Earlier this year, I wrote a blog aimed at demystifying artificial intelligence and providing some alternatives to long-term, custom projects. Indeed, these large-scale endeavors can be daunting, especially for an organization that is just beginning to dip a toe in the water with AI. Not only can smaller, “quick win”-type projects deliver immense value relative to their investment, but they can also build momentum and support in the organization for larger-scale analytics initiatives.
But just how quickly can you get an artificial intelligence or machine learning system off the ground, delivering value for your organization? Quicker than you might think.
In this post, I’d like to describe five different AI-powered solutions that you can get started with quickly. In many cases these solutions can be deployed inside of 90 days — provided your data doesn’t require too much clean-up and you have a solid understanding of the business objectives you are working towards. In fact, depending on your specific circumstances, there are off-the-shelf solutions that could get you started even quicker than that.
Traditionally, market segmentation was more art than science. Marketers would develop traditional personas based on standard demographic factors like gender, age, and marital status, and then attempt to neatly bucket their customer base into one of these. Even those that took a more data-driven approach would have to compile information from research and then eyeball some commonalities.
You can use clustering to help you build segments that are more robust and accurate, and save time while doing so. Because this analysis is AI-driven, you can analyze a much larger data set with many more features, including psychographic and behavioral factors. This will enable you to build more distinctly defined segments and define the customers within each with a greater degree of confidence.
AI Accelerator Program: Your 6-Week Jump-Start
Quantify Survey Responses
Sometimes, the best feedback from a customer is their own words, but words don’t boil down neatly into percentages and graphs. (Please don’t get me started on word clouds.) Too often we fall back on rough metrics, like Net Promoter Score and Likert scale-type questions, which don’t fully get at the questions we truly need to ask. What we really need to do is take a deeper look at narrative responses to find what patterns emerge.
Fortunately, there are ways that machine learning can help boil these words down to numbers that paint a meaningful picture, using a discipline called “natural language processing,” or NLP. You can use NLP to scan for keywords that point to the most relevant topics that come up in people’s replies, as well as score the tone of the feedback to assess whether it was positive, neutral, or negative. With these methods, you can get statistical data on people’s sentiment (positive, negative, or indifferent) on various topics (for example, features) without asking even a single quantitative question.
Increase Pricing Accuracy
Whether you’re looking to find the optimal price point for your own products or services or attempting to determine an optimal bidding strategy when competing in a market, advanced analytics can help you zero in on an appropriate range.
Leveraging historical data, machine learning can detect patterns in data and use it to predict a range of values, including the low and high ends of a price range. For example, if your company happens to make regular wholesale purchases of a certain product, you might want to take a more scientific approach to bidding rather than relying on gut. Using a regression analysis, you might find that the ideal price point is influenced not only by obvious factors like quantity and weight, but also weather, time of year, and other external factors.
These models can range from very simple to very complex. Nonetheless, you can often get a viable model going pretty quickly and reduce the element of “gut” involved in price setting.
What is it that makes an employee likely to stay? What are the warning signs that one might be on their way out the door?
We’ve heard the conventional wisdom about what it means to be an engaged employee, or what makes one a flight risk. But with modern realities such as the Great Resignation, the work-from-anywhere and hybrid work movement, and changes in what employees value in their careers, there are new red flags to watch out for.
Instead of trying to figure out a range of predictions like the pricing example, here we just want to answer a yes-or-no question: is this employee a threat to leave? Again, machine learning can help analyze a large dataset and find the right predictors, drawing upon data like performance reviews, meeting attendance, and survey results. Ultimately, you’ll be able to figure out which employees are more likely to leave than others and take steps to target engagement efforts towards them.
Detect Unusual Activity
Financial institutions of all sizes have leveraged artificial intelligence to identify likely fraudulent transactions. In this case, we can’t really provide predictor features, because the definition of fraud may vary from customer to customer. A carpenter in Texas may spend $1,000 at a home improvement store in Dallas, and a pub owner in Maryland may spend the same amount at a restaurant supply store outside Baltimore, but reverse those transactions and they become suspicious. This is anomaly detection, and in this case, it is also considered “unsupervised” learning as we don’t need to label the dataset as “fraud” or “not fraud” – it will simply find instances that are so far removed from the others as to raise a red flag.
While many banks already use this method for fraud, there are applications in other disciplines as well. Cybersecurity professionals can use anomaly detection to monitor suspicious logins, policy violations, or possible attacks. In manufacturing, it can be used to identify production flaws or predict equipment maintenance schedules to reduce costly repairs.
So Why AI?
None of the items discussed here are new activities. Some are things you might doing now, but artificial intelligence and machine learning can help you do them better, faster, more accurately, or in a way that brings in much larger quantities of data. Here we talked about clustering, natural language processing, linear and logistic regression, and anomaly detection. In many cases, you can deploy simple approaches in a matter of weeks that will start delivering real value for your organization.
Where would you begin tomorrow?
Interested in launching or expanding AI within your team or organization? Our Advanced Analytics team would love to answer any questions you have. Feel free to schedule a quick call here.