Many times, while ordering food, one parameter a customer takes into account is the estimated time of delivery (ETA in food delivery jargon). At a rudimentary level, restaurants used to estimate this depending on the type of food and where it has to be delivered. But these were times when reducing delivery times was not really a concern.
In this modern, post-pandemic world, delivering food in time is of criticality and a few extra seconds can be a deal-breaker in some cases. At LogiNext, we’ve been working with several top QSR (Quick Service Restaurant) chains like McDonald’s, BurgerKing, Starbucks and KFC to ensure under 30-minute deliveries. And the envelope is being pushed to under 15-minute deliveries.
In such a scenario, it becomes extremely important to predict the ETA before a customer places an order so that the entire supply chain is orchestrated in an efficient manner. And this is where Artificial Intelligence plays a critical role. Predictive ETAs can be calculated using two different modes- one is dependent on the delivery driver and the other is dependent on the best estimate.
Best Delivery Driver Mode
The objective of this mode is to identify the best delivery driver to whom the order can be assigned as per the specified profile and predict ETA considering the delivery associate’s details. The following parameters are considered here: Average Food Preparation Time, Incremental Preparation time (time taken to prepare each subsequent quantity of an item), load multiplier (includes peak hours), Pickup time, transit time between the two locations considering traffic patterns and even more granular details.
Best Estimate Mode
The objective of this mode is to calculate the best estimate of the Pre-Order ETA based on factors such as the availability of delivery associates, other orders waiting to be assigned at the branch, and so on. This is a more generic way of estimation and takes into account more parameters. The earlier one is laser-focused on optimizing for delivery driver management.
A generic formula for predicting ETA for food delivery would look like this:
Predicted ETA= Max {Pickup Time, (Order Preparation Time*Peak Hour Multiplier)} + (Default Service Time Per Order*X) + (Delivery On Road Transit Time)
The Giant Stride
Calculating ETAs is a massively complex problem owing to all the possible on the ground situations that can arise. One of the major strides when it comes to predicting ETAs is route planning and route optimization. There are several planning properties that contribute majorly to making the above a reality:
- Planning Objective: When planning the orders, one would want to achieve a goal rather than just plan orders. The goal can be to optimize overall trip cost or find the best possible route for fleet or to create trips which distributes the amount of work among the selected fleet.
- Route Constraints: To achieve the goal, one may want to modify the various route constraints that can be applied to optimize the trips.
- Fleet Constraints: Along with the route constraints, one can also apply fleet constraints which are applied on the selected fleet to optimize the trips.
- Advanced: There are certain other operational constraints that one may want to apply and a configurable system allows for this.
Role of AI
At the base of all these planning properties and delivery modes is the power and beauty of Artificial Intelligence and Machine Learning. Tracking billions of location data points, all the parameters can be predicted with greater accuracy and this makes the end customer experience ever richer. From an engineering perspective, this is an amazing problem where a data scientist gets to push the limits of how much can a machine predict the future? The more variables we can account for, the higher will be the precision and accuracy. And this is why AI plays such a big role in the future of supply chain and logistics management over the globe.
This article is authored by Pooja Patel, Vice President, Product Delivery, LogiNext
