Do you want to improve the accuracy of your demand forecasts for inventory planning? If so, you’re likely to achieve solid results with one or both of these approaches:
- Explore better forecasting methods.
- Make your forecast errors easier to manage.
This article focuses on ways to make forecast errors easier to manage. We’ll share important ways to increase forecast accuracy without creating more work.
Unfortunately, forecast error is inevitable
Probability theory tells us that when you flip a coin, the chance of it coming up as “heads” is 50%. So you’d be smart to predict 1,000 heads and 1,000 tails in 2,000 flips.
But we all know from personal experience that in 10 flips it’s possible for a coin to come up heads eight times. And it’s only a little more likely to come up heads exactly five times. This is why statistical forecasting tends to be wrong, especially when the number of occurrences is small.
Statistical demand forecasting systems use statistics and probability theory to predict future demand. They do so by projecting demand forward, based on the history of prior demand. But statistical forecasting methods are blind to the effects of the many factors that may deviate from history.
Such deviations may include:
- price changes or promotions
- short supply of products
- random changes in buyer behavior
- changes of weather and
- other factors that can occur between the moment you generate a statistical demand forecast and the time you record actual demand.
Any such factor can destroy the accuracy of a statistically generated forecast. This is why all statistical forecasting systems are built on the assumption that the forecast will often be wrong.
Exception management improves forecast accuracy
To manage inevitable forecast errors, many demand forecasting and inventory planning systems use a process called “exception management.”
(Some demand-forecasting systems ignore exception management, as if their mathematics were so sophisticated that they don’t need it.)
The exception-management process first identifies system-generated forecasts that are inaccurate. Many do so by looking at the forecast for each item in each location. Then the system compares these forecasts against actual demand for the same period.
When the system finds a significant difference between forecast and actual demand, it flags the items as exceptions.
The system then presents the exceptions to users (demand planners, forecast analysts or replenishers) whose job it is to assess factors that may have led to forecast inaccuracy.
The users may override the forecast, change the forecast model or identify the forecast as correct despite the exception. Users may also adjust the demand history if they think it’s an anomaly.
Exception management enables knowledgeable analysts to correctively nudge the system. They use their knowledge and experience to correct a forecast that would otherwise be based only on history.
Exception management comes at a cost
As valuable as it may be to invite humans fix the errors of statistical forecasting systems, we face a series of tradeoffs:
- We can waste time wading through a sea of forecast exceptions that may not need adjustment.
- We can review forecast exceptions that will have only a small effect on our demand and inventory plan.
- We can incur high labor costs for doctoring the components of our forecasts.
- We can ignore forecast exceptions at the expense of lower forecast accuracy
Studies have shown that “anomalous demand” generates more than 80% of exceptions. It’s pointless to adjust forecast errors that result from anomalous demand. They are, by definition, unlikely to occur again.
This means that as much as 80% of the time spent on forecast exceptions is wasted.
So how much unnecessary work do forecast exceptions create?
The short answer is, it depends. The labor required can be rather low or quite high, depending on the number of SKUs you carry and the number of locations where you carry them.
Forecast exceptions can add many labor hours a week
Let’s consider an example. Suppose you operate a single warehouse where you stock 40,000 items.
If your forecasting system normally generates forecast exceptions for about 10 percent of your items each week (the standard rate, based on decades of empirical evidence), that comes to 4,000 exceptions a week.
If your forecast analysts spend an average of 30 seconds per exception, that’s 120,000 seconds or 2,000 minutes, or 33 hours a week. If you employ three forecast analysts, that’s 11 hours per analyst every week.
While these exception volumes and labor hours may sound high, three analysts can easily manage the workload. The labor cost is probably worth paying for. You’re likely to achieve so much better forecast accuracy that you can improve in-stock performance or reduce inventory. Maybe you can do both.
Regardless of company size, forecast exceptions can be unmanageable
Let’s suppose you operate 700 stores, each with 15,000 items. That comes to 10.5 million SKU-store locations.
If your forecasting system still generates forecast exceptions at a rate of 10% per forecasting period, it will produce 1.25 million forecast exceptions a week.
Assuming the same level of labor productivity as before – 30 seconds per exception – the weekly labor requirement grows to about 10,400 hours.
How many replenishment analysts would you need to review those forecast exceptions? And what would be your labor cost?
True cost may be especially high for smaller organizations
In smaller organizations, the planner, analyst or buyer often has more duties and is less specialized than in bigger organizations. The effect of even a small amount wasted time may be big in a small organization – especially if it keeps inventory managers from pursing more profit-oriented activities.
Here’s the bottom line. Whether your organization is large or small, it will pay for you to use a forecasting system that…
- produces fewer exceptions
- manages some exceptions for you or
- enables you to manage exceptions much more efficiently.
With good analytics, you can manage much bigger numbers
Many forecasting systems use the conventional approach to managing forecast exceptions.
Fortunately, good alternatives are available. For one, you can use your analytics system to help manage your forecast exceptions.
Here’s how we do it at Blue Ridge. Our CLARITY analytics system enables inventory planners and forecasters to focus only on the exceptions for items that have the greatest effect on their company’s key performance metrics.
For example, you can specify that if the total annual demand for an item in a location is less than $18 in revenue (or any other value you want to use), then remove it from the list of exceptions.
You can also specify that you don’t want to see exceptions for any items whose inventory value is less than, say, $50.
By layering the appropriate filters, you can reduce thousands or even millions of exceptions to the few that most need human attention.
With this filtering capability, you can achieve higher forecast accuracy by involving your analysts only with the right items. And you can do so without sending your labor costs through the roof.
Have you found better ways to manage forecast error? If so, please share your experience.