How Data Analytics Fails Forecasters

G. Dan Hutcheson

The Chip Insider®

Strategy and Tactics: Data Analytics: How it fails forecasters.

Data Analytics: How it fails forecasters — In general, data analytics are great management tools but poor leadership tools. Data analytics models incorporate some level of artificial intelligence based on probabilistic weights derived from relationships found between data streams. They forecast well when the future is like the past. But all too often, the future stubbornly fails to follow the lessons of the past. They are artificially intelligent when real intelligence is required. The judgmental skills needed to know when data analytics will work and when it will err is the difference between managerial forecasting and leadership forecasting...

With the onset of COVID-19 in 2020, data analytics failed forecasters in this way because the artificially intelligent models had captured the high degree of correlation between macroeconomic and semiconductor data series. Remember, what AI systems cannot account for is the extraordinary event not seen before. What these models could not account for - and only judgement could - was that economic growth was decoupling from semiconductor growth because ... This is proof of the principle that 'little data' is often far more important than 'big data.' In this case the little data was the data rates that most were overlooking and the 'big data' in the lake was all the strongly correlated macroeconomic and semiconductor data. Ironically, this very problem would be the seeds for the great shortage of 2021... The great shortage of 2021 was first noticed with the auto chip shortage late in 2020 and broke into the open with a WSJ story soon after the New Year... But its seeds were planted with the COVID shutdown and the managerial forecasting approach of the Auto IC companies Then, four different things happened that could not or were not accounted for in models ... Fourth was More than Moore silicon production needs... a strategic inflection point for the semiconductor industry.