Is your business ready for AI? Artificial intelligence has become common conversation in business media as we seem destined to eliminate the fiction part of science fiction. The technology brings with it high hopes.

A Deloitte survey of what it calls "aggressive adopters of cognitive technology" revealed three-quarters expect AI to "substantially transform" their organizations within three years.

But AI isn’t a one-tech-fits-all solution. It’s an array of systems, processes, tools and algorithms — all still developing rapidly.

Adopting artificial intelligence will require substantial investment from companies, in both technology stacks and talent to understand and utilize these instruments. Thus, deployment of AI will require a confidence in return on investment. Based on a new report in Harvard Business Review, the most likely ROI will come from two primary areas.

Following an examination of more than 400 actual use cases, the authors of the report determined that marketing/sales and supply chain management/manufacturing are the two business functions most likely to benefit from artificial intelligence.

In fact, the study estimates they’ll make up more than two-thirds of AI opportunities, at least in the short term.

The authors found AI has the most potential impact in areas that provide companies the most value. Take a retailer, for example. Marketing and sales provide significant value, and personalized promotions can deliver a 1 to 2 percent increase in incremental sales for brick-and-mortar retailers, according to the report.

Advanced supply chain forecasting can reduce inventory costs, and value in manufacturing is derived from predictive maintenance, saving money through operational functions.

The report further asserted that the most potential for creating value through AI was in applications that already benefit from advanced analytics. In other words, where AI can greatly enhance some of the existing tools or systems, as opposed to applying AI in areas that currently see little traction from advanced data use.

But effectively deploying AI is no small task. Gartner estimates as many as 85 percent of big data projects fail for various reasons (remember Microsoft’s racist chatbot?). In addition to the financial investments for talent and technology, fueling AI with the substantial data required to produce useful results presents a challenge, as does maintaining integrity of that data to account for security and privacy laws.

A lot goes into AI implementation, and another Harvard Business Review article offers a few key steps to give artificial intelligence deployment a better chance for success.

The first step it suggests is to clearly declare your purpose. Have a distinct and specific end-game. Know why and how the deployment will drive actual results.

Secondly, choose wisely what to automate. AI works best by automating routine processes. Apply it to cognitive tasks, freeing up your human manpower for more social responsibilities.

An example provided is the Apple Store, where technology has been utilized to automate unseen tasks, but you’re still given a human greeting and live customer service rep. AI still isn’t the futuristic, self-thinking robots of movies, and works best in conjunction with humans applying the social skills.

Finally, choose your data wisely. More isn’t necessarily better data. Yes, deep learning has an appetite for data, but dirty data can produce flawed results or bias in outcomes.

Certainly, the use cases and applicability of AI will increase over time, but for now artificial intelligence appears best applied to a small few — and specific — business functions. Algorithms have shown a propensity for problems when asked to do too much.

This article from Fast Company poignantly suggests that AI is better at making really smart toasters than "wonder-boxes."

So, if you’re hoping AI can help take your business to a new level, take a careful and considered approach to where and how you apply it.