Let’s use a very simple example. You take your car to the shop and they determine that two of your tires have a tire tread below 2/32". That’s the legal limit for tires. Past that point, it’s been determined that the risk of accidents, blowouts, hydroplaning, etc. are too high. That’s predictive analytics. True, it’s very simplified predictive, but predictive nonetheless. It’s telling you that, if you don’t change something about the tires, the likelihood of something bad happening is too high.
So, you know you have to do something about these two tires…But what…?
You may not realize it, but there are a lot of options to be considered when replacing two bad tires. What brand do you buy? Do you replace all 4 or just those 2? From whom do you buy the tires? Who installs them? Is it even worth buying new tires, or should you consider selling your car? If so, for how much? Do you still replace the tires before selling? And so on...
Even something as simple as reaching a tire tread threshold brings up dozens of response options. How do you know what the best choice is? Well...like most people, you use the advice of “experts” and some amount of “gut feel.” Even so, you won’t know until after the fact whether you made a wise choice. Further, you certainly won’t know if you made the best choice, because that relies on a completely different type of math (optimization!).
Predictive can tell you when you should consider taking action, but it can’t tell you what action to take. Thus, if you rely on gut feel or standard practices to make decisions, sometimes you’ll end up selecting a bad response. The more complex the decision, the greater the likelihood becomes that you won't select (or even consider) the best plan.
So, by making poor decisions in response to predictive analytics, predictive all of a sudden becomes significantly less valuable.
Let’s get more technical now.
As we said above, predictive analytics uses forecasts and statistical models to assess and suggest what could happen.
In Supply Chain, as an example, predictive analytics looks at all the possible situations within demand planning, costs, profits, inventory optimization, logistics, etc. and allows organizations to better anticipate anomalies throughout their network.
Predictive modeling has been around for a while and has a wide range of cross-industry and cross-functional applications. No matter your role or industry, there’s a good chance you could be leveraging predictive analytics as shown in these examples:
Just to name a few.
Predictive analytics is gaining momentum with the rapid increase of Internet of Things (IoT) devices. These devices embed sensors in equipment to monitor and transmit data continuously to IoT platforms. Twenty billion units are expected to be deployed by 2020.
With this influx of technology, predictive analytics is more powerful than ever. The management consulting firm Deloitte estimates that the global chemicals industry — which employs 20 million people with annual sales of $5 trillion and is the backbone of many end-market sectors like automotive, construction, and pharmaceuticals — could recognize billions in savings through IoT and analytic solutions.
Here’s the value that manufacturing companies see from predictive analytics:
Then what’s the problem?
Predictive analytics is a form of analytics that optimizes one line of business’s (LOB) operational efficiency, for example, only to leave other LOBs lacking similar results. Even worse, the new and improved LOB negatively impacts other LOBs.
Let’s dig deeper...
Imagine a chemicals company that uses predictive modeling to find out when equipment is due for maintenance and at what point parts need to be replaced. Data from sensors’ historical events are leveraged for statistical analysis to forecast the likelihood of events. While this action can significantly cut maintenance costs and better forecast spend, predictive modeling does not provide insights into the best courses of action.
We can’t say it enough...predictive is not actionable!
It wouldn’t tell the company whether it’s worth paying the money to fix/maintain a machine. It wouldn’t tell the company where to move the production/processing of products while the machine is down. Predictive only gives you a small piece of the equation, but it doesn’t help decision makers execute on decisions.
There are two primary ways that decision-makers develop action plans and respond when predictive data reaches a threshold.
Both forms of decision-making can be categorized as prescriptive analytics. (Want to learn more about the types of prescriptive analytics? Visit our interactive web page: Your Ultimate Guide to Prescriptive Analytics.)
So, the chemical company’s machine is down and will require maintenance. Let’s see how each decision-making approach impacts the company.
The first approach to leveraging prescriptive analytics for decision-making is largely hypothesis driven. In short, these involve coming up with several hypotheses — derived from gut feel or past business experiences — and quantifying them with a spreadsheet or an OLAP based tool (think Anaplan).
With a hypothesis-driven approach, the chemical company is limited to a finite scope of response options, all of which must be obvious to the human eye. A hypothesis-driven approach would suggest it’s time to shut down the machine for maintenance. Production from that machine then gets allocated based on rules that were developed by a plant manager a while back.
“When a machine is down, we move run overtime to meet demand”
Or
“We pre-build a certain level of inventory that we can dip into when a machine is down
And so on…
Let’s look at the other approach
The second main approach to answering the question “what should I do when my machine needs maintenance” is constraint-based optimization modeling.
Here’s how it works, in short.
A model of the chemical company is created, one that represents everything from raw materials, to the end customer. This includes intricacies within all of its manufacturing operations, things like manufacturing lines, labor costs (S/T and O/T), available inventory, available capacity, transportation costs, etc.
When the machine shuts down, instead of limiting herself to static rules that don’t necessarily match the real-time business realities, the plant manager runs a handful of optimization scenarios in order to determine the best path forward.
She is not limited by static rules. Not only that, she can see the impact of her decision on the larger value chain, things like profitability, transportation costs, labor hours, etc. Optimization can enable her to make a decision that meets one or more of her objectives while still respecting constraints in other LOBs.
The plant manager can look at the real-time state of the business and of her plant and use data to answer questions like:
As you can see, our plant manager has a lot more optionality!
Optimization has several benefits over hypothesis-driven decision making:
Here’s the point: predictive analytics is unable to provide both the physical (resource) and financial/economic outcomes of decisions. And that’s why executives should really care.
You should be paying attention to how you respond in the event of a high-risk situation, not just making sure you respond.
Wouldn’t you want to know if that machine that’s due for maintenance is a necessary asset? Or if fulfilling that demand could simply be delayed a day or two and sold for a discount?
What about that advertising offer you serve certain visitors of your e-commerce site: wouldn’t you like to know by exactly how much you should discount a product in order to maintain profitability and ensure a 95% view-to-sale rate?
Predictive analytics tools leverage an incredibly powerful form of advanced analytics. What the market fails to address, however, is that they will never be as valuable as they could be without the appropriate tool that offers actionable, optimized insights.
Until business executives embrace the need for actual decision support, they’ll only be getting a fraction of the value from their advanced analytics investments.