Can You Predict What People Will Pay?
Analytics-based pricing-optimization can impact a company’s bottom line, says David Simchi-Levi / Photo: Quim Roser
How do you accurately predict what price customers are prepared to pay in order to maximize both sales and margins?
Felipe Caro and David Simchi-Levi, experts in supply management, joined IESE MBA students and alumni on campus in Barcelona in March to explain how modeling and machine learning can be used to do both.
Fashioning a Global Success
Felipe Caro is associate professor of Decisions, Operations and Technology Management at UCLA Anderson School of Management. He has enjoyed a long association with Spanish-headquartered fashion multinational, Zara.
Zara’s global success stems from its leadership in innovation in the ready-to-wear sector, said Caro. One of the principal challenges the fashion retailer faces in its operations, he said, is how to optimize item pricing during sales periods – a conundrum the Inditex brand has attempted to overcome by applying powerful modeling techniques.
The Highest Possible Low Price
With markdowns, the object is to maximize margins while selling as much stock as possible. Unsold – or salvage -- stock is stripped of its branding and sold by weight to countries where Zara has no outlets, for a fraction of cost price, explained Caro.
Historically retail managers have had no fixed methodology to use with markdowns. They have had to rely on their own (sometimes limited) experience and "gut instinct" to suggest markdowns, which in turn would be prone to discussion and ratification across different levels of the organization – a time-wasteful process with unpredictable results.
In 2005 Caro began to work closely with Zara’s CFO to develop a tool that would help accomplish markdown goals – both across its own chain and external, franchise stores in any territory.
Tried and Tested, Then Rejected
Caro and his team developed a mathematical methodology that would take the guesswork out of the process, by dynamically combining demand estimation with price optimization. They tested their method it in the Republic of Ireland and Belgium – with success.
Zara immediately set about designing a usable tool for all its managers. However, they found that in many territories there was reluctance on the part of managers to deploy the new tool itself – and in some cases, shop managers were adopting a mixture of the new system and old methods.
The Human Factor
Part of the problem Caro learned was skepticism over the forecast.
"No prediction can be 100 percent, and this combined with the fact that managers were only being told how effective the tool had been in their territory at the end of the discount period meant that they were unenthusiastic about using the tool."
Zara responded by giving managers an ongoing figure for each week and enabling them to compare figures for the corresponding week in the previous year.
"It's important to understand how people think if you want them to use your model" said Caro.
Online Retail – A Multi-Billion Dollar Business
David Simchi-Levi is professor of Engineering Systems at Massachusetts Institute of Technology. His research focuses on developing and implementing robust and efficient techniques for manufacturing and supply chains.
The challenge he outlined to MBAs also related to understanding how people think and behave: specifically, how willing people are to buy goods online at an optimal price.
Simchi-Levi says that the application of machine learning can be used to not only work out the optimal price for maximum sales margins, but also to avoid endangering market share; significant news for the online retail industry, which is currently worth $300 billion, and predicted to grow at an annual 10 percent.
Real Time Retail
Online retail has the advantage over bricks-and-mortar of being able to gather and act on data in real time. But the very sophistication of these types of business, says Simchi-Levi, is also one of their biggest challenges. Sites offer daily deals and "flash sales," where prices can remain low for just 48 hours. Added to which, high demand uncertainty, short product lifecycles and retailers with limited inventory can present complex challenges in terms of supply chain, he said.
Exploration Versus Exploitation
The solution, says Simchi-Levi, is developing a "balancing act that has to trade off exploration and exploitation:" the longer the period the system has to become more accurate at predicting customer behavior, the less time it is left to actually utilize the information. And of course the opposite is true, too.
Moreover, any workable solution has to accurately predict what price the customer is willing to pay; secondly, it has to figure out how much time should that price be offered; and thirdly, the order in which the different prices should be on offer.
Different types of retailer require different approaches, too. Simchi-Levi cited the example of Rue La La, which operates flash sales and has a severely limited inventory – products that brand-name suppliers won't allow to be sold at reduced prices after the initial offer.
In addition to maximizing returns, the business is concerned that while higher prices may result in greater profits, they might damage market share.
To address this, Simchi-Levi has combined machine learning and optimization techniques; an approach that is set to yield a 10 percent revenue increase for Rue La La with little impact on demand.
"Rue La La’s use of this new price-optimization application demonstrates how analytics can change the way a company operates. We have created a cutting-edge, demand-shaping application that has a tremendous impact on the retailer’s bottom line," said Simchi-Levi.
Felipe Caro and David Simchi-Levi were keynote speakers at the first IESE Fashion Operations Conference.