The Key to Better Business Outcomes from Data Science

In a Harvard Business Review article, Alessandro Di Fiore, founder and CEO of the European Centre for Strategic Innovation (ECSI) countered the assumption “that companies with more data scientists have a better chance of generating business impact.” Based on both in his consulting work and research, he has come to the conclusion that hiring a larger number of data scientists does not necessarily produce better results for a business.

The same observation was made to me in a recent interview with Henry James, founder and deputy CEO of Fincross International, who said that what he’s seen at businesses with vast resources to invest in data science is that they can, in fact, do better with a team of five than of 50.

Extending AI to Those with Domain Expertise

What really makes the difference for a company, Di Fiore, pointed out, “is the democratization of access to AI tools and decision-making power among managers and employees which creates more tangible value.” He went on to observe, “Best practices show how democratization can bring about quicker and better distributed decisions, making companies more agile and responsive to market changes and opportunities.” (To learn about how some businesses are already using AI, check out AI Today: Who Is Using It Right Now, and How.)

While he doesn’t care for the term “democratization” and prefers that of “team sport,” Todd Hay, Ople’s COO, agrees with that view. As he explained in an interview with Techopedia, he envisions the shift from rarefied and centralized AI to the masses as analogous to the adoption of spreadsheets, a useful tool that should be used by all businesspeople.

“Subject and domain experts are in the best position to assess a prediction that can impact the business,” Hay said. But with a setup that puts data scientists in charge of those predictive models, “they’re excluded from the process.” That’s not to the benefit of the business.

Though he concedes that the data scientists have the expertise in math and statistics to judge if a model performs well or not, they don’t have the capability to determine which questions they should be putting to the AI to solve. And that gap between model expertise and stakeholder expertise is what accounts for the fact that “70%-80% of case models are never used.”

Understanding what Goes Into the Decisions

There are further ramifications to not being able to understand the way the model works. In regulated industries like healthcare, insurance or finance, Hay said, the concern is being in a position in which they have to explain the decision-making process to auditors and not being able to do it.

Rick Saletta, Ople’s senior sales marketing executive of AI, machine learning & data science, noted his agreement in the interview and said this is why businesses are now looking to develop “transparent AI,” also known as explainable AI. As we saw in AI’s Got Some Explaining to Do, in the absence of a clear explanation of how AI reaches its conclusions, you cannot be sure it is “bias-free.” He added, it is no longer acceptable to shake off the business’s responsibility to operate fairly by saying “the AI did it.”

Lessons from the Rise of the Internet

The fear that remains in the face of AI operating like a black box is holding back businesses from reaping the full benefits it makes possible. That’s a mindset that has to change, according to Hay. He suggested AI today is like the internet in the late ’90s. That means that there will be some spectacular failures like Pets.com and other such misfires due to people not being quite certain how to apply the new tech. And fear of new tech holds people back, he said: “It’s new and scary and very complicated.”

But there is also great opportunity for those who figure it out. “All the stuff that we’re seeing now was opened up by internet because people were willing to try out new things,” Hay said. It’s the same situation now with AI enabling people to find what “they didn’t even know they should be looking for.” They also should not doubt their own capability, as many “have more skills in the company than they thought they do,” particularly “subject matter experts and people who know the data.“

Making Technology Accessible Now

“We want to see how every company can take advantage of AI now — today,” Hay declared. In order for that to happen, it’s necessary for AI to be made accessible outside the circle of data science experts. “The number of competent data scientists in the world is far below the number of companies that would benefit from it,” he explained. Accordingly, the key to getting more business problems solved is “not training more people to be Andrew Ng but by making the technology available to people.”

Indeed, that is the wave of the future, according to Gartner, which predicted this year will see an increase in “self-service” analytics. Significant progress in AI, as well as complementary technologies like “SaaS (cloud) analytics and BI platforms are making it easier and more cost-effective than ever before for nonspecialists to perform effective analysis and better inform their decision making,” observed Carlie J. Idoine, research director at Gartner.

When that is put into place in a business, and more employees overcome their reluctance to help themselves to the benefits of AI, it can truly become a participatory rather than spectator sport within the organization. That shift can have tremendous impact. (If you haven’t thought much about AI for your business, here are some implementations you might want to consider: 5 Ways Companies May Want to Consider Using AI.)

Reducing the Risk by Reducing the Time and Cost

“People are so afraid of spending six months to run a hypothesis,” Hay explained, because it is such a major investment of time and money that can ultimately fail. However, if AI is not reserved for these major moonshot projects with a longer time horizon but for more common tasks that are completed more quickly, possibly even on a daily basis, they become “more like a spreadsheet,” meaning an accessible, inexpensive tool that people are not afraid to try out, even working through several different ones to find the one that best fits their needs.

However, Idoine cautions that doesn’t mean that businesses should just expect their employees to pick up on how to use and adapt it to their needs on their own. She insists that “training, support and onboarding processes are needed to help most self-service users produce meaningful output.” Accordingly, it is necessary to provide “the right guidance on how to get up and running quickly, as well as how to apply their new tools to their specific business problems.” And that — rather than increasing the data science team’s numbers — is the key to better solutions to business problems.