The quest for the accurate forecast 

Published on February 28, 2025 within Article

An accurate forecast is a good forecast. Or is it always right? It's a little more complicated than that – and above all more interesting!

If we take the example of the weather forecast, most people would probably agree that it should absolutely be accurate. But what about the revenue and cost forecasts we create in our organizations and companies: What function do they actually serve? Is an accurate forecast in a business context always equal to a good forecast? 

We have asked management consultants Knut Fahlén and Hugo Bluhme a few questions to hear what they think about the purpose of the forecast and how it should be designed to create as much value as possible. 

Knut, what would you say is the big difference between a weather forecast and the economic forecasts produced in companies and organizations? 

The big difference, I would say, is that we can't do anything about the weather, which is the basis for the weather forecast. Even if we hope it will be sunny this weekend, we are completely powerless to change the situation. An accurate weather forecast would probably be considered a good forecast by most people as it gives us the opportunity to adapt accordingly.

With a forecast in a company, there is probably – at least if we extend the time perspective – an opportunity to implement activities to change the situation to what we want to happen – that is, to close the gap between the forecast and our set goal. If we take it to the extreme, few people want the forecast that shows bankruptcy to turn out to be an accurate forecast. However, such a forecast is extremely valuable for being able to act in time.

Unfortunately, it is unusual for organizations to use forecasts in this way.

So what is a good prognosis in your eyes? 

A good forecast is one that gives decision-makers insights into the future that lead to action. Because it requires that we dare to look at reality exactly as it is. We don't do that if we talk about "stretching a forecast" - then we mix up the concepts. A CASE can be stretched – a forecast should always be realistic.

How do you create a realistic forecast, it doesn't seem that simple?

Predicting the future is not easy. Especially not now in a world that is increasingly characterized by uncertainty and rapid changes in the market. That is precisely why those organizations that have the ability to use forecasting correctly have a competitive advantage.

The difficulty in most cases lies not in the methodology and technology, but in mindset. There is no “one size fits all” but all businesses need to do their homework and understand their external environment factors and internal revenue and cost drivers. In addition, management needs to accept that we actually cannot fully predict the future. It may sound obvious, but sometimes I think we reason as if management and finance functions had a crystal ball at their disposal. Read more about how Beyond Budgeting changes the way we look at forecasting here..

You talk about methods and techniques, what are these?  

To simplify, there are basically four approaches to forecasting:

1. Human judgments  
2. Historical data and statistical models
3. Driver-based models 
4. Predictive and AI-based models

In the first approach, the traditional bottom-up perspective is dominant, where the forecast is designed in the worst case as an updated budget. The problem with this method is that it is very time-consuming, but also that it involves subjective assessments that risk introducing bias, such as optimism, conservatism or groupthink. But the method should definitely not be overlooked completely. Bottom-up forecasts, where experts and employees close to the business contribute their assessments, can be used to complement data-driven models and provide a more nuanced picture.

An important principle to agree on is whether or not measures should be built into the forecast. If this is done, my recommendation is to be fully transparent about the impact the measures are expected to have as these in many cases involve uncertainty.

What we see far too little of are organizations that use data models of various kinds in their forecasting work. Here, the development and application of AI has in a short time provided even greater opportunities for realistic forecasts. Read more about how companies are implementing data-driven forecasts and rolling forecasts for better decision-making here.

How can AI be used in a forecasting context?  

There is definitely leverage with AI, says Hugo Bluhme, but it should be seen as one technology among many and the quality of the solution depends to a large extent on which AI technology and what basic data you use.

For example, using machine learning algorithms, we can improve the accuracy of forecasts by identifying non-linear relationships in large data sets. Using prediction with ML models and pre-trained neural networks via decision trees and regression models can help predict complex changes that are not obvious with traditional methods. There is also a place for language models as assistants in forecasting. They can help you test curve fitting, give you insight into your data, or help you or a developer code the solution you are looking for more efficiently.

What advice would you give to an organization that wants to improve its use of forecasts?  

I would say that the first thing is to agree on what the forecast or forecasts will be used for, Knut begins.

I agree, Hugo continues. Based on the purpose, the next step is to design the forecast model and choose suitable methods. Yes, Knut continues. And this is best done by analyzing your business logic and revenue and cost structure in the beginning, and then seeing what data decision-makers can act on. Often, a forecast designed with a hybrid model is the best solution. I completely agree, says Hugo, once you have sorted out the business logic, it becomes much easier to identify areas where investment in advanced prediction will have a direct return.

A hybrid model – what could one look like? 

It can be a forecast based on statistical models to understand trends but supplemented with driver-based models to include influenceable factors. Predictive models (e.g. AI) are added to improve accuracy and detect unexpected patterns. And last but not least, human judgment to secure strategic and market-specific insights.

Any last messages on the way? 

I want to encourage people to dare to test their skills, says Hugo. Learning by doing, quite simply. With the right expertise in place, it is no big deal to develop a few test versions that you test in a critical situation over a period of time. I also want to emphasize the danger of working too zealously with accuracy as a quality measure. A good forecast is one we act on. If the forecast for, for example, sales is too low, we will of course act and hopefully change the outcome to a better scenario. Then following up on the outcome according to the old forecast as a quality measure is doing the work a disservice.

It's a good approach, Knut continues. Changing the mindset fundamentally – the very idea of what the forecast is and should be used for – is a bigger challenge than the technology itself, and it's easier to do if the change is allowed to emerge through gradual insights and conclusions about what works.