Why blog again about optimization and analytics? Because the current way of having optimization be part of analytics is a bit misleading. Let me first say I assume that optimization is part of analytics here. Granted, a previous post of mine supported a different view, but the idea that mathematical optimization is part of the broader category of analytics is gaining momentum. For instance, the INFORMS society is pushing for it with its Conference on Business Analytics & Operations Research.
INFORMS defines analytics as
"Analytics -- the scientific process of transforming data into insight for making better decisions."
At IBM we use a definition quite similar. It could be depicted this way.
The analytics process
The Analytics category is then refined by distinguishing three sub categories
Descriptive Analytics. It is about providing tools to look at data, look at aggregates, and be able to query and drill down. It includes Business Intelligence in particular. For instance a sales manager at a retail store chain may want to see sales per region per month per main market segment. She may also want to be able to drill down both geographically, and by product category, down to the sales of a single product (SKU) for a single store. Descriptive analytics helps understand what happened so far.
Predictive Analytics. It is about further understanding and analyzing data. It includes statistics, machine learning, data mining. For instance, predictive analytics could be used to detect trends in sales for our retail store chain. Trend detection would leverage past sales levels, as well as past price level, weather condition (we know that warm weather increases drinks and ice cream sales), etc. from this we could get a predictive model that relates sales to price and weather condition. Then, using this predictive model one could extrapolate and predict mot likely sales levels from weather forecast. Predictive analytics helps understand why things happened, and what is most likely to happen next
Prescriptive Analytics. It is about making decisions or recommending actions. This is where mathematical optimization plays. For instance, the inventory manager of the retail store chain can plan his inventory replenishment at the store level and at the distribution center level using sales forecast as input. Prescriptive analytics helps decide what we should do next.
We also like to say that the value provided by analytics increases as you go from descriptive to descriptive then prescriptive. But why is that so? Because all of the above analytics is useless till the point of action. It is only when you start changing your business that you create value. You can stare at data for years, it won't change your business if you don't act. You can analyze data for decades, it won't change your business if you don't use the resulting insights to make action. It is only when you act that you unlock the value of analytics. More on this can be found in my analytics is a mean to an end post.
Then we see why the value increases as we move from descriptive to descriptive then prescriptive. Value increases because we get closer to the point where the value of analytics is actually unlocked. Said differently, it automates a larger piece of the analytics process defined above.
The analytics journey
This way of presenting analytics and optimization within it resonates quite well. Yet, it has its own issues, including :
Predictive analytics conveys one of its business value quite transparently: it can be used to make predictions. People don't need to be explained why predictions are useful. Everybody values meteo forecast for instance. Prescriptive analytics does not convey similar business value: yeah, it could compute prescriptions. So what?
Prescriptive analytics may refer to medical prescription in layman's world. It then conveys the idea that it is a technology to provide cure after something wrong happened. This is certainly a possible use case for optimization, but it is far from being the only one.
Predictive and prescriptive sound almost the same. This leads to an easy confusion between the two as it is hard to remember which is what first time you hear it. I've indeed met many people who confuse the two.
For the above reason we've been looking for another way of describing optimization as an analytics category. Here is our current favorite:
Here's the elevator pitch that goes with it:
When you know the likely future then you can take advantage of it, and make better plans. Optimization is a great way to do it.
We tried it and it resonates well with customers. Here are possible reasons why:
Proactive analytics conveys one of its business value quite transparently: it can be used to act knowing what might happen. People don't need to be explained why this useful. People do this every day, from going to the gas station before the car runs out of gas, to ordering or buying food before starving, etc.
Predictive and proactive sound different, and are much easier to differentiate.
Proactive contains active, and contains act. It hints at the action piece of the analytics process
Proactive analytics is the logical complement to predictive analytics: when you know the course of action then you can take advantage of it, and make plans.
Let me conclude by discussing a possible alternative, namely "Preventive Analytics". It shares most of the qualities of proactive analytics. However, it conveys the idea that it is mostly useful to prevent bad things from happening; This is great, of course. But it is a bit negative, and it does not cover the opportunities for using optimization to improve things that are already doing well.
As a conclusion, here is how I like to present analytics nowadays. What's your take on it?
Proactive analytics in the analytics journey