In business-to-business transactions, each side has an incentive: Sell high or buy low. For sellers, knowing how the buyer will respond to an offer is difficult. But new research from Maryland Smith is helping to figure it out.
For many sellers, their main issue is making pricing decisions without the benefit of having complete information about their customers, says Maryland Smith’s Ilya Ryzhov. Prices vary from customer to customer, and they fluctuate based on the product being sold and how much buyers are willing to pay for it. That can easily complicate transactions and create headaches for sellers, he says.
“The goal is to learn the largest amount that a client is willing to pay,” says Ryzhov, associate professor of operations management at the University of Maryland’s Robert H. Smith School of Business. “Any higher price than that would be rejected and any price lower than that risks leaving money on the table. You have to do your best with these yes/no answers.”
Published in Production and Operations Management, Ryzhov coauthored the research with Maryland Smith’s Michael Fu and UMD PhD student Huashuai Qu, and three others from Vendavo, a Denver-based consulting firm that specializes in B2B pricing science.
Together, the researchers analyzed historical data that included information on tens of thousands of B2B transactions involving a single seller and a large number of buyers. The main challenges for sellers, Ryzhov says, included the high cost of failure, only being able to see binary yes or no outcomes from the buyer, as well as the sheer enormity of the data with thousands of unique products and buyers.
“This statistical model tries to consider these characteristics so that you’re not limited to learning about something only when you sell it,” Ryzhov says. “If you have 10,000 different products that you're selling, you want to be able to use each sale to learn something about products that are similar to the one you sold. It is important to learn quickly because B2B contracts are high volume. If you lose a lot of sales, that's a very high opportunity cost.”
To overcome these challenges, the researchers employed an approximate Bayesian inference model to help pricing decisions self-adjust to reduce the risk of very inaccurate estimates. That led to the development of a patented “Bayes-greedy” pricing strategy, Ryzhov says, that optimizes an estimate of expected revenue by averaging over large numbers of possible revenue curves.
“What we’ve developed turned out to be quite fast to run,” says Ryzhov,” which is important for real-time pricing.”
Read the full research, “Learning Demand Curves in B2B Pricing: A New Framework and Case Study,” in Production and Operations Management.
Media Relations Manager