The Cost of Error
In January 2025, Tellusant started researching the cost of forecasting errors at large companies. Here we report some findings.
Introduction
Enterprises and countries mainly create value by improving structural capital¹. This is what is beyond contributions from labor and capital. Structural capital has grown immensely in importance over the past 50 years and is reflected in intangibles.
Key elements of structural capital are often said to be:
- Management processes
- Institutional rules
- Planning systems
- Forecasting
- Pricing algorithms
- Logistics optimization
In this article, we focus on forecasting and the cost of getting it wrong
Forecasting error
“We make the strategic plan demand forecast based on what happened over the past 5 years. I wish I could say that we have something better.” said the Chief Strategy Officer of a large company.
“Our category demand grows by 1.2 times GDP growth. That is an iron law.” said the regional CEO of a multinational company.
These are quotes from our strategy development work. In our experience, most companies use variants of this, expressed in more sophisticated language, for their strategic forecasting (i.e., predictions 3-10 years out in time).
To this should be added the naïve method. It says that demand stays the same as it is today. That is, zero growth. The naïve method is useful because it is what any other method should be evaluated against.
In a significant research effort, we calculated the capital expenditure at stake when using one of the three methods:
• Future growth is predicted by past industry (or company) growth
• Future growth is predicted by GDP growth
• Future growth is naïvely set to zero
FIRST We applied the three methods to 53 industries. We used 2013-2018 to predict 2018-2023, and 2008-2013 to predict 2013-2018. This allowed us to calculate the real prediction accuracy for two periods.
We used MAPE² (Mean Absolute Percentage Error) to measure the accuracy. The outcome across the 53 industries ranged from 15% to 22% error. Superb accuracy is 5%, high is 5-10%, and good is 10-20%.
SECOND We used the list of the 40 largest companies in the United States and matched them against the 53 industries. We applied the MAPE for their industries to each of them. This meant some companies have as low as a 4% error, some as high as 40%.
We overlayed this on capex in 2018 (and 2013) to see how much capital was at stake. The graph shows this in total for the 40 companies.
62 out of 313 billion dollars of capital expenditure in 2018 is at stake by 2023. A truly staggering number. Having excellent strategic forecasting reduces this to 15 billion dollars.
CONCLUSION Having superb strategic forecasting capabilities likely is the single most important driver of shareholder value. Of course, some companies are already well on their way to achieve this, but most are not.
Why not just cut, increase or redeploy capex if forecasts are off? Because capex is not easily redeployed.³ Strategies are in large part developed to reduce this problem.
Thus, we advice companies to look over their forecasting methods for 3-, 5- or 10-year planning. These are not the time series methods used for operational planning. Instead, a more sophisticated approach is required. The good news is that the methods exist and can be put to great 𝗺𝗮𝗻𝗮𝗴𝗲𝗿𝗶𝗮𝗹 use.
¹ Called Total Factor Productivity (TFP) or Solow residual in economics.
² Technically, MAPE was adjusted to avoid right-hand skew that exaggerates the error.
³ It is not only capex that is at stake. Human resources, especially in R&D and management, are hard to redeploy if strategic forecasts are off.
¹²³⁴⁵⁶⁷⁸⁹⁰