Stop saying forecasts are always wrong! It’s a pet peeve of mine because when you say this, you’re only hurting yourself and this field. Besides, you are wrong! Forecasts can be 100% correct.
Every business decision is built on a forward-looking projection which is, in essence, a forecast. So, if you are saying that forecasts are always wrong, then you are also saying every business decision is wrong. And we know that is not the case.
Saying “forecasts are always wrong” should not be an automatic response we have or an underlying assumption about this field. Why? Because when we say this and instill this belief in others, we lower the confidence in the forecasts we make, therefore reducing the effective of them. After all, if our forecasts underpin the demand plans used by our stakeholders in S&OP, and we expect them to be used, they must be credible.
Perpetuating this myth also justifies poor actions in both Demand Planners and our colleagues in other departments. As Demand Planners we use it to preempt criticism when our forecasts don’t meet that magic number, and our colleagues in other departments use it to justify their negative behaviors.
This field requires some thick skin, so why make it worse by putting yourself down and telling yourself and everybody around you that your forecasts don’t add value?
Forecasts Don’t Have To Be ‘Right’ To Be Valuable
And I get it, forecasts will rarely be 100% accurate. But that entirely misses the point.
We need to get away from this idea that the forecast number has to be correct to be valuable. Forecasting is not about hitting an exact number, but about providing insights, and for that, we don’t need an exact number; we need to understand uncertainty and the probability of something happening.
Uncertainty, far from being ‘wrong’, is important because it allows people to make better decisions because if you understand the uncertainty in your forecast, then you can plan for it better. If you’re forecasting a fixed number, you don’t have an idea of what the uncertainty is surrounding that number, meaning you cannot plan the business to manage that uncertainty.
After all, uncertainty is inherent in forecast models, the forecasting process, the supply chain, consumer behavior etc. Uncertainty is inherent in business period. And we mustn’t hide that uncertainty by saying ‘forecasts are always wrong’.
Focus On Probabilities Rather Than Single Points
And that brings us to probabilistic mindsets vs single point mindsets. Not everybody working in demand planning is comfortable with ambiguity or uncertainty – but we need to be. As forecasters we’re often predisposed to wanting fixed, absolute outcomes. But that just isn’t the case in business forecasting due to the uncertainty inherent in consumer behavior, supply constraints and a whole load of other business variables.
This where we can learn from data scientists because they think probabilistically. They think not in terms of a single number, but in terms of the probability of something happening. This kind of thinking considers the uncertainty/risk associated with a forecast which is what we really need.
So if you’re saying forecasting is always wrong, you’re lost in a false right/wrong dichotomy. The reality is that business planning is not black or white and it is impossible to say with 100% that something will happen.
Range Forecasting Beats Point Forecasting
If we roll one die, the probability of an exact outcome is equal every time we roll. Every time we roll, there is the same probability of rolling a 2, 3,4,5 and so on. But rolling two dice is a different matter. The probability of rolling two dice and getting 2 is different to the probability of getting 6, for example.
Let’s say I forecast that we roll a 7 plus or minus 2. That gives us a range that covers 5,6,7,8 or 9. That gives us over a 60% probability of our forecasted range occurring. If I forecast a 7, there’s a 16.7% probability that we’ll be correct whereas if I forecast a range, I’m right over 60% of the time. This is the difference between forecasting a point and forecasting a range. If this were a business forecast, which is more useful? The forecast range is more useful.
Why? Because with a single point forecast, you are throwing away good information. With a range forecast, we have a center point plus the top of the range, the bottom of the range, and the magnitude of the variability. Those are 4 critical pieces of information that you don’t get with a single point forecast.
You don’t know where that point fits into a range of uncertainty and you don’t know the magnitude of variability, i.e. you don’t know if you are roughly right or precisely incorrect – you could be very confident or not confident at all.
So let’s be less precise but more accurate. But how do you actually provide that range?
Coefficient of variation (CoV) or Demand Variation Index (DVI) can determine the inherent variability in a given dataset. With CoV we square the magnitude from the mean whereas as DVI takes the delta from the mean without squaring. But both essentially look at the same thing, looking at historical data and how much variability from the mean that data set has. That tells us how much variability we may see going forward. And there we have a forecast range that incorporates the uncertainty we need to be aware of to properly plan the business.
There is more to it than that, but I hope this serves as starting point to think probabilistically, get comfortable with ambiguity, and start providing valuable range forecasts that consider the inherent variability in business.
For further information, read this primer on probabilistic forecasting published in the Journal of Business Forecasting.
This article was taken from this episode of IBF On Demand, the leading podcast in the fields of demand planning, forecasting, S&OP and predictive analytics.