Best Buy: random sales forecasts?

Written by Mohamed Barraoui, on 25 October 2017

January 2005, Richfield, Minnesota. Jeff Severts, newly appointed Director of Consumer Product Marketing and Brands at Best Buy, decides to tackle a major issue: reducing the gap between sales forecasts and actual sales...

With the support of the Chief Strategy Officer, the new marketing director decides to conduct experiments to improve sales forecasting estimates. To do this, Severts draws inspiration from the book "The Wisdom of Crowds" by American journalist James Surowiecki.

Severts launches an initial experiment within his department. He sends an email to about a hundred voluntary employees asking them how many Best Buy gift cards they think can be sold in February, with a trend curve from previous months as the only guidance. To motivate the troops, Severts promises a $50 reward to the person who provides the most accurate forecast!

The paradox of the wisdom of crowds is that the best group decisions come from many independent individual decisions. - James Surowiecki

All responses are consolidated. Severts then notices that the group's forecast is ten times more accurate than that of the experts in the field!

"The crowd has won!" he announces in his email sharing the results. While the experimenters rejoice, his team of experts feels somewhat humiliated by the experience...

Severts decides to conduct further experiments, ensuring in his communication to explain his approach to avoid it being perceived as a critique of anyone's work.

His second experiment is no longer restricted to his own team but is open to all Best Buy employees. This time, the marketing director asks them to estimate the company's revenue during the holiday season.

Once again, Severts provides them with little data to guide them. He doubles the incentive to $100 for the employee with the best forecast. Participants, under a pseudonym, have 14 weeks to make their estimates on a simple shared Excel document.

At the end of the year, the result is clear: while the experts' simulations are accurate to 93%, those of the crowd are accurate to 99.9%!

In January 2006, Severts decides to launch a third and final experiment: the implementation of an internal sales forecasting platform, "TagTrade" (developed by Consensus Point), initially open to a hundred voluntary employees. The platform allows employees to propose a sales forecast for a virtual stock of products.

The marketing director then solicits both the expert forecasting department and, via TagTrade, his employees to predict sales for a new laptop service offering.

A week before the service is launched to customers, the market prediction was 44% below the official prediction! It subsequently plummeted.

As a result, when the initial sales figures confirmed the market's estimate, Severts terminated the service offering and began redefining its scope. He then revised and corrected it in TagTrade to have a prediction ready for the start of actual sales by mid-September. The market rebounded, and Severts declared that he benefited greatly from it.

After validation by Best Buy's management, the sales forecasting program resulting from these various experiments was extended to inventory management, customer service, and customer engagement.

TagTrade is now open to all 115,000 Best Buy employees in the United States. The 2,100 participating volunteers receive one million dollars in virtual currency to "trade" over a nine-month period. The top "traders" earn a $200 gift card.

TagTrade's prediction is not only more accurate than traditional estimation models (by over 5% in some cases), but it also accurately predicts delays or, conversely, on-time delivery of several major initiatives, including new services and store openings.