As the rate of data collection has skyrocketed over the last several years, so too has the quest to extract meaning from every single point of data.
And while all data tells a story, there are criteria every story must meet in order to convey a meaningful lesson. No lesson? No reason to make an operational change. No operational change? No chance of better performance and profitability.
Introduction, conflict, and resolution
All successful stories contain the three elements of introduction, conflict, and resolution. The same is true for the stories told by analytics. In order to properly understand what the data is telling us, we need:
- Introduction, or context: What does this data mean? Why are we collecting these numbers? How relevant are they to the goal we are trying to predict?
- Conflict, or failure: What are the criteria for failure in the data we are measuring? How will we know when to sound the alarm?
- Resolution, or success: Conversely, what are our best results? What ideal conditions should we be trying to emulate?
Without these factors, it’s impossible for us to get the proper meaning out of our data.
Example: A large manufacturer was trying to predict mechanical failure of a piece of equipment. They had done their homework, collecting significant data involving run rate, maintenance, and more.
There was just one problem – the part had never failed before! And without knowing the conditions that led to failure, it was impossible to extrapolate any meaningful prediction about when it would occur.
Filtering and distilling data
Once we have collected sufficient data to support our goal, we can start to look for meaning. One of the most important steps is to look for signals; i.e. features of our data that share a high degree of mutual information with the variable we are trying to predict.
When two sets of data share mutual information, it means that a change in one is accompanied by a change in the other. This does not suggest any specific pattern – linear, exponential, or otherwise. But it does means that a change in one variable is accompanied by a change in the other. Knowing the first of these pieces reduces our uncertainty of the second.
Signals alone are not sufficient to analyze our data. But they are an important indicator that our data contains meaningful patterns.
Once we have recognized these signals, we can often identify the three or four most important variables impacting the result that interests us. From here, we can take the most direct path to identifying what conditions lead to failure or success – without getting distracted by peripheral factors with a much weaker impact on the goal we want to measure.
Moral of the story
Big data is a valuable tool. It can reveal meaningful insights about your business. But it’s just a tool.
Analytics is not inherently useful unless the data encompasses the failures and successes we want to measure. As data takes its place as the Next Big Commodity, we need to start thinking about the most efficient ways we can use it to drive our decisions.
In other words, it’s not about the data and the cool tools you can use to look at it. It’s about the result we want to improve or refine. In manufacturing, it’s about the chosen operational improvement that leads to better business results.
Decide on your goal. Find the relevant data. Focus on it.
Now you’re on your way to better business decisions.