Cindy Howson: Best Practices for AI and the Joy of Hosting a Podcast
When I offered recent podcast guest Cindi Howson the opinion that data science has become much simpler, she had a ready response:
“Are you telling me it’s not hard anymore?”
I had to laugh. But Howson knows her data science. Now the chief data strategy officer at the company ThoughtSpot and host of the podcast The Data Chief, Howson has been in the data business for three decades. Starting with installing an early Oracle database in the late 1980s, she later founded a business intelligence company named BI Scorecard, then went on to work as a VP at Gartner, where she modernized that firm’s data and analytics programs.
Her point about the challenges of data science is spot-on: even though our tools are enormously powerful, we work constantly to avoid AI bias, creating trustworthy AI, and AI safety. But we’ve also made big strides.
“I do think about what is hard now versus what was hard back then,” she told me, thinking back to her early career. “There were no data groups to connect with, no best practices. Everything we did we had to define from the start—even security policies. I remember going to bed at night thinking, ‘I can’t do this.’”
But of course, she could. And did. At ThoughtSpot, she is helping create a new way to deploy search and AI to find insights in company data and put that data in the hands of business teams.
“The technology today makes things possible that were impossible just a few years ago,” Cindi says. “In 2020 we saw the acceleration to the cloud, and ThoughtSpot has pivoted to full SAAS [software as a service]. We’re thinking big, and our customers are thinking big.”
Still, just as we do here at DataRobot, Howson wrestles with how to get AI to return on your investment. Some studies have shown that as many as 87 percent of AI projects fail to deliver the desired results. Infamous cases include IBM’s Watson for Oncology project, which was canceled after spending $62 million, and Apple’s Face ID failure.[i]
Cindi has some best practices in mind. One is to be clear about the vision for a particular AI project. “That vision will answer the question, ‘What is the business value or use case?’ Let’s say you want to improve customer retention. Then you will run some experiments that work towards that. And you’re going to budget-box it and say, “Here is how much money and headcount we’re going to dedicate to this.’ Maybe you won’t operationalize this, but you’ve time-boxed it, and you are aligned to the business use case.”
It is also important that data scientists have a detailed understanding of the business they’re working in. “If somebody is pursuing a new degree in business analytics, maybe in the computer science department, I’m a little concerned,” she says. “I worry they’re not gonna have enough business application experience.”
One challenge, Cindi says, is convincing HR to apply the right metrics to hiring. “You would think recruiting and retaining top talent would be a high priority. But HR is often given short shrift in terms of data and analytics. I think people expect data science to be the answer for everything. But I think you need to recruit people with a combination of skills.”
“I love being a host,” she says. “I love finding people’s stories and giving them a stage for sharing. I learn so much from my guests.”
That is a sentiment I can endorse. It was great to talk with Cindi, as it is with all my guests.
[i] “Five Biggest Failures of AI, Why AI Projects Fail?”, ThinkML, Oct. 14, 2020.