There’s no question that machine learning and predictive analytics have entered the public consciousness. A few things made this possible. For starters, compute power has gotten much faster and more economical, and data transfer speed and storage costs improved dramatically, all of which have enabled artificial intelligence (AI) algorithms to scale up to the highly glorified big data. But, enterprises should be wary of buzzwords and the cure-all promises of artificial intelligence before they jump in. Although business leaders may have a good sense of the outcomes they desire, many lack knowledge of what it takes to get there, like differences in data sources and types, and the nuances of different types of machine-learning models.
As more vendors have jumped on the AI bandwagon, many terms, like predictive analytics and machine learning, have become very popular marketing labels, something enterprises should keep in mind when looking at predictive analytics solutions. Understanding these labels is the key to overcoming hype, and getting real value with true self-service solutions for your operations teams to directly manage their business challenges.
There’s an old joke – I’ve heard many versions – about the guy who’s searching on the ground around a lamppost. A friend comes by and asks what he’s doing. “Looking for my keys. I dropped them over there,” he says, pointing toward some bushes. “Then why are you looking here?” asks his friend. “The light is better,” says the guy. Known as the Streetlight Effect, this story describes something called observational bias, referring to the habit of looking for things where the search is easiest. It’s also a something to consider when talking about predictive analytics and machine learning. The streetlight effect can show up in predictive analytics when algorithms are applied either where developers think they’ll find the best insight or where they’ll find an issue.
If a company says it’s applying machine learning, for instance, but it’s just one tiny feature, and developers still need to manually configure search parameters (such as ‘bug’ detection in game development), then there’s only a marginal impact on functionality and operations for the analysts. Until recently, most machine-learning models predicted exactly what they were trained to predict, so their forecasts were only as good as the data used for their training. A new generation of true machine-learning algorithms, which don’t need to be told where to look and learn from experience, promises to completely change how many businesses will operate. Eventually these should deliver truly self-service operational forecasting to analysts, without relying on Data Science and Data Ops teams. Which are you being offered?
The other big issue is that data sources themselves are quickly changing. Up until recently, we’ve been limited to the use of “data at rest” – data sets that are self-contained and fixed. “Data at rest” provides valuable historical context which makes it possible to make business predictions based on past experiences. But, Internet of Things (IoT) technology, connected data sources like sensors, and social media feeds now provide us with new sources of meaningful data that changes over time.
The application of real-time data has opened up an exciting new branch of predictive analytics called anomaly (or outlier) detection, or the ability to recognize unusual, anomalous behaviors as they occur. When we’re talking about data-driven insight for businesses, that ability to identify problems you didn’t know could arise and which could lead to issues that have to be dealt with down the road, is obviously worth gold.
In fact, anomaly detection is already being leveraged. Waze, the popular crowd-shared navigation system recently turned to a Tel Aviv-based company called Anodot, to help it find problems which couldn’t be anticipated. “Anodot is helping us find anomalies and identify issues that drivers may not feel, but we can see the differences when looking at the big picture,” says Dr. Orna Amir, Analytics Group Manager, Waze. “The changes won’t make the drive faster. But, using Anodot helps the company understand user preferences, like detecting trends in certain countries, or features that are not performing to their full potential and that will help us provide a better experience to our users.”
Anodot’s platform looks at the big picture of the data across the board. Rather than being told where to look, the machine learning system analyzes all data history and data streams to detect anything out of the ordinary, and then signals an alert for further action.
Shipping giant UPS also recently announced they’ve started using a new generation, real-time predictive analytics algorithm to optimize the movement of parcels across its delivery network. UPS introduced its ORION algorithm back in 2016, but this new Harmonized Enterprise Analytic tool, which was developed in-house, will now provide functionality that used to be spread across different applications.
“Today, we use data extensively to plan,” UPS chief information officer Juan Pereztold The Wall Street Journal. “But the more real-time data we can get on the state of a package, the better visibility we can get about any exceptions in the network, helping us generate improved plans to manage the network as a whole.” Predictive analytics will enable UPS to forecast demand, so it can direct its trucks and planes to locations where they’ll be needed most to improve service and save money. The next version, says the company, will employ artificial intelligence to also determine best actions in real-time, to eliminate human bottlenecks in the decision-making process.
Despite these success stories, a 2017 McKinsey Global Institute study found that many business leaders remain uncertain about what exactly AI-based technologies can do for them. The good news? The same study found that, “AI adopters were 27 percent more likely to report using AI to grow their market than companies only experimenting with or partially adopting AI, and 52 percent more likely to report using it to increase their market share.”
The key is knowledge. As predictive analytics continues to advance, it’s important for business leaders to learn and understand all the definitions and sort the hype from reality. If a vendor tells you it’s using machine learning, find out exactly what the machine will be doing and what you will still be doing, and how that helps empower your operations and analysts. Don’t hesitate to ask questions and clearly verbalize your goals. The more you know what to expect from predictive analytics, the sooner and more effectively you’ll optimize your business outcomes.