The data-driven insight is clear: if you watch a film starring Nicholas Cage, you face a high risk of falling into a swimming pool and drowning.
Andy Pulkstenis was clearly trying for a laugh, and he got one. The director of advanced analytics at State Farm Insurance in Bloomington, Ill, was leading a breakout session at Analytics Experience, a conference for customers of enterprise software giant SAS Inc. this past Fall.
While much of the event was designed to inspire organizations about how they could use data science and its corresponding tools to help influence their corporate direction, Pulkstenis was offering a series of cautionary tales. This includes the potential bias that might be hidden within artificial intelligence (AI) applications, false positives or simply confusing correlation with causation. Or, as Pulkstenis put it, “Your boss thinks X must be causing Y” simply because two data sets (like Cage’s filmography and drowning statistics) are put side-by-side.
“There’s this sense that if you have big data and machine learning it will now solve all your problems. That’s not true — it will create more problems,” Pulkstenis told the audience. “We’re assuming the algorithm is more important than the data, when the majority of the algorithms we want to use are essentially free, and the majority of the data is not.”
While most organizations have “observational data” about their customers or competitors, Pulkstenis explained, many stop short of using experimental data based on structured, scientifically guided testing.
“Too often we’re fishing for insights in observational data without any valid, rational hypothesis,” he said.
This raises issues that are likely to become a bigger topic in B2B organizations and elsewhere over the course of 2019: are we moving too fast into analytics without really understanding what it is? If the marketing of martech and AI vendors doesn’t delve into enough detail, will the objectives businesses have for the tools be met? And are we really developing a data-driven workforce, or one that merely thinks it is?
There’s no question that the initial hype around analytics and AI is creating a shift in how post-secondary schools are trying to prepare the next generation of business professionals for the workplace. Since 2014, for instance, programs have popped up at schools across the U.S. and Canada, including the University of Michigan’s Michigan Institute for Data Science, the University of California, Irvine, Data Science Initiative and the College of Charleston’s data science program. Even traditional business schools are incorporating analytics more deeply into their curricula.
“A decade ago, business students were looking to go into a prescribed path like marketing, banking or the consulting profession,” Erika James, John H. Harland Dean of the Goizueta Business School at Emory University, told B2B News Network in an interview late last year. “We’ve seen at our business school an inordinate amount of interest among students who want to understand information in a way that allows them to do what they want to do.”
Companies aren’t waiting around for graduation day to get value out of analytics and AI tools, of course. Christian Nelissen, head of Enterprise Data and Analytics at Canada’s TD Bank, describes his firm as a data company, rather than a financial institution, since moving data around is the majority of what its team does today.
According to Nielsen, the propensity for errors, or the kinds of problems Pulkstenis discussed, come when those on the analytics team aren’t sitting close to the business, metaphorically speaking. In other words, data-driven insight is seen as something you need to outsource versus weaving it into the organization’s entire approach to doing business.
“(The analytics team) will get an question to solve via e-mail but 80 per cent of the time once that’s answered, what (management) comes back with is, ‘What I really meant was X,’” he said, adding that many data scientists and analytics teams often go above and beyond despite being kept at a remove by management or other lines of business. “My favorite analogy is where someone in analytics will say, ‘I know you asked me for A, but I did B, and while I was doing B I found C, because I know you’re interested in C as well.”
Analytics is unlikely to be closely aligned with the rest of the company if the company is not closely aligned to the customer, argued Tricia Wang, a data science consultant with Sudden Compass who also spoke at the SAS Analytics Experience event.
While enterprises may need to educate themselves further on bias in AI and concepts such as multicollinearity, Wang warned that quantitative tools should not be seen as a replacement for understanding your customer. Instead, they should be seen as a way of activating that understanding.
“Company-centric companies invest in data and wind up knowing less about their customer,” she said. “What we see here is that digital is often treated as a checklist. Once you do it, you’re done, and it’s a technical process — people are brought into the room who are not close to the customers. Often sales and marketers are left out.”
James, who recently joined the board of SurveyMonkey, had similar sentiments, calling for a more holistic approach where human experience is factored into analytics, especially as the limits of the tools are better understood.
“The corporate sector has invested tremendous resources and partnered with universities to ensure that that (analytics) skill set is a part of the environment. I think they do so almost to an extreme, and lose sight of the fact that organizations are units of human interaction,” she said. “There is an empathy and emotional intelligence that comes with running and being part of an organization or a business. If we focus too much on making sure they’re technically adept with skills, I worry long term about the viability of organizations writ large.”
Nielsen suggests what he calls the “Christmas party test”: the business should be close enough to the analytics team (and its customers) that you couldn’t think of celebrating the holidays without them in the room. He also suggests extending the idea of “customers” beyond household purchasers.
“Personalization is just as important for corporates as it is for consumers,” he said. “It can be as simple as recognizing a company’s date of incorporation or the date that they first banked with you,” he said. When the bank notices a company’s energy bill is higher than usual, for example, it offers an opportunity to help them with a specific pain point by looking at their energy usage and offering advice, versus merely selling them another service.
“I think most of the time what customers dread is that sort of, ‘How’s everything going?’ (outreach),” he said.
Getting the customer relationship right, along with a more nuanced view of how analytics teams can contribute value, is probably a better bet than hoping a vendor’s new release will save the day, Pulkstenis added.
“It’s like if you had invented the car and you live on an island and you can’t get to the mainland because there’s no bridge, and you so you invent a Ferrari next,” he said. “If you’re struggling implementing simple data science solutions, improving the algorithm is not going to help you that much.”