Intelligent Order Sequencing: How does it work?
In the past year, the outbreak of COVID-19 has pushed many restaurants, fast food chains, and cafés worldwide to adapt to new ordering channels. These channels extend beyond traditional in-store purchases and drive-thru orders to include mobile ordering. The advent of mobile ordering has added complexity to order production, with orders coming in from multiple new channels. Examples include 1st party mobile app orders, 3rd party mobile app orders (such as Grubhub, UberEats, Instacart, etc.), and website orders. The added complexity from these channels has made it difficult to ensure product quality and customer prioritization, plus introduced scalability challenges.
Our client, a large, international coffeehouse chain, was facing challenges brought on by the rapidly changing consumer buying patterns, including:
- How to best optimize customer order prioritization between mobile-app, in-store, and drive-through channels.
- How to improve sales and customer satisfaction.
- How to time order production based on various factors such as the customer’s estimated arrival time, any orders already in the queue, and barista workflow.
The Intelligent Order Sequencing solution from Neal Analytics addressed the customer’s concerns by creating a custom machine learning algorithm that uses real-time data to determine an order’s priority. This algorithm helped the customer maximize their production efficiency by improving throughput during peak hours and capturing more demand. The customer noted the solution offered substantial improvements, even during the solution’s trial process. The improvements included a shorter average customer wait time, fewer orders awaiting pickup, and a sharp reduction in waste. It is forecasted to positively impact revenues by over $200 million in >10,000 locations.
How it works
Production sequence optimization is like peeling an onion; what seems like a simple problem of improving on “First Come, First Serve” is much more complex than merely matching estimated arrivals with estimated production times.
Our consultants start with engaging with various teams to:
- Fully scope out current data’s unique dynamics and priority business considerations
- Evaluate key factors and possible targets for optimization
Neal Analytics tests hundreds of algorithm configurations against a digital twin, live benchmarking, or both in a lab setting. The tests are conducted several times to observe each KPI’s movement, and the results are reviewed with the customer’s stakeholders to find the optimum balance that meets their criteria. Reviews include visual representations of the sequence output to build a solid story and a clear understanding of the options, each with its benefits and drawbacks.
Once we understand how our target outcomes react to each algorithm, we can select the suitable algorithm for each store/moment. The results guide optimization development by providing data for the reinforcement learning algorithm to consider and leverage in training.
Deploy & Evaluate
The final production algorithm is designed to handle edge cases and can always fall back to traditional “First Come, First Served” timestamp logic if need be. The code runs locally in stores, and while the algorithm does receive various data from various web APIs, the trained model should not be dependent on a network connection. Therefore, it needs to fall back to a simpler First Come, First Serve logic in the event of certain edge use cases or potential errors. This fallback capability makes the solution robust enough to support the real-world traffic regardless of whether an organization has just one, hundreds, or even thousands of open locations every day.
Learn more about Intelligent Order Sequencing
If you’d like to learn more about Intelligent Order Sequencing, please contact us.