The volume of the conversation around enterprise artificial intelligence (AI) is now so deafening that it’s often difficult to separate the signal from the noise. According to an analysis of transcripts by Bloomberg Chief Economist Michael McDonough, since mid-2015, the number of corporate earnings calls that include some mention of AI has skyrocketed. Now, 80% of companies report they have AI applications in production.

In most cases, what’s missing is a more nuanced discussion about what AI actually entails, especially in a business context. Within the same group of companies that reported 80% AI adoption, 33% said the biggest barrier to that adoption is that AI technology is “nascent and unproven,” a sentiment that was second only to concerns that a lack of talent and understanding (34%) and information technology (IT) infrastructure (40%) are impeding progress.

Even if the majority of companies have embraced AI in some capacity, it appears they are still woefully underprepared to meet the demands of building and capitalizing on intelligent systems. As Forrester notes in its predictions for 2018, the AI honeymoon is over: Enterprise companies that have been “naively celebrating the cure-all promises of artificial intelligence” need to start putting in the work to support it, or else.

Is What You’re Doing AI?

It’s impossible to have a discussion about the state of enterprise AI without first acknowledging that the term has become a catch-all for everything from pattern recognition to chatbots. Like data science, AI lacks a universally accepted definition — so much so that Forrester’s vice president and principal analyst Mike Gualtieri, at a symposium hosted by DataScience.com last month, broke the concept into two distinct parts: pure AI and pragmatic AI.

Pure AI, Gualtieri noted, is the “sci-fi stuff” — sentient robots and the like — which is far outside the realm of enterprise AI. This distinction has some concerned that industry adoption of the term is muddying the waters, but many of the machine learning-based applications being embraced by businesses today are able to perform tasks that once required human intelligence, such as image recognition.

“Really, what AI is about is pragmatic AI,” Gualtieri explained during his presentation on machine learning platforms at DataScience: Elevate in San Francisco. “Pragmatic AI is narrower in scope, but it can often exceed human intelligence. Google can win a game of Go, IBM can win ‘Jeopardy!,’ but you can’t snap your fingers and apply that to another domain very easily. It’s not generalized intelligence. It’s very specific.”

Neither example cited by Gualtieri is science fiction-worthy, but techniques like natural language processing (NLP) and deep learning have the potential to help companies save time and improve operations on a massive scale. That’s not conjecture: A DataScience.com client once leveraged NLP to parse thousands of customer support inquiries and online reviews for information about the most critical issues facing its product, eliminating hundreds of hours spent on manual searches. The improvements the company made based on that information resulted in a 500-point increase in net promoter score.

As Gualtieri said, “It’s acceptable, in today’s world, to say that if you have one machine learning model that you’re doing AI, and that’s okay.”