Individuals living in the higher latitudes will tell you that ice dams on your roof are an omen of bad things to come. I learned this first-hand last week after record snowfall in the Boston area. To minimize future damage, I set out to purchase a roof rake to clear the snow off of my roof. I drove to one big-box home improvement store – out of stock. I called the major competitor -also out of stock. As a long shot, I decided to call the small TrueValue Hardware store down the street. They had plenty of roof rakes in stock, with a number of models to choose from.
This personal experience motivated me to discuss weather’s effect on supply chains and the methodologies employed to manage the variability, disruptions, and uncertainty associated with weather events. How are supply chains currently planning for and adjusting to weather events, and how do their processes differ from one another? Was TrueValue just lucky in this event? Maybe they tend to overstock all items? Or maybe they planned better than their competitors?
Planning for Weather Events
I view planning for weather events as something that can be done fairly well for the long-term (season to season) or for the short term (7-10 days), but not so well between those time horizons. The company Planalytics provides functionality that focuses almost exclusively on weather related analytics and other capabilities to manage weather’s business effects. Steve previously wrote about Nestle’s use of Planalytics for weather-driven demand planning, noting that past weather events and their effects are embedded in sales history that is then used for forecasting. It is important in forecasting to identify and remove these effects to correct for past anomalies, then explicitly add back inventory expectations related to weather. Forecasting product demand stemming from discrete, infrequent weather events is a probability related exercise that is likely to result in a regional inventory decision, rather than for a specific site or location. Therefore, depending on the specific case, it is likely that a probability forecasting algorithm similar to those used by spare parts inventory optimization would effectively support such planning.
Mid-term weather forecasting still appears to be the “wild-west” of meteorology. But I did recently came across a company with a website (http://www.weatherplanner.com) that provides “interval forecasts” for as long as one year in advance. I am unsure of their record of accuracy but I plan to test it over the next couple months. But assuming variation from one winter (or hurricane season) to the next is essentially random, the challenge then falls to adaptation at an operational level.
Operationally Adapting to Weather Events
Demand sensing applications are used for refining replenishment plans by leveraging POS data as it becomes available. Traditional demand sensing may be beneficial to updating replenishment plans for ongoing weather trends like the one we are experiencing in Boston, but it’s not likely to be beneficial to isolated, short-term weather events. These situation require accurate and comprehensive inventory visibility across the network, and adaptability to allocate inventory to appropriate nodes in the supply chain. However, this needs to be done in an environment where the weather is also causing transportation delays and capacity constraints. As Mark Derks noted in his post on the subject, these weather events can greatly increase the cost of shipping due to tight lead times, routing guide substitution, and accessorial charges. These cost increases can be mitigated by analyzing your transportation network and developing a plan to leverage available resources.
There is also some interesting development in routing applications worth noting. Steve previously discussed the application developed by Quintiq that uses past demand and delivery routes as a framework for future demand, then adapts this framework with real orders as they occur and runs an optimization algorithm to route the deliveries efficiently. One of the core benefits of this methodology is that the plan-to-actual variance provides a cost-to-serve estimate that can be used to evaluate the cost-benefit trade-off for demand opportunities. I believe a similar methodology would be beneficial to regional fulfillment of inventory driven by discrete weather events.
Another example of sophisticated routing that is currently used by select carriers is the Esri GIS solution. Steve and I were briefed on its use in transportation. The Esri GIS solution allows carriers to overlay their routing with maps, routes, traffic patterns, and weather events. This allows users to determine and forecast delays on various routes, determine alternate routes, and compare the impact of various alternatives. This differs from hub and spoke models because the distances and times are more accurate, and it is also more sophisticated in its ability to select a given geographical region, and apply a multi-factor analysis to the region. These factors can include weather, altitude, as well as slope and grade.
Forecasting Black Swan Events?
During lunch today, I went out to purchase another snow rake (my parents borrowed mine). This time TrueValue was out of rakes. I also checked ACE Hardware, and they were all sold out as well. I went to the Planalytics website earlier in the week to view the case study videos as Steve suggested in his post from a number of years ago. Interestingly, ACE Hardware is a Planalytics user and they have a video presentation on the site. It looks like proper weather planning wasn’t enough to meet the uncharacteristic demand from the recent storms. This leads me to the distinction between variability and uncertainty. Some events, termed “black swan” events in the economics trade, are so rare they often not worth planning for. I guess that’s when insurance comes into play.
Stay warm and well.
