Forecasting demand is never easy; here are some tips for making the process work for you.
It's 7:00 p.m., and your favorite restaurant has already run out of the day's special entrée. You wonder, didn't they expect it to sell? Another time, you save money at the "end of season blow-out sale." The cool summer slowed swimming suit sales, and while the store is clearly losing out, you are the lucky bargain hunter, paying less than half the original price.
Restaurants and clothiers are not the only ones to suffer from the challenges of predicting demand for goods and services. Whether selling back-to-school items or greeting cards, service calls or insurance, businesses need to be able to confidently forecast demand. It's essential: Having a good idea of what's coming may improve customer service, but it can also make a big difference in sales and profits.
Where do you begin? There are nearly as many ways to forecast as there are business owners. Some companies forecast informally - a gut feel, or a few numbers scratched on the back of a cocktail napkin. Some pay statisticians to work with expensive software to develop predictions of the future. Regardless of method, few forecasts are ever right.
The issue is not getting an accurate forecast; the issue is getting one good enough for the decisions at hand. It must be worth more in the information it provides than it costs to develop. So how do you do that? Software alone is not the answer, and may not be the answer at all. As you work to improve your ability to see the future, consider these four key steps to shaping a forecasting strategy that will work for your business:
First, understand your business.
Your understanding of your business is more important to good forecasting than almost any other factor. Don't leave forecasting to the software or the equations. Make sure you can say, "This seems reasonable" for both the forecast model used and the forecast itself.
For example, your industry may have an annual show in November that always pumps up demand. Since your historical data reflects that bump, a statistical forecast of future demand would also plan that bump. But what if that show is being moved to October next year? A good forecast would plan for the show-induced increase in sales to happen earlier. Your understanding of your business is what would tell the forecast model to incorporate that change.
Develop a relevant forecasting process.
Forecasting cannot be an event; it must be a process. To get forecasts that will help you make informed decisions, you must define steps to ensure you are using the right data, that the forecast models used make sense, and that the forecast is used in the way it was intended. You need to define responsibilities and timing requirements. It is less important where the responsibilities lie, and more important that they are defined, accepted, and executed within the timing rules of your process. Ignoring these issues makes a good forecast is a matter of luck, not planning.
As an example of "right data," consider this: Some of you use shipment or billing history as a basis to forecast the future. Ask yourself, does that data really reflect what and when the customer wanted? If you have a history of late deliveries or product substitutions, then customer order data could better reflect what the customer wanted. If you are going to use the past to predict the future, make sure the history data you use reflects the assumptions you want to make about the future.
Understand how the forecast will be used.
"Please have a forecast on my desk at 8:00 a.m. tomorrow morning." That assignment cannot be effectively accomplished until you know what kind of decisions will be made using the forecast. For example, the decision to work overtime this weekend requires a much more near term and more detailed forecast than the decision to buy land to build a new facility two years from now. The end goal of forecasting is NOT to generate a forecast; it is to support improved decision-making.
Every time you create a forecast, you must choose a level of detail and length of the planning horizon. A forecast can be for dollars, product family units, or part number detail; it can be for annual, quarterly, monthly, daily, hourly time buckets; it can look out a day, a quarter, a year. Once you understand the decision that will be made using the forecast, you can construct the forecast appropriately.
Choose an appropriate model.
There are lots of forecasting models you can use. Some are as simple as projecting this month's sales based on last month's sales, while others deploy very sophisticated mathematics. There is no reason to assume that fancier models are better. I have seen a simple 3-month or 6-month moving-average forecast beat out more complicated models many times.
If you have forecasting software, review the numbers it provides in assessing the accuracy of different models. If you don't have forecasting software, a simple spreadsheet can be invaluable. To check how well your model works, use it to predict the last 3 months and compare it to what you know did happen. Is the forecast model you used close enough for your purposes? If so, use it. If not, try another model and see how well it would have done. Keep doing that until you find a model that seems to work for you. Your knowledge of your business will be critical in choosing models that make sense to try.
Key point: Avoid models that require more mathematical or statistical expertise than your forecasters and users of the forecast have. If your business requires complex models, and some do, then make sure the appropriate personnel are trained in their use and interpretation. Don't trust the software alone.
Forecasting demand well enough to enable better planning and decisions is important to most businesses. These considerations can help you get closer.
Got a comment on this topic? Post your thoughts in the Inc. Forums.