As groundbreaking as new technology can sometimes be, the benefits aren’t always obvious. This isn’t the case for artificial intelligence (AI) — numerous articles have already been written about what it is and how it works, how to apply it and the benefits to businesses, large and small.
What isn’t so clear is how to go about determining exactly where, in your own business, it can and should be practically applied, and reconciling yourself with the fact that, sometimes, the best application of AI isn’t as sexy as you’d hoped. In fact, in many cases, an effective use of AI isn’t sexy at all. And in other cases, it may not actually provide the best return on investment (ROI).
The trick is identifying the difference.
Here are four questions to ask yourself and your team to identify internal processes that could benefit from the application of AI.
What processes in your organization require binary decisions?
Binary decisions of yes/no, true/false or category A/category B may only be part of a person’s workflow, but isolating this work is a good place to start when looking for areas where machine learning can be applied. Is this a fraudulent transaction? Would this customer like this dress? Is this atypical computer activity? If you can’t identify questions with binary answers at the outset, then the process of applying AI could be substantially more difficult.
What decisions do you make that are based on complex inputs?
While the final decision may be a simple yes or no, machine learning can be extraordinarily effective when a large number of factors are considered to reach a correct answer. Music or movie recommendation services are a good example of this. By gathering massive amounts of information about its users, a music service like Spotify or Apple Music can make great recommendations to a single listener based on that user’s activity, compared with the activities of all its other listeners.
What decisions could be automated?
If the inputs on which a decision is being made are not something a computer can make sense of, machine learning would not be applicable. For example, customer service associates might be making decisions based on a variety of data points, but the source of those data points is answers given by a customer over a phone. At this point, a human is still better positioned to extract and analyze that information.
Could you get a tangible benefit from such automation?
This is where "sexy" tech can get you in trouble. Looking for a cool place to implement the technology without regard for ROI is a fool’s errand. If machine learning can’t save a company time or money relative to a particular process, then what’s the point? Many companies still need a "fear barrier" to effectively determine how and if AI can help.
While asking these four questions can help you identify processes where machine learning could be employed effectively, it doesn’t guarantee that it can be employed effectively. You still need to test and confirm its applicability.
Sometimes Humans Are Still Better At The Job
At my organization, we recently put AI to the test in two areas. In one, it worked tremendously well. In the other, it didn’t.
My company, DiscoverOrg, provides sales and marketing teams with actionable B2B sales and marketing data that can help our customers build their sales pipeline. As such, we have to actually identify that “actionable data.” To this end, we regularly ask whether the companies in our database have recently launched large-scale projects or initiatives.
One of the data sets we review to answer this question is job postings. Originally, phrases were extracted from these postings and sent to an analyst to evaluate if the language indicated the existence of a large project or initiative.
So, we had identified a binary decision, based on complex inputs of a specific data set.
Once we applied machine learning to the process, the AI began to learn from the decisions of the humans. More specifically, it learned from the analysts’ nos. Phrases that originally prompted a section to be reviewed were eventually filtered out because the analysts always deemed them irrelevant. Our analyst efficiency increased as the volume of the sections they had to review fell.
Another data point that we provide is reporting structure. For the past 10 years, since we were founded, our analysts have had to ask and identify where someone sits within a company’s organization, which takes a lot of time and results in voluminous amounts of data.
We engaged a firm that specializes in machine learning and ran a pilot project to see if the question “Who does this person report to?” could be answered by AI using that data. After two months, it was determined that humans were better at the job because of the significant variability in reporting structures between organizations.
These two scenarios distinctly illustrate that sometimes AI works … and sometimes humans are still better at the job.
Ultimately, when determining whether AI can be useful to your business, try to avoid getting caught up in how sexy the technology is (and what the business environment is saying about how valuable it should be). Instead, focus on outcomes specific to your business. Can the tech improve your organization’s decision making, workflow efficiencies and ROI?
If the answer is no, then move on.
