Here’s a business insight: The future is not caused by the past. So then, why are your predictive analytics facing backwards?
The domain of predictive analytics implies high precision. Yet, predictive analytics is too broad to ensure a valid process behind the forecasts—predictive methods are too easily misapplied. Decision-makers should not ascribe confidence to insights derived by predictive analytics without a meaningful understanding of typical pitfalls. Many predictive analytics methods are not suited for complex business problems. It’s far too easy to jump onboard with the latest artificial intelligence (AI) applications and business intelligence (BI) tools, and to be lured in by flashy dashboards and countless metrics. Unfortunately, misapplying predictive analytics methods simply guarantees arriving at exactly the wrong answer in a quick and stylish manner.
Typical business operations are simply not driven by the past. In fact, many events and processes may be determined by the future. The influence of one event upon another is not unidirectional when observed in actual operations. For example, we may decide to take our wind turbines down for preventative maintenance today—ahead of the normal schedule—because sustained winds are expected in the days that follow. We may elect to employ repair kits on our fleet of aircraft before observing component failures because we anticipate new failure modes due to aging thresholds expected in the future. We may intentionally over-work some of our vehicles this month because they will soon be overhauled, while others are expected to operate long-term on limited preventative maintenance. The future has an impact on today. While past events certainly influence impending outcomes, we must consider the full process and the relation of events along a continuum that historical methods simply cannot handle.
Suppose I’m looking through a narrow slit in a fence, and a snake goes by. I’ve never seen a snake before, so it is mysterious. Through the fence I see first the snake’s head, then I see a long trailing body, and then finally the tail. Then the snake turns around and goes back. Then I see first the head, and then after an interval the tail. Now if I call the head one event and the tail another, it will seem to me that the event head is the cause of the event tail. And the tail is the effect. But if I look at the whole snake I will see a head-tail snake and it would be simply absurd to say that the head of the snake is the cause of the tail, as if the snake came into being as a head first and then a tail… when we talk about different events, at different sections or parts of one continuous happening.
– Alan Watts, 1972
Over the next few weeks, we’ll examine the hazards of misapplying history-driven predictive analytics when examining complex, uncertain systems and operations, and will introduce a truly precise alternative that maps mathematical relationships between today’s conditions and future outcomes without the volumes of constraining limitations inherent to popular predictive tools. This approach generates high-resolution data from future operations and thus expands the analysis to link present conditions to the many possible future outcomes while clearly mapping event relationships, uncertainty, and risk.
- Gazing into the Rearview Mirror: Examines the fallacy of rear-facing predictions.
- Changing a Lightbulb with a Hammer: Why machine learning may not be your tool of choice.
- When Statistical Models Fail: Sampling for frequentist inferential models can quickly unravel.
- Beyond Inferential Modeling: Fooling ourselves by adjusting the significance level.
- Misleading Correlations: Fooled again, by fancy graphs, dashboards and the patterns they reveal.
- When Traditional Forecasting Doesn’t Fit: Traditional forecasting is inaccurate in the face of complexity.
- Rearview BI Misleads Business Strategy: Shaping future plans while facing backwards fails on at all levels.
- Thirty Years of Data Science Performance: Exploring predictive analysis experience spanning decades.
- Clockwork’s Differentiators: How do we deliver predictive analytics for complex business problems?