60 Minutes recently aired an episode that showed aerial drones from Amazon delivering goods right to the front porch of a customer. Jeff Bezos, the CEO of Amazon was touting their continued experimentation in the path to achieving same day deliveries.
Same day delivery by drones is ridiculous. It is not just that the FAA is behind schedule on regulating drone flight paths, and getting to real regulations is likely years away. It is also, as a Wired article pointed out, the Amazon DCs are too far away from metropolitan areas to make these deliveries feasible. The hub and spoke logistics model that Amazon practices just will not lend itself to same day deliveries.
So if this path toward quick home deliveries is nonsense, what is feasible? This is something that Logistics Viewpoints has written several articles on (Best Practices in Omni-channel Home Deliveries; Omni-channel Last Mile Logistics; An Amazon Study Shows Why Same-Day Deliveries Matter).
I was already planning a new article on this topic. The 60 Minutes piece pushed this article to the front of the queue.
I had an interesting conversation with Victor Allis, the CEO of Quintiq, on this topic. Quintiq is a supply chain optimization software company and Victor is one of the brightest guys you could talk to in this industry.
Victor argues that to enable quick home deliveries we need to start forecasting which consumers will be ordering, what the volumes will be on given days, and where those consumers live; based on the forecast, a routing optimization engine then prebuilds routes for the next day that include inventory locations (stores) and homes. In this neighborhood of the city, for example, the forecast might say we will get 2 orders on Wednesday. The routing engine is prepopulated with the addresses of people in that neighborhood that have ordered in the past. Then as real orders come in for that neighborhood, the placeholder addresses are swapped out for real addresses based on real orders in the prebuilt route.
The forecasted orders will not, of course, be exactly in the quantities or neighborhoods forecasted. But that is the beauty of real-time routing solutions. There is an instance of the routing engine that contains the stops and routes, and there is also an offline routing engine that continues to look at the routes and ask itself “if we swapped this stop on this route with that stop on the other route, could we save money?” In short, over the course of the day, forecasted routes morph into real routes.
The forecasted orders gives the real time routing engines a head start on building routes and responding to demand. But will the forecast really be good enough? If it is too widely divergent to the actual orders that are placed, there really is no head start at all.
Forecasting 101 will tell you, first of all, that over time an accumulation of data should lead to improvements in forecast accuracy; and secondly, that a higher volume of orders can be more accurately forecasted than intermittent low volumes.
To achieve higher volumes, multiparty retailer/courier collaboration would be very helpful. If one courier was delivering orders for several retailers in a metropolitan area, route forecasts would be more accurate while the cost per delivery would go down.
Nevertheless, there can be big surges in orders based upon the season and retail promotions. If the courier company can flex and add new couriers that use their own vehicles, then demand spikes can be easily accommodated. However, if there are transportation capacity issues (the number of delivery vehicles is static), variable delivery fees can be used to shape demand fulfillment. In effect, a buyer is told if we can deliver between 3 and 4 pm, the cost is $5, if you want it between 6 and 7 pm, the fee will be $25.
Using delivery price as a lever won’t be easy; it requires integration between the routing engine and the web store. If a courier has to integrate to several retail web stores, that integration could get really complex. On the other hand, there are also advantages to not depending on temp workers for deliveries.
Victor mentioned that the kind of process he was describing would not occur over night. He though it might be 10 years before it became widespread. I will venture a forecast of my own. We will see robust forecasted route optimization far sooner than we have goods delivered to the home by drones.