Magnificent Models | what are you predicting?

The next agenda topic in our executive meeting was IIoT and predictive service. Avid Andy had a feeling that this was going to be a controversial   topic as there was a lot of ambiguity surrounding the definition of data science-based modeling.  This was an opportunity to jump right in and start a brief discussion on the meaning and impact that data science-based learning models could have on the organization.  He began by defining a couple distinct types of models:

1. The market today is filled with lots of misleading information, our competitors all say they can predict, we ourselves have manufacturers stating that they can also predict, and our customers have an expectation that we are predicting. My question is, predicting what? Aligned to which customer outcome? If we do not have a clear picture of all the elements combined with the customer outcome all we will get are unintelligent responses such as your automobiles "check engine soon" light.
2. What level of confidence will you need from the predictive model? For instance, let's say that you were trying to predict when the air filters at any given jobsite are going to need to be replaced. Sounds relatively straightforward and the worst thing that would happen is that you may send somebody a bit too early or a bit too late. Another scenario might be when the battery needs to be changed in a pacemaker. The predictive algorithms and the confidence level that you are going to need to have in your outputs will need to be much higher on the replacement of the pacemaker battery.  This all boils down to the level of investment on what your sensing and how you're expected outcome when leveraging that data.

Once you understand the customer's objective (s) you need to combine that with your own business objectives. Is the reason that you are using predictive modeling is to give you an edge in the market? If that is the case than the models that you choose should be single models, possibly only looking at one element and having the right to claim, rightly so, that you are predicting an outcome. However, if you are trying to turn around or mitigate risk within your organization, you may need to look at multiple pieces of equipment and their predictive models to see how, when combined with one another, create different perspectives which may alter your course of business. These are very different approaches, they all need predictive models and data, certainly investment, but their level of sophistication is vastly different. Both are valuable, the bottom line will be how well aligned the model is to your business conditions and environment.

The final thing to remember, don't shy away from collecting data, even the data which you may not see a direct use for today. This data is commonly stored in an inexpensive format often referred to as a data lake. If your ambitions are to mature your predictive models and evolve them eventually into prescriptive models, you will need all the data you can get your hands on.  Andy was beginning to lose his team members in the executive meeting and decided to summarize using a few bullet points;

• grab as much data as you possibly can
• leverage other people's models to create a holistic model of your own which can align to customer outcomes
• start small with laser focus and understand your market and the models value

Learning models, and the algorithms contained within, can drive incredible value to your business. It is key that you understand how the outputs of these data science-based objects influence action with in your current operating environments. Keep in mind that IIoT, data sciences, and even workforce sciences, are as much about the tools as they are about the cultural and market-based changes required to truly evolve your operation.

-----

Next post:  your face is our face

Thoughts?  feel free to leave replies or direct message

See all "last mile worker" posts here:  http://lastmileworkersolutions.com

-----