When the facts change, I change my mind. What do you do?

I just finished an excellent Forbes article on the surprising difficulty enterprises are having in recognizing financial impact from applying AI to marketing applications. I say “surprising” because my introduction to AI, in late 2014, delivered immediate and unquestionable returns. The Forbes article explains that, according to the Forbes accountability report, “70% of AI initiatives have shown little to no return for marketers.” 

The heart of the issue is that AI has become a buzzword in the c-suite. Solution providers now refer to everything in their portfolio as “AI.” In this environment, business stakeholders need to have a laser-focus on how and where AI can add value. Start with a material business problem--not a small optimization challenge, but a big, hairy opportunity--to transform the way a business drives results. My introduction to AI in 2014 provides an example of what such a framework looks like from an executive perspective.

The Problem

I was on the hook to increase digital traffic, online conversions, and, ultimately, margin dollars. I think almost any stakeholder can relate to this challenge. Notice that nowhere did I mention AI. That’s KEY POINT #1: Applying AI to marketing doesn’t start with AI, it starts with a clearly-defined business problem.

The Hypothesis 

On some random Tuesday in 2014, during the NFL season, CBS Sportsline sent me an email recapping my Fantasy Football Team’s performance over the weekend. WOW, I thought, an automated, fully-customized, data-driven, editorial communication recapping one of the thousands of “make-believe” contests from that weekend had just landed in my inbox. I thought...could this be applied to marketing? Could a platform leverage data to construct hyper-personalized content that would make marketing communications more effective and increase traffic, conversion, and margin dollars?

The Search 

The answer was, not exactly, at least not in an editorial sense, but there were other, better options to address the core business problem. I came across a company (who happened to have an AI platform) that claimed to be able to understand how language resonated with audiences, segments, and individuals. Essentially, they had developed a proprietary database of tagged language, giving it the ability to use AI to identify the words and phrases that resonated with individuals and then used that insight to create content that was mathematically proven to drive double-digit lift in conversion rates across marketing touchpoints. KEY POINT #2: You must identify a very specific opportunity to address the business problem. In my case, this opportunity was to create better language based on data, which leads to KEY POINT #3: True AI requires a very large and unique dataset contextually related to the problem and opportunity (in my case, a database of millions of tagged words and phrases) that can be used as the basis for the AI.

The Test 

“Bullshit,” I thought. There’s no way choice of words has that type of impact on customer behavior--but hey, I’ll test anything. KEY POINT #4: Business leaders must be willing to experiment and make decisions based on the data they receive, despite existing biases. There is no point investing in AI if teams are unwilling to change behavior based on the insights delivered. As John Maynard Keynes (or Paul Samuelson, depending on your source*) famously said, “When the facts change, I change my mind. What do you do, sir?”

The Epiphany 

That initial test worked, and eventually unlocked the power of words across the marketing program, exceeding all expectations. This takes us to KEY POINT #5: Once AI proves its ability to add value, it must be scaled across the enterprise to unlock value. In this example, as data-driven language experimentation expanded across the enterprise, the AI platform was able to generate more learning, resulting in more accurate underlying models and, ultimately, more business impact.

The Bottom Line 

As with most new technologies that really work, a lack of clear return is often an execution issue, not a technology or platform problem. In 2020, it pays to clearly define the problem, identify specific hypotheses, implement true, data-backed AI to address the hypothesis, then embrace and scale the results. This is admittedly a massive oversimplification but it’s also a proven framework for driving material financial returns from AI. 

Keynes/ Samuelson quote attribution