As I make the rounds speaking to the leadership in my organization, many have concerns about the state of our analytic capability. I have heard many times that “we don’t have predictive capability.” When I’ve dug into this comment, I find that 1) it isn’t true, and 2) predictive capability isn’t what the speaker was talking about in the first place.
One of the most important parts of transforming an organization into a data-literate and data-driven powerhouse is understanding where your capabilities lie currently so you can assess the gaps between your current and desired state. Part of that understanding is in understanding the capabilities.
This article isn’t intended to give you a thorough understanding of the capabilities, but rather to help articulate the difference between accepted definitions for capabilities in advanced analytics to the folks who might not be so familiar with them.
So here’s the quick breakdown of descriptive, predictive, and prescriptive analytics and what they’re used for.
Descriptive analytics are the first step in any analytics program. They do exactly what you think they should do from their name – they let you break down big data into smaller, useful, describable chunks of information.
The primary purpose behind descriptive analytics lies in the past. They summarize what has happened historically, or give you a synopsis of what your metrics are currently.
Some examples of descriptive analytics in human resources are course enrollments, course completion rates, survey results, number of new hires, number of current jobs for hire, average length of time for new hires, and so forth. These are typically rolled into business metrics and while they might be used for performance analysis, they are purely backward looking.
There is a subset of this that we call diagnostic analytics, which are collected simply to see why a particular thing might have occurred. They are still backward looking, but they have a different purpose.
Predictive analytics are your next evolution up the analytics ladder. These include everything from stochastic modeling to machine learning, taking trends observed in historical data and projecting them forward to make predictions about a system’s future behavior.
The term “predictive” is something of a misnomer, however. I think of even the most powerful predictive analytics as simply a weather forecast, providing you probabilistic outcomes of what might happen given the current condition of your environment.
My boss asked me what the role of intuition and experience would be in this new data-driven organization we’re trying to create. Quite simply, we’re all gamblers when it comes down to it. There are lots of cases in everything from the tactical to the strategic environment where we have to make the gut call. Predictive analytics just give you more accurate odds and more likely outcomes than you had before.
Think of it, again, like a weather forecast. There might be a 79% chance of rain in the next hour, but you’re still the one who decides if you need an umbrella or not, or if you can wait to go outside until later.
You see a lot of this in the human resources domain when it comes to predicting employee churn and turnover and when certain jobs are likely to become available.
Let’s step up a level again to prescriptive analytics. Here, not only do you predict outcomes but you begin recommending (or prescribing) an action or a set of actions and predicting the consequences of those actions.
Prescriptive analytics take the predictive analytic model and simply build in a “what-if” engine based on a feedback system that tracks outcomes produced by outcomes taken. Based on the outcomes of this what-if engine, you’ll get a recommendation for an action.
Is this science fiction? Do you need ‘droids to accomplish this? No, we’ve already got it. You likely open your Waze when you’re getting ready to get on the road and trying to beat the commute. The GPS is one of the simplest examples out there of prescriptive analytics. Social media engagement strategists use a version of this same system to figure out, instead of the best time to drive, the best time to engage an audience.
In the human resources domain, you might see prescriptive analytics used to determine if a development initiative or incentive might have a positive impact on employee performance, or if the population you’re targeting with a bonus or special pay might respond with higher retention.
These seem like very simple use cases, but remember that the higher you go, from prescriptive to predictive, the more data you need!
So what comes after descriptive, predictive, and then prescriptive analytics? Logic would dictate that it’s not only a system that can predict possible responses to action but then decides to take an action based on the course of action with the best predicted consequences.
And yes, we call that artificial intelligence.
What do you all think? Did we break this down well enough, or are we missing a few things in there?