What are some of the advantages and disadvantages of using judgmental vs. statistical forecasting techniques?

When deploying a Judgemental forecasting approach, gathering qualitative information from multiple sources with varying backgrounds and establishing consensus is important. One method of achieving this consensus is through the Delphi technique, where insights are gathered from a panel of experts while granting anonymity to the source of the insight, largely through asynchronous methods. This eliminates common undesirable interactions between group members, such as domination or disengagement, allowing experts who may otherwise keep their opinions to themselves to provide valuable insights quickly. Consensus is reached through several phases of structured feedback until a pre-determined consensus level or acceptable range is reached.

Judgmental forecasting leverages the expertise, insights, and intuition of individuals or groups to make predictions about future outcomes. One advantage of judgmental forecasting is its flexibility and adaptability to unique situations or environments where historical data may be scarce or unreliable. Experts can incorporate qualitative factors, contextual knowledge, and subjective judgments into the forecasting process, allowing for nuanced predictions that account for complex real-world dynamics.

Judgmental vs. Statistical Forecasting **Statistical:** Pros: Objective, data-driven, repeatable Cons: Relies on historical data, may miss external factors **Judgmental:** Pros: Incorporates experience, adapts to change Cons: Subjective, prone to bias, inconsistent **The ideal approach often combines both:** Statistical methods provide a solid foundation. Judgmental adjustments can account for external factors and expert insights. **Example:** A statistical forecast predicts sales, but a salesperson might adjust it down due to an upcoming competitor launch.

Spot on thinking! Stats drive the thinking process. Industry experts add wisdom and strategic process sophistication while challenging the predictions.

Judgmental forecasting is particularly useful in situations with limited data or high uncertainty, and it allows for creativity and innovation. This is often seen in new product launches, emerging markets or long-term strategic planning. Plus points include the ability to consider qualitative factors, quickly adapt to changes not captured in statistical models, and originate novel ideas. However, it's not without flaws. Cognitive errors can impact the accuracy of predictions, biases can creep in through societal influences, and it can be both costly and time-consuming given the information gathering process from diverse sources.

Statistical forecasting is a powerful tool when used right, but it is not a silver bullet or a plug-and-play solution that any company can use with minimum investment in understanding the way it works. I have seen companies implementing an advanced suite of stat forecasts and then wonder why should they trust the automatic outcome given by the machine. Common question to address are: what algorithm should I use? based on what variables (just sales time series or multivariable? at what level of granularity (SKU, customer and day for instance introduce too much noise, at total company level is not specific enough...), how much history to consider (2+ years, last 6 months...), what horizon? is it for all products (new launches, end of life..)?

Statistical forecasting techniques rely on empirical data, quantitative analysis, and mathematical algorithms to make predictions based on historical patterns, trends, and relationships. One advantage of statistical forecasting is its objectivity and empirical rigor, which help mitigate biases, subjectivity, and cognitive limitations inherent in judgmental forecasting. By systematically analyzing historical data, statistical techniques can identify underlying patterns, correlations, and causal relationships that inform future predictions. Additionally, statistical forecasting enables scalability and automation, allowing organizations to process large volumes of data efficiently and consistently across different contexts or domains.

Short input from me, Judgemental or Statistical Forecasting depends on: • Data Availability and Quality. • Complexity of the Situation. • Expertise and Resources.

When I led an FP&A team at a public company, we experimented with eradicating judgmental forecasting. Instead, we relied on statistical forecasts, using advanced machine learning algorithms. We ran this experiment for several months. These were the results: - Statistical forecasting had a higher forecast accuracy than judgemental forecasting alone. - But, a combination of judgemental and statistical forecasting yielded significantly better results. Here is how we explained the outcome: Statistical forecasting suffers from emphasizing noise while removing all biases. Judgemental forecasting takes nuances into account while introducing some human bias. As a result, the combination of the two forms an excellent package.

Combining the power of statistical forecasting with judgements from seasoned leaders in the business is a winning combination. Statistical forecasting improves accuracy by relying on the relevant historical data and trends – helps to steer clear from overly optimistic or very pessimistic models and helps reduce human bias. The judgement leverages the unique insights and experience that leaders can provide, helping the forecasts be more accurate, especially in dealing with complex situations. Models leverage data and analytics and are good at identifying long-term trends and patterns, but not equipped to adapt quickly to sudden changes or unanticipated events. That is where having judgmental input is extremely beneficial.

Judgemental and Stat forecasts are better together, but not mixed. "Forecasters are not smarter than anyone, they just have their ignorance better structured". It is important to have a process where the stat forecast used as baseline and the manual judgemental overlays are recorded separately. In this way we can run a "Forecast value added" analysis checking if each layer of input improved or worsened the accuracy of the final forecast, gaining intuition on who's judgemental forecast improves the final forecast and who is usually more biased and off target. Without a proper Value Added analysis, Judgemental forecast loses most of its value.

In the intersection of forecasting techniques, an exciting player emerges - Generative AI. Its ability to generate novel content from existing data patterns strengthens statistical forecasting. More intriguingly, in unpredictable or data-sparse environments, Generative AI can simulate potential scenarios, complementing judgmental forecasting's flexibility. In essence, Generative AI could fuse human intuition with data-driven analytics, creating a unified forecasting approach. This intriguing blend invites us to explore further the synergies between human judgement and AI, potentially transforming forecasting into a more dynamic practice.

Combining judgmental and statistical forecasting techniques can offer a comprehensive and balanced approach to forecasting that leverages the strengths of both methods. By integrating qualitative insights from experts with quantitative analysis of historical data, organizations can develop more robust and accurate forecasts that capture a broader range of factors and uncertainties. One advantage of combining judgmental and statistical forecasting is the ability to incorporate both subjective judgments and objective data into the forecasting process, thereby reducing the impact of biases and cognitive limitations associated with each approach.

The root issue preventing reasonable levels of accuracy is statistical forecasting itself. It does not play nice with others. Its limitations will always remain limitations of whatever hybrid approach it is included in. These hybrids are generally limited to either pick one over the other or override values from the other. To achieve a symbiosis where parts strengthen each other, statistical forecasting needs to be replaced with probabilistic forecasting. This allows judgmental, machine learning, scenarios, and a probabilistic core to lead to highly accurate forecasts (95%+ versus ~50% whenever statistical forecasts are involved). The key reason statistical forecasts, like ML forecasts are precise at the expense of being accurate.

Las técnicas de pronóstico crítico son adecuadas para situaciones en las que se requiere una comprensión más profunda de las tendencias emergentes o para pronosticar eventos raros o inesperados, mientras que las técnicas de pronóstico estadístico son más apropiadas cuando se dispone de suficientes datos históricos y se requiere una precisión numérica. La elección de la técnica adecuada dependerá de la situación específica y de los objetivos de pronóstico.

When selecting forecasting techniques, organizations should consider factors such as data availability, forecasting objectives, organizational capabilities, and the nature of the forecasting problem. A hybrid approach that combines judgmental and statistical forecasting methods may be suitable for complex or uncertain environments where both qualitative insights and quantitative analysis are valuable. Additionally, organizations should invest in training, education, and collaboration to build expertise in both judgmental and statistical forecasting techniques and foster a culture of data-driven decision-making.

This has been a long-standing dilemma in the forecasting industry. The bottom line of the diacussion, however, always remains the same — judgemental and statistical forecasting go hand in hand. If for a certain case one technique is being used over the other, it is still essential for the other technique to fall in line with the observations. For example, one might consider to use judgemental forecasting when there are several macroeconomic variables involved in the forecasting process, however, such a judgemental forecast should coincide with the statistical observations from a similar situation that prevailed earlier, which makes the argument stronger.

We found that context is key. For example, how reasonable a fit is the forecast compared to others in the industry. Usually, the same external forces impact all players in an industry or sector. The difference on the players in the industry are their respective strategies, products, markets, key customers, distribution, deep pockets, training, business culture and so on. But the industry forces are the same and this should be the starting point of any forecast. Stakeholders such as banks or investors will ask how your forecast compares to others in the industry. This is a sanity check question, and it gets back to the context is key because peer comparisons are common. Forecasts should be robust enough to answer these questions.