Just as fashion trends seem to be cyclical, so too do technology trends. The fluorescent running clothes I purchased on closeout several years ago are once again trendy, just as a basket of technologies that were “hot” a decade ago–VR, connected devices, and artificial intelligence –are once again front and center in the technology press.

It’s easy to be skeptical that these technologies will finally gain market acceptance, just as my wife is legitimately skeptical of my fashion foresight versus my ability to find stuff in the bargain bin that’s so outdated it’s about to become fashionable once again. However, the difference this time around is that these technologies are being applied largely to consumer applications, and thus have massively scalable applications rather than solving narrow problems that ultimately became too expensive for the technology that was available.

SEE: AI is booming, but can the benefits live up to the hype?

Understanding what’s available

The first step in formulating a useful AI strategy is understanding the current capabilities and determining if they’re relevant for your business. Despite television ads where people sit down with AI platforms like IBM’s Watson and have a relatively normal conversation, we’re not quite at the point where you install some software or buy a cloud subscription, and suddenly have access to a sentient and ultra-intelligent being.

Most recent advances in AI technology have been on the software side, particularly around machine learning algorithms. These are essentially highly complex models that allow a program to iteratively refine its pattern matching capabilities. For example, machine learning allows the smartphone in your pocket to recognize pictures of Mom once you point it at a few examples.

The problem with the current state of machine learning algorithms is that they’re largely dependent on the quality of the data that are provided. Image processing is an easier problem for machine learning to solve, since the data are in a standard format (a photograph), versus determining reasons for customer churn or learning a new language, where data are not standardized and it’s difficult to connect cause and effect and derive a set of rules.

Furthermore, preparing the algorithms and datasets is a complex and costly endeavor. You can’t install something like Watson, type in the URL for WebMD, and expect to have a medical expert within a matter of hours. Even if the hardware and software is relatively low-cost, the costs of development and data “massaging” can be massive and ultimately have an unknown result. Simply put, in a lot of cases we’re not sure what problems current AI technology can effectively solve.

SEE: IBM Watson’s latest gig: Improving cancer treatment with genomic sequencing

Ideal use cases

That leaves us a few years from the vision of a hyper-smart AI that can do anything from booking our business travel to determining whether our new product should be launched in the Asian market in January or April. Where AI has made strides in the workplace is with robotic process automation (RPA). At a basic level, RPA is the next generation of screen scraping technology, whereby a user can “teach” a software-based robot how to do a process visually, and then let the robot do the work. Rather than interfaces or APIs, the robot mimics interactions with the mouse and keyboard, so business logic that’s part of the existing interface is accounted for. This also allows streamlined “training” of the robot, since you demonstrate the process you’re automating using mouse and keyboard rather than code.

If your business has a significant processing component, for example managing applications for financial services or memberships, or requires frequent keying and referencing between systems, RPA could be relevant. An easy test is to walk the floors of some of your processing centers and look for dual-screen setups, where operators are comparing data between systems or rekeying information from forms into different systems. There’s some fear that RPA will negatively impact the job market, however many of the tasks where RPA is relevant have been offshored or outsourced since they’re relatively rote, low-skilled tasks that could ultimately free your people to focus on higher value activities.

Another reasonable use case is pattern matching, much like the facial matching technology that’s now common on phones and online photography sites. This is relevant in areas like security to detect intrusions or sharing of confidential data, or in industrial environments to detect signs for equipment failure before they occur. Where there is a relatively clear cause and effect relationship, this pattern matching capability can be quite robust and sort through volumes of data that would be impossible for a human.

SEE: Forget the plow: Robots and facial recognition for cows will be essential tools on the digital farm

Plotting the course

Some of the current advertising around AI technology has created unreasonable expectations. Key to planning how you can leverage AI is understanding what this technology can and cannot do, and including the large cost of “training” the AI and preparing the associated data. Thankfully, there are several AI technologies available through cloud providers, allowing companies to experiment and see if this technology is relevant to their business. Key players have made their offerings available, with IBM’s Watson, Microsoft, Google, and Amazon Web Services leading the way in making AI tools available in the cloud. Do be aware that despite easy access and often free pricing tiers, AI and machine learning are not for the faint of heart and will require skilled developers to begin exploring use cases.

Look for areas where rapid pattern matching, based on definable rules, exist for your initial experiments. For example, using AI to identify missing signatures on a scanned loan application would be a better initial foray than trying to develop an AI to identify fraudulent loan applications. As with any new technology, start with the simplest use case that provides demonstrable value and then scale from there.

At this point, unless there is a clear and obvious use case, and a return that justifies a large investment with uncertain return, AI should be a key part of your R&D portfolio and a technology worth observing and following, as it has the potential to augment human decision making, as well as automate routing tasks previously sent offshore. However, like all emerging technologies, the “hype” surrounding AI is exceeding the technical capabilities, so limit your experimentation to narrow use cases, and temper expectations as the technology develops.

Also see:
Baidu launches open AI platform to give businesses an easier on-ramp than Google or Microsoft
White House details AI’s scary impact on jobs, and how government can fix it
The secrets to big data project success for small businesses