Part Three – Estimating Demand Based on Data

 

A few weeks ago, we introduced a series on how the “maximum capacity utilization” edict has potentially driven profits DOWN in several industries. We outlined four tactics that can help you move away from this philosophy:

• Improve communication between supply and demand groups
• Estimate demand based on data, rather than roll-ups of sales forecasts
• Consider total costs of your supply chain, including moving output to higher-demand areas
• Using software to optimize your business profitability

This week, we’ll focus on “Estimating Demand Based on Data.”

In my earlier days prior to pricing, I did a lot of supply chain work. In the mid-to-late-nineties, the concept of optimizing your production capacity existed, and tools like i2’s Rhythm product, were forerunners in the optimization world. These were great products, but had one big weakness – the demand-side inputs were largely roll-ups of sales forecasts. While this has been an acceptable way of business planning for quite some time, and certainly can improve your business, it might not provide sufficiently granular information to deliver the best results.

There really is no need to rely on this roll-up, though, as you are sitting on a treasure trove of big data – your demand history for the past several years! By applying statistical techniques such as regression and ARIMA models, and exponential smoothing algorithms, you can get a good sense for demand, based on your historical data. Also, applying unconstraining techniques will help estimate the larger market demand, even if you do not have information about business you failed to win. PROS has found the application of more advanced models, like Bayesian forecasting, is called for in some cases.

Since these techniques can be complex, you should maintain diligence around the big tradeoff in demand forecasting – Occam’s Razor vs. improved accuracy with complexity. I subscribe to the former.

It’s also very important to understand how seasonality affects your business. For example, one of our customer’s businesses experienced high demand both on weekends and at the end of the month. When the two coincided, demand really went up. Some of our statistical methods did not work as well on this model, so our team developed a simple forecast based on year-over-year demand with some growth that took these unique factors into account. The result: A simple, easy-to-understand and accurate forecast that drives the optimization process.

Getting a demand forecast takes some time, and often requires some specialized knowledge. The manufacturers I know who have made an investment in understanding their demand have seen it pay off. Once they have a good demand forecast, their teams begin to make better decisions today, based on what they are confident will happen tomorrow. Your sales team may not be so inclined to give that customer the order, if they know a few more customers who’ve not yet ordered will likely do so in the next few days.

How can you use your big data to improve your demand estimates? If you’re looking for guidance, let us know. Our experts stand ready to help.

Doug