When companies implement a demand management or replenishment system, the goal is usually to improve customer satisfaction while holding less inventory. Pinnacle Propane implemented demand management, replenishment, and order promising solutions from John Galt Solutions with the goal of improving service – reducing what they call “out of gases” – while reducing transportation costs. The implementation also involves leveraging weather data to improve forecasting. This is also rare.
Pinnacle Propane is based out of Irving, Texas. Pinnacle Propane was founded in 2010 through the acquisition of several retail propane companies in Central, South and West Texas. Over the years, Pinnacle Propane expanded across the U.S. Their operations include bulk gas storage and delivery, cylinder filling and community distribution gas systems. In 2017, Pinnacle Propane was acquired by a Dutch company, SHV Energy. SHV Energy is one of the world’s largest distributor of propane.
Gijs Majoor, vice president of supply chain and sustainable fuels, and Jacob Gladysz, the director of logistics explained the Pinnacle Propane business and their journey to improve their forecasting. When it comes to distribution, primary distribution involves getting big bulk deliveries by pipeline and truck to their manufacturing sites. Secondary distribution involves using their trucks to deliver propane from their manufacturing and intermediate storage facilities to the customer.
The forecasting projects were focused on secondary distribution. Initially, the forecasting project was focused on deliveries of 20-pound cylinders to businesses in 42 states – both large retail chains and Mom and Pops. Pinnacle Propane would deliver full cylinders and take away the empties. That business line has been divested to enable full focus on their core business line, the bulk business where the lessons learned are now being applied.
The bulk business includes delivery by tanker trucks to both residential and commercial customers and community distribution sites. In a community propane distribution system, a central propane tank in a common area on the edge of a development pumps propane by pipeline to homes in the development.
Bulk gas for commercial customers involves deliveries to segments like oilfield services and agricultural customers. The bulk business also involves filling the tanks of businesses that use propane as an energy source. For example, a warehouse might have a big propane tank they use with forklifts that run on propane.
It makes sense that the forecasting project began with the now divested 20-pound cylinder distribution line. Forecasting is harder there. With bulk distribution, businesses like Pinnacle can apply telemetry devices that measure the amount of propane in a tank and transmit that data at regular intervals back to Pinnacle Propane. This data leads to a better baseline forecast. 20-pound cylinders do not have these sensors.
The goal of the project was to make timely deliveries while reducing out of gases. Mr. Gladysz explained that we wanted to “anticipate what our customers needed and when they needed it.” Better forecasting would allow both better service and fewer deliveries to customer sites, which reduced their fleet’s transportation costs.
Forecasting was based on three years of historical shipment data, information on whether the retailer was planning a promotion, weather data, the inventory carrying capacity at a customer site, and the point-of-sale data that some large retailers could provide.
The amount of propane used by customers is influenced by the season (there are fewer cook outs in the winter), holidays (demand goes up), and weather – there will be fewer cook outs if it rains.
Pinnacle was delivering to roughly 20,000 locations. Each location was geocoded. They then used weather data from 20 regional weather stations. Each customer location was assigned to a regional weather forecast.
The company hired a demand forecaster who had a background in forecasting. There was a learning curve associated with coming up to speed on the application. Then, with all this data, the cloud-based John Galt forecast solution was used to do multi-tier regressions. In other words, different forecast models were applied to different classes of customer. Certain algorithms worked better with customers that could provide point-of-sale data, other algorithms worked better for smaller customers.
Mr. Gladysz pointed out that abnormal weather patterns could adversely impact future forecasts. If it rains in the Southeast on the 4th of July for example, the system will predict lower demand for the following year. The data that goes into the forecasting system needs to be massaged on an ongoing basis to to account for anomolous events.
Because the John Galt system is cloud based, it makes integration to other systems, particularly other cloud-based systems, easier. Nevertheless, integrating to legacy on premise applications is often not easy. It certainly was not for Pinnacle.
Initially, the cylinder forecast was not good. It took several months to learn how to use the system better and improve the data used in by the forecasting models. From a data standpoint, they learned that the data they had in their enterprise system regarding the amount of inventory a customer could hold was often wrong. The forecast might think that at a customer location, that customer could hold 100 cylinders, when in fact at that location there was space for 150 cylinders. When this happened, too many deliveries were made. If Pinnacle thought a customer could hold 150 cylinders, when they could only hold 100, gas outs become more likely.
This data was captured via driver updates and quarterly data scrubs where they asked the field sales force to review the data and look for abnormalities. Their systems also could be used to pinpoint errors. “If the system said that a customer would max out with a delivery of 100 cylinders,” Mr. Gladysz explained, “but there were 3 deliveries of 200, we knew the data was wrong and needed to be updated.”
Customers also needed to be educated on the new system. A farmer that is used to receiving a propane shipment every week, might have only needed a shipment once every week and a half or two weeks. Moving to less frequent shipments was a source of concern for some of the smaller customers. “We did lose customers among the smaller players based on the reduced shipments and the loss of trust,” Mr. Majoor said.
Mr. Majoor continued, “initially communication was a challenge” even with the larger customers. “But we built trust.” We sent out the actual results and had calls with buyers at these organizations.” These large retail chains saw that fewer out of gases translated into higher sales. The loss of a small number of smaller customers was more than made up by savings in transportation. Transportation costs went down by 15%! And for a distributor, transportation costs are a significant portion of the overall costs.
For the main business line within Pinnacle, the bulk business, Pinnacle is also using a route optimization solution called RoadNet from Omnitracs. Both the forecasting and routing solutions contribute to lower transportation costs. First the miles need to be reduced by reducing the number of deliveries. That occurs when the forecast is handed off to the replenishment system. The replenishment model factors the locations of warehousing and delivery sites, the available trucks with their varying capacities, available drivers and their remaining hours of service, travel time to each delivery location, and the service time to make each delivery. Some cylinders were filled at factories, other cylinders were filled at warehousing sites where the inventory had been forward deployed. Once the rough cut replenishment plan is in place, the routing solution generates the final routings.
The company is in the process of implementing the John Galt solution for their bulk business. The goal is not to just forecast the timing and number of deliveries by customer in the next few weeks, but to forecast the number of drivers that will be needed in the coming months. This is a highly seasonal business where demand for propane for heating is higher in the winter. More truckers are needed then. It takes four to six weeks to hire a driver and get them up to speed. There has been a driver shortage for some time. This kind of planning is very important.
With a standalone forecasting system, sales and shipments are forecast. Then the actual shipments occur, and the forecast can be compared to the results. This gives the application a feedback loop that allows the demand models to be fine-tuned.
But the feedback loop is more complicated when it comes to driving transportation savings. The integrated routing/forecasting solution is also forecasting the number of hours a driver will be on the road. Then the driver’s telematics device records the actual drive times. This data improves the routing. Improved routing in turn can lead to a forecast requiring even fewer shipments.
As the forecasting and routing becomes more accurate and more granular, flexibility increases. For example, if there is a cold snap in Missouri, there may be enough hours of service capacity to send drivers from Texas to Missouri to meet the increased demand.
What advice did these executives have for other companies that might be looking at a forecasting solution? Mr. Majoor said “start with the goal. When you have that goal, tailor the system with that goal in mind. Otherwise, you will lose yourself in all the opportunities.” In short, the solution is functionally rich. Keeping the goal in mind keeps the project on track and improves the return on investment.
Mr. Gladysz added. “But once you have that rigid goal, be flexible in the way you get there.” You may be visualizing the process unfolding in a certain way. That might not be the process that can be implemented most quickly to achieve the goal.