PreciTaste AI agents processed over one year of historical POS data from which a real-time customer demand prediction model was trained. The Restaurant Brain was integrated with the existing POS systems so that the model could be continuously improved in real time using live POS data from the store.
The high turnover of employees presented a perfect opportunity to showcase PreciTaste’s ability to coordinate kitchen staff and provide simple, easy to follow cooking instructions through Crew Interaction Management screens. As orders from the POS or predictions from the Restaurant Brain were generated, a display in front of each cook station gave clear instructions to employees for how much food to prepare to meet the demand in the lobby and drive-thru with a simple count of what food items to cook and how many. Instead of having to parse through every order on the KVS, PreciTaste’s system reduced the cognitive load on the kitchen staff and filtered out only the relevant information they needed.
Coordinating the crew through cook command screens allowed for the implementation of intelligent batching. Instead of frequently cooking small batches of burgers and fries as the orders come in, the AI enabled the kitchen staff to cook larger batches less frequently, freeing up their time and allowing them to perform additional tasks like preparing salads and bringing food out to guests.
After the initial deployment, the AI-powered predictions improved the speed of service by 21% during rush hours (Fig. 1). The AI predictions tracked incredibly close to actual orders received at the restaurant (Fig. 2).
Fig. 1: Comparison of service times before and after installation
Fig. 2: Comparison of hourly AI Predictions vs actual hourly sales for April 2019
New hires in the kitchen were trained to the cook command screens. Managers reported that training was significantly easier with the screens taking over the bulk of the decision making and the new hires reported feeling confident behind the grill and fryer.
After deployment in the restaurant, efficiency in the kitchen rose by 53% (Fig. 3). Batch sizes increased over time as the batching algorithm self optimized.
Fig. 3: Average daily batch sizes for May 2019
The AI-managed restaurant is the optimal way to predict demand at a restaurant. The system continuously learns at each location, optimizing its algorithm from the moment it’s installed. Crew Interaction Management eliminates the learning curve for new hires and drives kitchen efficiency. As this customer saw firsthand, the PreciTaste Restaurant Management suite is revolutionizing the way QSRs manage their kitchens.