Every year, The Walt Disney Company employs at least 70 field agents to survey over a million park and hotel guests, employees, and travel industry professionals. That data is transferred to a team of 35 analysts who use it to predict park attendance, hotel bookings, how long attendees will wait for a particular ride, and how many times they will ride it over their stay. Using this approach, Disney has been able to produce a five-year attendance forecast with a low, 5 percent average error and has been able to produce annual reports with a 0 to 3 percent error, according to Operations Management, Sustainability and Supply Chain Management.
Forecasting is the art and science of predicting events. In a business setting, an accurate forecast can be the difference between making a profit or taking a loss. Baked into larger business operations, forecasting can be overwhelming for many small business owners who struggle to know how a forecasting model can benefit and integrate into their business.
Local Chick-fil-A owner, Dave Hamel, faced a forecasting challenge in August of 2018 when Kansas State University Athletics announced a partnership to sell Chick-fil-A’s famous chicken sandwiches at all 2018 home football games. It was a huge opportunity, but leading up to the first game, he wondered exactly how great the demand would be.
Putting Together a Game Plan
Although the Chick-fil-A corporate office provides software to generate in-store forecasts to anticipate daily sales volume and assist with supply/food orders, it did not provide a way to forecast how many sandwiches he would likely sell in a stadium of 50,000 fans on opening day. The sandwiches would need to be cooked and prepared in the restaurant and then transported in heated containers to Bill Snyder Family Stadium for food safety and quality purposes. Hamel’s own game clock began ticking at that point, because each sandwich has a finite shelf life to be sold, after which it would need to be discarded.
An incorrect forecast in either direction would result in an undesirable outcome: produce too few sandwiches, and Hamel risked generating unhappy fans on the first day; produce too many sandwiches, and Hamel would lose money in wasted labor and sandwiches. In addition to an overall estimate of sandwiches to be sold, Hamel wanted a more granular forecast to determine how many sandwiches would be needed throughout the five-hour window, starting two hours before kickoff when the stadium gates officially opened to fans. Would more sandwiches be sold in the 30 minutes leading up to kickoff or during the 20-minute halftime? The answer would guide the size and frequency of deliveries from the restaurant.
After consulting with K-State athletics, the food service company Sodexo, which runs concessions in the stadium, and other Chick-fil-A franchise owners who had served food at Big 12 football games, Hamel received additional guidance on how to generate a ballpark number for first-day sales. The input from these resources outlined multiple factors that would influence sales volume on any given day, including:
Start time of the game—morning, afternoon, vs. evening
Weather/temperature —cooler games would result in more sales of hot sandwiches
Number and location of concession booths
Opponent and game result—blowout vs. close game
A Costly Fumble Against South Dakota
With general guidance, but no concrete data, Hamel began creating estimates for the first football game against South Dakota. Attendance for the 6 p.m. kickoff was expected to be high, and initial forecasts for the day called for cool weather and a slight chance of rain. With a good amount of publicity surrounding the partnership between K-State and Chick-fil-A, and a need to avoid a sandwich sell out, Hamel set a very aggressive goal: three times the sandwich sales volume of a typical Saturday.
Using that high goal as a starting point, Hamel coordinated with his leadership team, consisting of Ashley Napier, director of catering/marketing, and Jack Fisher, director of team development, to create a detailed schedule for the 72-hours leading up to and through the football game. It outlined when food shipments needed to arrive, when frozen chicken would need to be placed in thawing cabinets, how many employees would be needed to filet, bread, and cook the chicken, and the delivery time of cooked sandwiches to the stadium. By breaking the overall goal into 10 minute segments, with projected surges in demand right before the kickoff and during halftime, Hamel was able to predict staffing needs for point-of-sale registers and additional behind-the-counter personnel.
In an effort to improve the fan experience, and to capture more discrete sales data during the game, Hamel purchased Square™ point-of-sale registers and requested K-State install hardwired internet to the two Chick-fil-A booth locations in the stadium. K-State Athletics quickly fulfilled the request on short notice, anxious to see if the technology could translate to other concession booths.
When game day arrived, Hamel’s team was prepared to deliver on his ambitious goal, three Saturday’s worth of sales, crammed into a single pregame and four quarters. It would require a Herculean effort, but as the stadium parking lots filled with tailgating fans early in the day, cooling storm clouds went north and the stadium temperature continued to rise. Without the expected reprieve, the heat index soared to over 100 degrees.
When the gates opened, an expected rush of fans anxious for a Chick-fil-A sandwich didn’t materialize. Many fans ordered cold water and soda in an effort to combat the oppressive heat, but within the first 30 minutes, it was clear that sandwich sales were far below the initial forecast. After an hour-and-a-half, sales were trending to less than a third of the initial target. Hamel slowed down the production at the restaurant but was optimistic that halftime sales could possibly bounce back to projected levels.
However, just as the Wildcats were sluggish in a slow start that resulted in a 27-24 victory, foot traffic to the booth was equally sluggish when fans, who had spent hours tailgating in the hot weather, opted out of a “fresh and hot” sandwich. The halftime “rush” had about as much urgency as the Wildcat defensive line that day, and final sandwich sales finished at less than a third of the original goal. Even though the production of sandwiches had been slowed, Hamel’s staff circled the stadium during the fourth quarter, handing out dozens of free sandwiches to security and parking staff, and hundreds of sandwiches remained, most to be thrown away.
Hamel, disappointed, took solace in what he had learned from the launch: first, his team had successfully demonstrated their capacity and ability to increase sandwich production to handle the original sales goal; and second, he now had minute-by-minute data of the night’s sales that would provide a baseline for a more accurate forecast. This would be needed the following week when the team played host to SEC powerhouse Mississippi State. “I paid some expensive tuition tonight,” Hamel said, “but I believe the education will be worth it.”
Mid-game Adjustments Pave the Way for Better Results
Hamel, a former K-State football player, is no stranger to halftime adjustments. And despite a disappointing first act, he knew he had six more home games to refine his results. The Mississippi State game would be a morning game, with kickoff scheduled for 11 a.m., giving fans less time for tailgating and placing lunchtime right after the first quarter. After speaking to other concessionaires, he knew that sales would be higher for this game. On the morning of the game, a cold front passed through Manhattan, and the game-time temperature was expected to be only 60 degrees — perfect weather for a nice and hot chicken sandwich.
With a revised goal and minute-by-minute forecast based on the previous game’s data, Hamel’s efforts and education positioned himself for dramatically different results. By comparing real-time sales data with his forecast model every 15 minutes, he was able to see how far ahead or behind he was from his overall forecast and continually recalculate a projected final sales total for the game, which allowed him to adjust sandwich production at the restaurant to align with the real-time, end-game estimate, helping him minimize wasted sandwiches.
The earlier start time and cooler weather resulted in nearly double the sales of sandwiches, a result that left a big smile on Hamel’s face as well as his team’s. The frustration from the previous week had melted away. Looking forward, Hamel was excited about a third straight weekend with a home game, this time against the University of Texas at San Antonio (UTSA).
Trusting the Game Plan
The UTSA game, with a 3 p.m. kickoff, presented an additional forecasting challenge. The weather was expected to be in the ‘80’s, neither too hot nor too cold. Hamel wondered whether sales would be slow like the first game or brisk like the previous week. Hamel received mixed feedback from his various sources, with some saying that afternoon games were better than evening games and others saying that they resulted in much lower sales. In response, he set a fairly conservative goal. As the gates opened that day, fans poured into the stadium. The first hour resulted in a stream of customers that was consistent but dramatically slower pace than the previous two games. After the first hour, the real-time forecast was estimating an end-game result of barely a third of the original forecast.
Hamel, an optimist by nature, initially resisted the story the data was telling. It didn’t seem possible that sales could be that low for this game. The weather was great, the football team had jumped out to an early lead—on the way to a 41-17 victory, and the stands were filled with excitement and energy. Rather than slow down production to match the dynamic forecast, he instructed his team to stay the course and continue cooking sandwiches to stay on track with the original forecast. He assumed that things would pick up before halftime.
As the game progressed, the forecast continued to slump. As halftime approached, the data showed the final sales total would be well below a third of the original forecast. Sandwiches intended for first and second quarter sales remained, and while still good to sell at halftime, an additional batch of several hundred sandwiches was already on its way, expected to arrive at the stadium within minutes.
When the second half concluded, Hamel and his team had several hundred remaining sandwiches once again. Chick-fil-A employees circled the stadium, handing out sandwiches to stadium staff members, and extra sandwiches were dropped off at the Riley County Police Department and fire stations around Manhattan.
Although disappointing, Hamel realized that the night’s results were markedly different from the first game. The data, combined with the forecast, had given him an early warning before a single down of football was played. However, the data’s story was upstaged by other qualitative signals, which proved to be misleading, and eventually, quite wrong.
Two weeks later, with the Texas Longhorns and cooler temperatures in town, Hamel made minor adjustments to his production levels to ensure they stayed aligned with the dynamic forecast. Although sales were modest due to a 2:30 p.m. kickoff, wasted sandwiches were minimal. Despite a 19-14 loss, the day was profitable.
Hamel is confident that armed with solid forecast data, a real-time point-of-sale system, and a smart, friendly and hardworking team, he will be able to replicate his successful dynamic forecasting for the 2019 season.
Brandon W. Savage provides customer experience, operations, technology, and management strategy consulting to companies throughout the Flint Hills area. He also is an instructor of strategy and operations at the K-State College of Business Administration. He and his wife, Cheryl, live in Manhattan with their eight children. Find him on Twitter @thecxpro.