Here’s something newcomers to predictive analytics often feel: “this forecast isn’t good enough for me to do anything with it”.

How can I better illustrate this than by an example: imagine an asset for which failures for two parts get predicted: a motor and a monitor. For both parts, the predictive analytics system is capable of predicting 40 out of every 100 failures with high accuracy.

Here are some of the questions the eventual user of of predictive analytics software should ask herself:

  • what’s the direct cost offset of acting proactively vs reacting after things have broken?
  • what is the indirect cost of the part breaking?
  • how critical is the part?
  • what would be the cost/benefit to improve this forecast beyond 40%?

Let’s first have a purely qualitative look at the above prediction. If I can accurately – with high enough confidence and enough lead time – predict 40% of all failures, what could be the impact on my business? Well, first of all, it would allow me to plan an intervention on the equipment. This means that I could do it at a time and place (in the case of moving equipment) of my choice. Second, it would allow me to ensure I have the required parts (or even the right maintenance team) on hand. Did you know that in some cases, more than half of unplanned maintenance duration is time lost waiting for parts? And third, if need be, it would allow me to reschedule my operations around the maintenance event.

Understanding the full value of ‘knowing’ is therefore very complex and many indirect factors affect it. I recently met with a client who said it would be hard to change the maintenance processes because they were already understaffed. Application of condition-based and predictive maintenance techniques is exactly what should free up capacity through optimisation schemes, pooling effects and better planning. Too much firefighting unfortunately leads to less efforts in prevention, which leads to… more fires!

Let’s now say that in this case there is no direct cost rationale for applying the above 40% forecast. Does that mean we should not react? If the machine you’re monitoring is a lathe that is part of a production line, failure of said lathe may result in waterfall delays and a large disruption of the production schedule. So, indirect costs may still warrant a proactive intervention.

Now consider the equipment being an airplane. Does the 40% prediction of a monitor warrant an immediate intervention? The answer may differ whether the monitor is a passenger IFE (In Flight Entertainment) screen or a cockpit monitor. Even in the former case, individual airline rules (some airlines value their customers more than others) may require a direct intervention. But when something (in this case: a predictive analytics solution) tells you there’s a 40% chance one of your engines may fail, things suddenly look different and doubtlessly any airline would decide to have a look at the engine before takeoff…

The conclusion therefore is that numbers need to be looked at in context as they don’t always mean the same thing and any good predictive analytics environment would allow visualisation of data relative to their impact (financial, safety risk, etc.). Bringing together all the relevant information in an easy-to-use and easy-to-interpret environment will lead to better business decisions, saving money… or even lives.