Executive Guide to Forecasting in Manufacturing, Part 8: The Role of Human Judgment

Most forecasts in organizations start with statistical models. Advanced planning systems generate baseline forecasts using historical data, machine learning, or other quantitative techniques. In practice, however, these baseline forecasts are rarely used as-is. Planners, sales teams, and supply chain managers typically review and adjust them before they become the final forecast used for business decisions.

In Demand Forecasting for Executives and Professionals by Stephan Kolassa, Bahman Rostami-Tabar, and Enno Siemsen explore this reality. Algorithmic forecasts provide a strong starting point, while human experts often hold contextual knowledge about markets, customers, or upcoming events that models cannot capture.

At the same time, judgmental adjustments are not automatically beneficial. They can introduce biases or organizational influences that reduce accuracy. The real challenge is therefore not whether to include human judgment in forecasting, but how to incorporate it in a disciplined and measurable way.

Key Insights from the Book

Human judgment plays a central role in real-world forecasting, complementing statistical models with contextual knowledge while also introducing potential biases that must be managed carefully.

Human judgment is an unavoidable part of real-world forecasting

In practice, forecasts are rarely fully automated. Most organizations start with a statistical forecast and then refine it using human expertise. Forecasters interpret model outputs, incorporate contextual information not captured in the data, and adjust forecasts when new events or changes occur. As a result, statistical forecasts are best seen as a starting point rather than the final answer.

Human expertise can add valuable domain knowledge

Human expertise can add valuable domain knowledge. Forecasters often have access to information that models cannot easily capture, such as upcoming promotions, customer negotiations, competitor actions, or early signals from market interactions. While some of these factors can be incorporated into statistical models, others remain difficult to quantify, especially insights gained through direct customer contact. Human judgment also helps identify meaningful relationships between demand drivers that are hard to detect statistically. In this way, it does not replace statistical models but complements them by adding relevant context to the forecasting process.

Cognitive biases can distort forecasts

Despite the value of expert insight, human judgment is also subject to systematic biases. Research in behavioral economics has documented several biases that frequently appear in forecasting contexts. Some of the most important include:

• Recency bias
  Forecasters give too much weight to recent events and react strongly to short-term fluctuations.
• Illusory pattern detection
  Humans tend to see trends in random data, especially when viewing charts or visualized time series.
• Trend dampening
  When real trends exist, forecasters often assume they will weaken too quickly.
• Representativeness bias
  People expect forecasts to resemble historical demand patterns and may introduce unnecessary variation into forecasts.
• Over-precision
  Forecasters tend to underestimate uncertainty and produce prediction intervals that are too narrow.
• Hindsight bias
  After observing outcomes, decision-makers often believe their forecasts were more accurate than they actually were.
• Service-level anchoring
  In supply chain planning, forecasters may unconsciously anchor to high service level targets, leading to persistent over-forecasting.

These biases illustrate why human judgment must be used carefully in forecasting processes.

Incentives and politics often distort forecasts

Forecasts sit at the center of competing objectives across sales, operations, and finance. As a result, judgment is frequently influenced by:

• target enforcement from leadership
• hedging to secure capacity or inventory
• sandbagging to lower sales targets
• second-guessing algorithms
• withholding information across functions

When forecasts become negotiation tools, they lose their role as an objective view of expected demand.

Measure judgment using Forecast Value Added (FVA)

An important message from the chapter is that the value of human intervention should be measured rather than assumed. One method for doing this is Forecast Value Added (FVA) analysis. This approach compares forecast accuracy before and after judgmental adjustments.

Research shows an interesting pattern :

• Large and infrequent adjustments often improve forecast accuracy because they typically reflect genuine new information.
• Small and frequent adjustments often reduce accuracy because they are more likely to reflect bias, noise, or organizational friction.

This suggests that human judgment should ideally be applied selectively and with clear justification.

Quantics Perspective

At Quantics, we strongly believe that expert knowledge matters.
A forecast should not ignore what commercial teams, planners, and market-facing experts know. Many relevant demand signals are not visible in historical data alone, which is why human input has an important role in modern forecasting.

At the same time, we see consistently in practice that human input creates the most value when it is structured, transparent, and measurable.

Balancing Systems and Expertise

We do not see forecasting as a choice between algorithm and expert. Instead, it is a process where:

• the system provides a robust and scalable baseline
• experts contribute relevant business context
• all adjustments are tracked and assessed
• performance is monitored to ensure interventions improve the forecast over time

Governance Turns Judgment into Value

Without proper measurement, human input can introduce bias rather than improve outcomes. When monitored through forecast accuracy metrics, override tracking, and value-added measurement, it becomes a controlled source of improvement.

The objective is not to reduce human involvement, but to focus it where it adds value and create feedback loops that make performance transparent across the organization.

Enabling Effective Collaboration in Practice

To support this in practice, Quantics provides unique capabilities tailored to enterprise manufacturing.

Our solution enables:

• efficient forecast overrides across all levels
• structured inputs through comments and reason codes
• full transparency by logging all adjustments
• clear visibility through forecast value added analysis

In addition, capabilities such as scenario planning, risk levels, and customer-provided forecasts help incorporate forward-looking information into the planning process.

Our goal is to help manufacturing organizations establish a controlled, transparent forecasting process that drives planning excellence.

Practical Takeaways

For executives, the message is straightforward. Human judgment has an important place in forecasting, but it should not be left unmanaged.

The most effective organizations:

• treat the statistical forecast as a baseline
• allow expert adjustments when new information exists
• measure the impact of those adjustments
• ensure incentives do not distort forecasts
• build governance around forecasting interventions

Forecasting should not become a contest between intuition and algorithms. Instead, it should be a structured process that combines both sources of insight.

In Conclusion

Forecasting is both a technical and a human activity. Statistical models provide powerful tools for identifying patterns in data, but they cannot capture everything that happens in markets and supply chains. Human expertise remains valuable when it introduces new information that models cannot access. At the same time, human judgment is subject to biases and organizational pressures that can degrade forecasts if left unchecked.

The most effective forecasting processes therefore combine both elements: a strong statistical foundation, supported by structured human input and continuous measurement of performance. When organizations strike this balance, human judgment becomes not a source of noise, but a powerful complement to data-driven forecasting.

At Quantics, we support enterprise manufacturing organizations in building exactly this balance. By combining advanced forecasting methods, structured workflows for human input, and full transparency on adjustments, we help teams turn forecasting into a controlled, scalable process that improves decision-making across planning, production, and supply chain operations.

Disclaimer

In this post, we share highlights from the book “Demand Forecasting for Executives and Professionals” by Stephan Kolassa, Bahman Rostami-Tabar, and Enno Siemsen, together with our own reflections on how these ideas apply to today’s manufacturing supply chains. This is not a replacement for the book, but rather a guide to spark thought and discussion. For a deeper dive, we encourage you to explore the book itself - available here.

FAQ

1. How do you actually know if a forecast override made things better or worse?

The only reliable way is Forecast Value Added (FVA) analysis, which compares forecast accuracy before and after a judgmental adjustment. If accuracy improves, the override added value; if not, it likely introduced bias or noise. Tracking FVA by planner, product, or adjustment type turns "we always improve the forecast" from an assumption into a measurable fact.

2. What's the difference between a demand forecast and a sales target?

A demand forecast predicts what customers will actually buy; a sales target is a goal leadership sets to drive performance. The two get confused when a forecast is inflated to justify a budget, or deflated to make a target easier to hit. Keeping them separate is key to accurate planning.

3. Should sales reps be allowed to override the system's forecast?

Yes, but only within a structured process, not as unrestricted, unrecorded edits. Sales teams often know about upcoming orders, negotiations, or churn risk that models can't see, so their input has real value. The safeguard is simple: require a reason code for every override, then review its impact afterward.

4. How often should you manually adjust a statistical forecast?

Less often than most planning teams assume. Large, infrequent adjustments tend to improve accuracy, since they usually reflect real new information. Small, frequent tweaks do the opposite, adding noise or bias instead. A good rule: override only for a specific, documented reason, not on a routine schedule.

5. Does AI-driven forecasting reduce the need for human judgment?

It reduces routine manual work, not the need for judgment itself. AI models handle pattern recognition, seasonality, and baseline accuracy at scale, freeing planners to focus on what models can't see: customer negotiations, promotions, market shifts. The goal isn't removing people from forecasting, but redirecting their time toward fewer, higher-value adjustments.

‍