Many people think that if they can just get the forecast right, all their supply chain problems will go away. As a data scientist who has been working on demand forecasting for companies in multiple industries for over 20 years, I’ve realized that we are never going to get the forecast precisely right.
There’s always going to be inherent uncertainty in the demand for products and services, which will cause us to get the forecast wrong. While this may sound counterintuitive and a bit pessimistic, I’d like to offer three things you can do to account for ambiguities, maximize the accuracy of your forecast and make decisions accordingly.
Think back to the beginning of 2020, for example. Who could have predicted there would be such a sudden, dramatic rise in demand for face masks and hand sanitizers? Who could have foreseen the interruptions in the supply chain that led to severe shortages of certain goods in grocery stores? And how could the travel and hospitality industries have known there would be such a monumental decline in business?
Yes, the pandemic has been an extraordinary example and something that fortunately does not happen very often. Still, many far less impactful events and activities occur regularly within supply chains and with consumer preferences, which affect forecasts: Orders get canceled, new orders come in, supplies get delayed or production problems delay shipping, to name a few.
Having accepted this unavoidable uncertainty, the next step is to develop the forecast with them in mind.
Essentially, what we’re trying to do with the forecast is:
1. Avoid ordering too much of a product or raw material, resulting in overstock.
2. Avoid ordering an insufficient quantity of product or raw material and losing out on sales due to stockout or production delays.
Too often in forecasting, we use the standard deviation of the demand — a standard bell curve — to make decisions based on the average. This works with a simple forecast. When lead times are high and demand is volatile, replenishing stock takes too long and overstocks leave you with an expensive inventory. You may be leaving money on the table.
Fortunately, there are ways to increase the accuracy of forecasts so that they are of greater value to the business.
Looking at forecasts from the perspective of stochastic analysis, also referred to as scenario analysis, allows us to better optimize inventory decisions. This is the process of analyzing future events by looking at alternative possible outcomes. It doesn’t attempt to show a precise view of the future but instead presents multiple alternative future developments. As a result, analysts can see a range of possible future outcomes and calculate the optimal inventory.
One way organizations may handle this is by doing a simulation. Start with a standard distribution curve, run 10,000 demand estimates from that distribution and calculate the profit and loss for each one. By averaging this across the range of demand estimates with constrained inventory, we get a better idea of the optimal level of inventory.
The stochastic analysis doesn’t rely on a discrete forecast (expected values) based on historical data, observations, and assumptions. Instead, it considers a range of possible developments and the likelihood of occurrence that could affect future outcomes.
Hedge funds do this very well. They use stochastic analysis to determine what the probability distribution looks like for the future price of an investment and use that distribution to evaluate the likelihood that they will make money on a particular trade. That’s how they look over a portfolio and make sure they’ve hedged in a way that they think minimizes risk and maximizes profits.
Being able to make decisions based on this intricate probability structure is more complex but will give you a better result than building to the discrete value for the forecast.
Forecast accuracy should improve with investment and time. If you’re truly committed to this, it’s important to measure actual results against your forecasts so you can continually improve your process and models.
Many organizations know that their forecast is not accurate, but they often don’t know why or by how much. It could be that they don’t measure how inaccurate their forecast is, or they may be using the wrong metrics or overly sensitive metrics. For example, they might have a bias that leads to an overestimate of demand or a bias that leads to an underestimate, neither of which is good from a forecasting standpoint.
Just having the discipline to do the required measurement can be a challenge, but it’s worth the time and effort. By measuring results more closely, we can gain a better understanding of the biases that might be built into predictive models, and how they affect accuracy.
The forecast may be sort of wrong, but we can live with that!
Getting a forecast right is something you’re just not going to be able to do. That doesn’t mean you don’t forecast. It means you understand and work within the limitations of the capability.
Remember, what we should be doing with forecasting is measuring the inherent uncertainty of demand. We won’t be able to eliminate the uncertainty; we’re trying to minimize it and manage accordingly.
This article was originally published as Demand Forecasting Is Always Wrong: Three Ways To Thrive With Forecast Ambiguity on April 27, 2021, on Forbes.