In the age of Amazon, differentiation on quality and service isn’t enough for retailers looking to not only survive but thrive. In order to compete in a narrowing retail landscape, retailers need to start competing with Amazon and other ecommerce giants on convenience as well.
This is no easy lift, as Amazon’s business model is centered on the fact that consumers have access to nearly any product at nearly any time. And with Amazon boasting 5 billion items delivered to Prime customers in 2017, this convenience is fast becoming the expectation for consumers. But using Location Intelligence, retailers can start to compete in delivery, and even offer premium services like set delivery windows, without breaking the bank.
To offer a premium service, in this case, delivery within a designated 2-hour window, that remains profitable for retailers of any size.
To build out our delivery services, we selected 9 store locations (we used Gap retail locations as a placeholder) in the Atlanta area from which we intend to deliver to local consumers.
The first step to create a manageable and profitable delivery network is to create distinct territories for each of our 9 locations. This territory is derived from 1.5 million points of historical delivery data. Instead of trying to map each historical delivery point to a store, we used CARTO’s intersect and aggregate functions to create distinct hexbin districts.
Next, our team sought to assign each hexbin to its proper store location algorithmically. They used CARTOframes, a Python package that allows developers and data scientists to interact directly with CARTO within their workflows and preferred tools, like Jupyter notebooks.
Using an Origin Destination Matrix (ODM), territories with the shortest overall drive time are assigned to the nearest unassigned hexbin, thus increasing that territory’s drive time. This technique is useful in providing baseline territories, but as you can see below, our map is a bit messy.
In order to clean up our territory map, the team, continuing to use CARTOframes, identified all of the ‘islands,’ or hexbins that are physically separate from their home territory cluster (or mainland), and worked on re-assigning them.
If an island was touching one mainland (see the two red districts at the top that are touching the green mainland), it was immediately assigned to that mainland. If the island was touching two mainlands, it was assigned to the touching mainland with the lowest overall drive time. This is done to keep the territory drive times balanced, and again it is all done within a Jupyter notebook.
Want to bring spatial analysis into your data science workflow?Learn More about CARTOframes!
As a result of this 2nd step, our team was able to build an efficient territory map within their data science workflow, that can act as the framework for our optimized delivery services.
Now that we have our territory map completed, we can plan out the day’s routes to match our consumer selected delivery windows. We can start by uploading the scheduled deliveries and use the intersect function to assign each one to the store location whose territory they fall into.
Coming up with optimized routes requires us to solve what is commonly referred to as a Vehicle Routing Problem, or VRP. This problem is made more complex by the constraints that we are imposing on ourselves. In this case, our primary constraint is the premium service we are offering in the form of delivery windows, so our routes not only need to be optimized to reduce our costs but must allow for accurate and timely delivery.
Using Google’s Optimization tools, another Origin Destination Matrix provides us with routes that are now fully optimized and meet the constraints that we set up in our VRP. Keep in mind that depending on the variables and constraints, these tools can take some time to run. Once we receive our routes, we can load them into CARTO to get a full picture of our day’s deliveries and the total drive time across all of our routes.
Widgets in the map below allow us to see the drive time per store location as well as per two-hour window (Noon, 2, 4, or 6).
Using the optimized routes and drive times we can get a picture of our total costs for each delivery window, based on labor and gas costs. When compared with the total value of the window’s delivery, we have a clear sense of the profitability of our newly optimized, premium delivery service.
If we find that our profits are not justifying costs, we can tweak the system to boost value. By creating a minimum total cart price for delivery, for example, we may drive down the cost incurred by having too many orders of too little value. A Location Intelligent solution allows retailers to stay flexible and adjust parameters.
Small and medium-sized retailers often have an advantage over ecommerce giants, in the form of local brand loyalty and recognition, but that advantage doesn’t hold up in the face of undeniable convenience. By using Location Intelligence solutions to optimize operations, retailers can begin to compete on convenience with premium offerings like delivery windows without breaking the bank. These solutions also drive a deeper understanding of purchasing patterns and enable changes to maximize profitability in near real-time.
If a retailer isn’t offering services like delivery, Location Intelligence solutions put them well within reach, and if they are, a Location Intelligent solution can take those services to a new level of efficiency and profitability.
Today at the Spatial Data Science Conference in New York City we announced CARTO Workflows, an exciting new tool that enables users of all types and skill levels to harness...Use Cases
Please fill out the below form and we'll be in touch real soon.