SCOTTSDALE, Ariz. — Mention data science and you get one of two reactions: excitement or glazed-over eyes.
Data science is a broad term that many don’t fully understand. But according to Matt Taddy, vice president of Amazon Private Brands, data science is the key to unlocking value for businesses.
Speaking on Thursday at the Amazon Business Reshape 2022 conference at the Hyatt Regency Resort and Spa, Taddy explained that data science — and understanding how to use it properly — has been key to helping Amazon (NASDAQ: AMZN) grow its Private Brands division, which includes Amazon Basics, Essentials, Elements, Wonder Bound and Mama Bear. In all, there are more than 40 Amazon Private Brands worldwide.
“Amazon uses science intensely in everything we do to solve customer problems,” Taddy told an audience of about 400. “We approach problems with a scientific [approach].”
When Taddy joined Amazon, he said the goal was to “inject technology” into areas of the company where it wasn’t. Talking to Amazon Business customers at the session, Taddy pointed to the keys to using data science, noting that many don’t utilize it correctly.
“Machine learning covers a really broad swath, but as leaders, what it does is detects patterns,” he said. “Data from past patterns can inform changes for the future.”
Saying that machine learning by itself is not that useful, Taddy illustrated with some examples of how Amazon leverages data science to improve products and pricing.
“It is very rare that you can look back at data [and find it useful for making decisions going forward]. How do you understand customer reaction to pricing?” he asked. “It’s also not that useful to how a customer is going to react to future prices.”
Strictly following machine learning in this case could cause bad decisions. The science will tell you that as prices go up, revenues go up. But that is not useful information in and of itself. Amazon, Taddy said, adjusts pricing weekly on many of its items and does it randomly so that it can gauge how the changes impact sales.
“[It’s] to not just have business as usual but to have actual data on how customers respond to price changes,” he said.
Offering advice to those at the conference, Taddy said to utilize machine learning to generate the data necessary to make decisions but keep the human element in the loop. His second point is to generate good data.
“Having good metrics and goals against those metrics is how you develop good quality leadership,” Taddy said. “As leaders, you have to know that you have to be skeptical of those metrics, if they are really driving us toward our goal of improved customer engagement.”
Having the right metrics is important, Taddy said, giving the example of return rates. Clothing and high-value items have the highest return rates. But if a customer wants to lower those rates and uses only the data around return rates, the solution would be obvious: Stop selling clothes and high-value items.
Obviously, though, that is not the right answer. Taddy said Amazon looked at this problem and created a new metric — return over expectation. This measured the return rate of an item in relation to the expected return rate.
“What I’m interested in is metrics on what the customer does next and whether that is going to lead to a long-term relationship,” he said.
Taddy advised the attendees to pick their spots to deploy machine learning, suggesting the first area to attack should be cost structure.
“What do materials cost? That’s a performance issue with a lot of clear metrics,” he said.
Finally, as the data is generated, Taddy said to be sure to funnel it back to production and create a “loop” of the entire process to validate insights. Failure to do this creates a one-way data flow and limits the usability of that information in truly creating change.
Click for more articles by Brian Straight.
You may also like:
Amazon Business wants to improve B2B e-commerce journey
Amazon Warehousing & Distribution is company’s latest foray into logistics
