Forecasting and demand management for new events using machine-learning algorithm


Football - demand planning and forecastingWhat does a machine-learning (ML) algorithm have to do with the Super Bowl?

When it comes to forecasting and demand management, a lot.

Consider this: According to the National Retail Federation, approximately 189 million people watched Super Bowl LI, and viewers spent an average of $82.19 on electronics, apparel and food specifically for the game, up from $77.88 compared to the previous year.

For events like the Super Bowl, retail demand planners create forecasts using data from a variety of sources to adjust product demand profiles in anticipation of which product, or group of products might be in demand the most.

This is a daunting task when one considers the variety of products available to football fans – from cheeseheads to cheezies and everything in between. In the past, only about one brand in 50 was able to precisely adjust their football-frenzy driven supply chain to meet demand during the short two-week window between the conference championship games and the Super Bowl.

The ability to forecast the uplift in demand reliably to guarantee consumer product availability and to evaluate the economic returns on the promotions has largely been a dark art. Without improved technology, very few companies can plan effectively in a promotion-heavy environment to help people jump on the Tom Brady bandwagon with an authentic Patriots jersey.

That’s where ML comes in.

In recent years, the ML algorithms have become more sophisticated, using data visualization to help subject matter experts determine the meaning of the results. Previously, when left without interpretation, ML algorithms produced data without context, providing no clear conclusion.

However, ML is advancing, refining models to determine correlations between data without human interaction. As more data enters the system, the system becomes more intelligent and the data becomes more manageable and subsequently, easier to interpret.

This type of analysis has applications beyond the football field, easily extending to science and engineering, and to fraud detection, genetic analysis and finance, just to name a few.

But back to football. Unleashing a machine-learning algorithm on the Super Bowl assists planners by helping them adjust to unpredictable product demand just by digging through historical data. To find the right correlation, the ML algorithm classifies products by similar event type, helping retailers strike the right balance with suppliers to ensure that giant foam finger gets to its destination on time.

By integrating ML algorithms with a decision support system, prescriptive analytics gives decision makers timely recommendations of which product or product types to focus on for the event, reducing “what-if” analysis.

For the supply chain practitioner, leveraging ML to help with demand planning and forecasting, and take advantage of opportunities like the Super Bowl, this is their “I’m going to Disneyland!” moment.


Iman joined us on December 7, 2015 as a Solutions Blueprint Developer. Iman graduated with a Doctorate in Industrial Engineering from Concordia University, and was working at Canada Post as process engineer for the last four years. At Canada Post Iman was designing logistics and transportation networks to fulfill customers demand with minimum operational cost. Iman enjoys reading, programming, and solving real and challenging problems. In addition to Kinaxis, Iman also is a part-time instructor of simulation course at the Sprott Business School at Carleton University.

More blog posts by Iman Niroomand


  1. I really liked this blog. Doesn’t the fact though, that once the two teams in the Super Bowl are established, the task becomes easier, and ML is not required, or does ML help predict how much of the team merchandise to have ready?

  2. Hi Joe, Thanks for your comment. It’s a very good question. I would say ML would predict how much the Super Bowl or any other event would impact the selling of the products, so ML would be helpful to adjust the demand of these products automatically. Second for sure ML can predict the team merchandise as well. This would be tricky a little because the historical data should be team specific and would require good historical data for learning and prediction.

    I hope that I have answered your question.

Leave a Reply