Introduction

Businesses today rely heavily on models—financial models, economic forecasts, consumer behavior models, risk assessments, and even AI-driven predictive analytics—to make decisions. Yet, no matter their sophistication, models are fundamentally only approximations of reality. The saying “All models are wrong, but some models are useful,” serves as a powerful reminder that models are tools, not absolute truths.

Understanding the meaning of this phrase and applying it to business strategy can lead to more flexible and effective decision-making.

Origins of the Phrase

The phrase was popularized by George Box, one of the most influential statisticians of the 20th century. Box made substantial contributions to time-series analysis, experimental design, and Bayesian statistics. His work emphasized that models are simplifications of reality, and while they may never fully capture the complexity of the real world, they can still provide valuable insights.

Box’s statement was first recorded in his 1976 paper, Science and Statistics, where he discussed the limitations and usefulness of statistical models in scientific inquiry. His main point was that no model is a perfect representation of reality, but that doesn’t mean models should be discarded. Instead, they should be continuously refined and evaluated for practical effectiveness.

This concept has since transcended statistics and found relevance in fields such as business, economics, data science, and artificial intelligence. In our experience, it applies to both statistical models and mental models. Often it is in the latter category, the unspoken assumptions about how the market works, that come back to bite you.

An example of a mental model that outlived its usefulness was the old mantra at General Motors that “a full dealer is a happy dealer.” The wisdom of this assumption was that when dealers had a lot of inventory, they were more motivated to sell cars. This led to an emphasis on selling to dealers in a way that hurt their economics because of rising inventory carrying costs. Ultimately, it created a cycle of price discounting to dealers and then having to offer rebates and discount financing to consumers to move the less desirable cars. This proved costly for GM and put their dealers at a disadvantage at a time when other car makers were managing common pools of inventory for dealers to pull from, thereby reducing their inventory costs.

Implications for Businesses

Most organizations use models to run their businesses. These range from simple mental models like GM example above to assumptions behind financial projections to sophisticated statistical models of consumer behavior. Increasingly, they may even rely on AI-generated predictive models, and sometimes models are embedded into the product itself (e.g., search engines or self-driving cars).

As use of more sophisticated models becomes more prevalent, it is essential that executives understand and acknowledge their limitations. The business world is littered with examples where over-reliance on models contributed, at least in part, to dramatic failure. Just a few examples:

• The 2008 Financial Crisis (Risk Models) – Investment banks and financial institutions used risk models to assess mortgage-backed securities, relying heavily on assumptions that housing prices would continue to rise. These models underestimated the risk of mass mortgage defaults, leading to a global financial collapse when the real estate market crashed. Businesses that relied solely on these models were caught off guard and suffered catastrophic losses.

• Blockbuster’s Demise (Consumer Behavior Models) – Blockbuster used outdated consumer behavior models that assumed customers would continue to prefer physical DVD rentals. These models failed to account for the rapid rise of digital streaming, leading the company to dismiss Netflix as a minor competitor. By the time they recognized the shift in consumer preferences, it was too late, and Blockbuster collapsed.

• Boeing 737 MAX (Engineering and Risk Assessment Models) – Boeing used predictive engineering models to optimize the 737 MAX aircraft. However, their risk assessment models underestimated the potential dangers of the new automated flight control system (MCAS). This resulted in tragic crashes and billions of dollars in losses for Boeing when the flaws were exposed. Overreliance on the model without adequate human oversight proved disastrous.

• JC Penney’s Pricing Strategy (Behavioral Models) – JC Penney attempted to move away from discounts and adopt an everyday low-price model, believing consumer psychology models that suggested customers would appreciate simpler pricing. In reality, customers had been conditioned to seek out discounts and promotions. The change led to a major drop in sales, forcing JC Penney to revert to its original pricing strategy.

• WeWork’s Business Model (Growth Projection Models) – WeWork’s financial models projected exponential growth based on office space demand. However, they ignored market realities such as fluctuating real estate prices and profitability constraints and underestimated the impact of work from home as a competitor. Their flawed business model led to their failed IPO and billions in losses.

How can you avoid making similar mistakes?

Practical Recommendations for Business Leaders

Given the inherent flaws in models, business leaders should consider the following best practices when using models to draw conclusions and make decisions:

1. Identify the Models You are Using

Often business leaders are not even aware of the models they are using, and it may be that the team actually has multiple or even conflicting models. As a first step it’s important to acknowledge and identify the models (mental and otherwise) that leadership and the team are using in interpreting information and making decisions.

2. Maintain Skepticism

Rather than accepting models as infallible, leaders should question their assumptions and limitations. Ask: What data was used? What biases might exist in that data? What factors might this model overlook? How sensitive is the model to key assumptions (e.g., will our plan work if oil prices fall not just if they increase?)

3. Use Multiple Models

Relying on a single model can lead to narrow thinking. Businesses should cross-check different models and synthesize insights from various sources to make well-rounded decisions.

4. Continuously Update Models

Markets, technologies, and consumer behaviors evolve. Static models quickly become obsolete, so businesses must update their models regularly based on new data and emerging trends.

5. Integrate Human Judgment

While models offer valuable guidance, human intuition and expertise remain essential. Business leaders should view models as decision-support tools rather than decision-makers. This is critical in rapidly changing markets where there is often insufficient data to make statistical projections

6. Prepare for Uncertainty

Since models cannot predict the future with absolute certainty, companies must build flexibility into their strategies. Scenario planning and contingency plans are crucial for adapting to unexpected disruptions.

These lessons remind us of a dramatic experience from early in our consulting career. Thirty years ago, we were working with a consumer products company. This was a market with few competitors and declining demand, but the products were wildly profitable. For years it was a functioning oligopoly, where one firm raised prices and the others largely followed, perhaps accelerating the decline in consumer demand.

The company built a “price elasticity” model to forecast demand that was statistically strong, but they failed to recognize its shortcomings – the underlying data only included price increases, never price declines, and because competitor prices moved in lockstep, there was no way to measure cross-elasticities. The combined effect was that they were predicting brand decline rates which tended to be fairly constant – leading to strong correlations, but not a real measure of elasticity.

In the mid-1990s, quality generic and store brand products were introduced in this category at prices more than 30 percent below the national brands. Because historically absolute prices were so highly correlated with sales, they falsely concluded that these relative price differences would have little impact.

Because of this conclusion, they aggressively pursued their own private brand business, realizing too late that they were accelerating the decline of their own profitable brands. In this category, the consumer switching proved mostly irreversible (the cheaper brands actually tasted about the same, so once consumers tried them, they had little reason to switch back to the national brands). Eventually their major competitor broke tradition and lowered their prices resulting in $100’s of millions of losses for the industry. Given the supply of quality alternatives, this result may have been inevitable, but losses were accelerated by our client’s expansion of their private label business.

Conclusion

The phrase “All models are wrong, but some models are useful” is a powerful reminder that business models—whether market assumptions, financial forecasts, risk assessments, or AI algorithms—are approximations, not absolute truths. Successful businesses understand this limitation and use models as tools for guidance rather than rigid frameworks for decision-making.

By identifying models used, maintaining skepticism, using multiple models, updating assumptions regularly, integrating human judgment, and preparing for uncertainty, businesses can make smarter, more adaptable decisions. In an era of rapid technological and economic change, embracing this wisdom will be key to long-term success.

What are your thoughts? Have you encountered business decisions where models played a crucial role—whether for better or worse? word for a six-sided die. So we close with the advice to take the time and get this part of market definition right – to skip this step is playing dice with your market strategy!