The Challenge of Marrying Data, Business Strategy at AI Summit New York

Panelists from Estee Lauder, StitchFix and the Stanford Institute for Human Centered AI joined Omdia’s Graeme Tester at the AI Summit New York to discuss how data science is informing overall business strategy and how companies can best use it within their organizations. 

While all agreed there’s significant value in being a data-driven company, they also agreed it comes with its challenges. 

“The emphasis is there,” said Arvind Karunakaran, assistant professor of management science and engineering at the Stanford Institute for Human Centered AI. “There’s also top management buy-in that they are all convinced that there’s value in being a data-driven company, but the challenge lies in setting the team structures right, especially now that every unit has access to data, and they think they own it.”

Integrating the data that goes to top management is the biggest challenge and overcoming that challenge is where Karunakaran says he sees a lot of bottlenecks. 

“As a technology challenge of combining data and harmonizing data sources, how do we get these teams to work together?” he said. “Each of them thinks they’re losing control and jurisdiction over certain things. If those are taken care of there is a lot more value data can create in informing business strategy.”

Sowmya Gottipati, vice president of global supply chain technologies at Estée Lauder,

said there are multiple layers in using data science to inform decisions and that leaves a gap between trying to figure out which business problem you want to solve and then translating it into that data. 

That gap, she says, is huge in that there’s all this data in the data lake and then the data science piece comes into it and then you bring in some insights and then it’s given to the various markets to make that data meaningful. 

“There is a whole process and the gap that needs to be filled,” she said. “It’s not just the data science and the data but the many different people that play in this chain if you will.” 

Gottipati says the most important place to start is in organizing the data. 

“In my supply chain role, master data management is extremely important,” she said. “Different people look at the data differently and they want to create their own reports. Master data and the governance around it is extremely important.” 

Kevin Zielknicki, data science manager at StitchFix, a company founded as a data-driven organization, uses data to help match its inventory with styles for its clients and their personal preferences, with classic recommendations programmed to collect a lot of data.  

“We have data woven into all the other aspects of the business as well,” he said. “Marketing to clients, organizing pick paths at warehouses, data science is involved across the business from an organizational perspective.”

StitchFix does AB testing for product development, using that data for intelligent decision making. 

“When we want to make decisions around what types of product experiences, changes we’ll roll out or not we heavily use AB testing to inform that decision making,” Zielknicki said. “We also do a lot of analytics as well to try and understand areas where customers are experiencing pain points to know where we want to invest more.” 

One of the biggest challenges companies face is overcoming the mental block of whether they have enough data to make decisions, Karunakaran said.  

“Everything starts with trying to understand the quality of the data that we have.’

Once you understand the quality of the data, then you can leverage it for making intelligent business decisions. 

“The avenues are enormous, some are manually developed and some decisions are intelligent,” Gottipati said. “For example, one huge area with various components in it is balancing demand planning versus supply planning. We have 30,000 SKUs now and a planner has to plan for the demand of the next six to 18 months across 30 brands, and across the region and affiliated retailers … it’s a very complex multidimensional problem.”

Gottipati says they have to run these models to forecast their demand against supply and work with the suppliers and feed it to the manufacturers. 

The use of data has evolved at both StichFix and Estee Lauder and as more data is collected, the more development of modeling approaches leads to the delivery of more value over time. 

“We had to start out with various simplistic models that were basically just looking at base rates and things .. but as we are able to scale and collect more data, we’re able to also develop more complex modeling approaches that make better use of the new types of data that we are collecting,” Zielknicki said.

He said StitchFix has gone through several cycles of developing new recommendations model frameworks that “we see increasing performance in terms of our client satisfaction metrics, revenue and such and that is something we are able to keep innovating on as we scale, as we collect data and deliver more value over time.”

“At the end of the day we want to build an agile and resilient supply chain,” Gottipati said, because events like the pandemic can interrupt the supply chain. “We need to be able to have resiliency in our supply chain so it’s resilient. 

She said having a more accurate forecast and understanding of how much inventory to carry, that’s what Estee Lauder tries to improve upon when using the technology and the data.

The current application the company is using for demand planning and supply planning has shown at least a 20% increase in accuracy, Gottipati said. 

As for the future, creating a unified data plan and making data sharing more seamless is where panelists see data science evolving.

Making things more seamless is a big goal for StitchFix, Zielknicki said when thinking about some of the challenges the company has faced since it evolved. 

The company had a lot of additional model complexity with different models running for different areas of the business and as it grew and expanded, there’s more making it important to keep track of all the various data systems. 

“Recently, we’ve been making a more unified model framework around all of these different products and services, business lines and regions so we can have a more unified understanding of our clients,” Zielknicki said. 

Gottipati said it’s been challenging organizing the data at Estee Lauder as the company grew organically over time with the acquisition of new brands, which brought with each different system. 

“The data is not uniform,” she said. “It’s across many different systems and many people want to access the data for many different reasons, so we need to create a much more unified data framework and governance around it on who has access, who has edit versus read versus when someone wants to create a report. I think that needs to evolve into the future.”