Know What You Want to Know & Know the Value of Knowing It
Strategy to surmount a key challenge
If you are serious about the word ‘strategy’ in an IoT strategy for predictive analytics then, make sure it is a strategy, not a framework, or reference design, implementation plan or technology roadmap. In Good Strategy Bad Strategy, Richard Rumelt asserts:
The purpose of good strategy is to offer a potentially achievable way of surmounting a key challenge.
The ‘key challenge’ in predictive maintenance is finding the warning signs of failure when what you are already measuring and tracking doesn’t predict failure with sufficient accuracy you can act upon it. Rumelt adds: A good strategy is, in the end, a hypothesis about what will work. Not a wild theory, but an educated judgment.
One ‘hypothesis about what will work’ is that if you look at the appropriate things at the appropriate time you can obtain insight into why systems fail and when they are likely to do so. When your visualizations fail to explain failures, then perhaps machine learning can help. Machine Learning can affirm if the data you already have can explain the outcomes you experience. Sometimes it can, but if it can’t explain with sufficient accuracy why you get the outcomes you do, then perhaps you are measuring the wrong things. The model’s inaccuracy is a result of not having data you can learn from. You are not alone. One of the conclusions in a 2012 Accenture survey, Getting Serious About Analytics: Better Insights, Better Outcomes was:
“Only 20 percent of organizations claiming to have a good performance management capability have any proven causal link between what they measure and the outcomes they are intending to drive.”
The Challenges of Knowing What You Want to Know
If you don’t have the data you can learn from, then IoT may be a way to get the data you need. Hypothesize what is causing the outcomes you are experiencing and then experiment to validate those hypotheses. But don’t pick any random hypothesis. As Rumelt suggests:
Good strategy grows out of an independent and careful assessment of the situation, harnessing individual insight to carefully crafted purpose.
The “carefully crafted purpose” in an IoT context is to acquire new data that enables new hypotheses to be tested. That is, to know what is worth knowing because if the hypothesis is true, you have a plan to change behavior and create a better outcome. This helps avoid measuring things you haven’t previously measured being added to your IoT to-do list just because they are easy things to do. And just like machine learning and analytics in general, you can start with small experiments to learn what you need to know. You may just start with one variable, in one site, or one cell, e.g. temperature or humidity, or viscosity to see if that reduces the error of the model, or at least validates it isn’t a discriminating feature.
The Value of Knowing
An IoT strategy can be directed by knowing what you want to know and the budget is set by the ROA (Return On Analytics) you expect, i.e., the value of knowing. The acquisition of new data isn’t an end in itself, it should be directly tied to improving predictions of equipment failure, or impairment of performance as measured by Overall Equipment Effectiveness (OEE). In his webinar Embedding Analytics for Growth: Creating a Data-Driven Culture, MIT’s Michael Schrage suggests that:
“Analytics are a means to an end, we are not analyzing for the sake of analytics we are analyzing for the sake of an outcome.”
Schrage wasn’t particularly concerned with predictive maintenance when he suggested this approach, but to predict failure requires an analytic model and the creation of these models for predictive analytic purposes are no different than creating models for improving productivity, increasing customer engagement, or increasing revenue. Good predictive models have a common purpose, to suggest next actions to create a favorable outcome. In predictive maintenance terms that could be to ensure the correct replacement parts are available, bring forward a planned inspection or reduce throughput to make it to the next scheduled outage. This is important because your predictive analytics priorities are driven by your analytic aspirations, which are driven by your overall equipment effectiveness plans. This implies you can borrow heavily from your analytics and data management plans and frameworks to create your IoT plans and activities. And when your experiments do provide an actionable insight and are worth scaling then the governance, retention, security, access policies and procedures from your data management plans can be recycled into your IoT production scale-out plans and activities.
New Data Providing New Insights Creates New Strength
Framing the IoT strategy to enable the ROA strategy within the context of your predictive maintenance strategy which is part of your overall equipment effectiveness strategy then you are more likely to maximize your ROIoT – Return on IoT. Rumelt explains in his book bad strategy has poor alignment, good strategy has strong alignment:
A good strategy doesn’t just draw on existing strength; it creates strength through the coherence of its design. Most organizations of any size don’t do this. Rather, they pursue multiple objectives that are unconnected with one another or, worse, that conflict with one another.
An insightful reframing of a competitive situation can create whole new patterns of advantage and weakness. The most powerful strategies arise from such game-changing insights.
Your IoT capabilities create a strength by providing new data you need to improve your predictive analytic models. Those experiments can create game-changing insights to create new predictive maintenance plans. Your new predictive maintenance plans add strength to your overall equipment effectiveness strategy. A good IoT strategy helps create new strengths with new data. It helps you calculate the value of knowing and guides you to know what you want to know.