Forecasting demand vs. forecasting "demand"

Editor's note: Wade Boudreaux is director of marketing at Danos & Curole, Inc., a Larose, La., marine contracting company.

For some companies, forecasting demand for products and services is about as easy as predicting the weather. Though there are many useful statistical methods allowing for better forecasting, sometimes it just doesn't seem to work out. In many cases, analysts may have a good handle on what their firm's key demand drivers are; however, they may be attempting to forecast the wrong factor.

For example, in 1997 the company I work for hired an MBA statistics major to develop an econometric regression model. The project's goal was to attempt to identify key variables that drive the demand for our products and services (we provide contract personnel, liftboats, construction crews, and fabrication service to offshore oil and gas customers). In this case the dependent variable used was "man-hours worked." Key variables were identified, and the model was actually pretty good. The analyst did wisely state during the executive presentation, however, that she felt the model was missing key explanatory variables, which were probably internal company data. Since good internal data was not readily available, the model was left as it was and mostly dismissed as a useable forecasting tool by executive management.

While I do believe that there are internal variables that could help better predict "man-hours worked," I also believe that the model would have been significantly more accurate if the correct dependent variable would have been used. In my opinion, this variable would have been "demanded man-hours," which would have been the number of man-hours demanded by customers that were both filled as well as unfilled.

Sometimes we get caught in a trap, associating demand with the number of products or services actually supplied. There are, however, supply problems that may have caused realized utilization or purchases to be lower than what was actually demanded. This is the case for many industries - especially the one that I work in.

In the case of my firm's industry - contract personnel services - there are huge supply problems associated with a lack of available talent required to meet customer demand. Recently, we have faced a shortage of approximately 75 people per month to fill demanded labor positions. These 75 people per month would amount to an additional 12,000 man-hours per month if the positions could have been filled. In the regression model that was created for our company in 1997, monthly figures for the dependent variable during the recent time period would have been approximately 12,000 less than actual demand for services. And isn't demand what we are actually trying to predict?

What to do

In order to begin making more accurate forecasts, company analysts should go through the effort of tracking actual product/service demand rather than merely associating it with purchases or utilization. This task is more significant for companies that regularly have supply shortages or are plagued in an industry that hinders delivery of products and services. Examples of these kinds of firms might be 1) home-construction companies, which might have to turn down work when interest rates fall and the new- build market for homes skyrockets, or 2) airline companies, which may have to cancel flights due to union activity.

The main point is: wouldn't it be better to keep historical information on what could have been as well as what actually happened? When companies try to forecast future demand, based on the current condition of key industry drivers, wouldn't they want to know what demand might actually be? The alternative to this method would nearly always produce a conservative demand outlook, especially for those industries that have large supply problems. In such a scenario, a firm might not attempt to begin beefing up supply to meet desired demand levels (that are, say, predicted to rise) and once again lose opportunities for more sales due to the unavailability of needed resources. And there is nothing more frustrating for a company than seeing opportunities for sales being lost due to product/service unavailability.

Practical use

To keep better data, simple additions to existing spreadsheets may only be necessary. Take, for example, one simple scenario that applies to my company's Marine Vessels division:

My firm owns and operates a fleet of oil platform service vessels that are rented to customers out on a daily contract rate. Price and utilization for these vessels fluctuate with the oil and gas market, which is plagued with highs and lows. The company currently has six of its 10 vessels on contract and four on the open (spot) market. My model for forecasting future demand for these four market vessels basically states that changes in demanded vessel utilization depend upon changes in the following variables: the wellhead price of natural gas; the utilization rates of shallow water drilling rigs in the Gulf of Mexico (GOM); changes in capital spending of oil and gas operators in the GOM; production levels in GOM offshore production and drilling; and the wellhead prices of WTI crude oil.

I believe that when these key industry drivers begin to rise, generally, so will the demand for our spot market vessels. To capture historical demand for utilization of these vessels, the following information is tracked in a spreadsheet, shown below.

Table 1

In the table, "On Payroll (Hours)" indicates that the vessel was working that day. "Off Payroll (Hours)" indicates that the vessel was available for work but did not work that day because there was no work available. "Unscheduled Down Time (Hours)" indicates that the vessel was not available to work that day due to unscheduled problems or events that did not allow the vessel to operate even though there was work available. "Scheduled Down Time (Hours)" indicates that the vessel was down due to scheduled repairs or events, not necessarily lack of demand. "Standby (Reduced Day Rate)" indicates that the vessel is working at reduced day rate, usually due to a transition from one customer to another or negotiated terms of a work order. "Grace Day (Downtime on Payroll)" is used for vessels under fixed-rate contract only and does not apply in this case.

Actual monthly utilization of spot-market Vessel 3 for a particular month would be calculated by adding the sums of all hours in the categories "on payroll" and "standby" and then dividing the total by the total amount of hours in that particular month. So, if the vessel worked a total of 312 "on payroll" and "standby" hours in April, then the percent utilization would be 43.3 percent (312÷720).

Actual monthly demand, however, for spot-market Vessel 3 for a particular month would be calculated by adding the sums of all hours in the categories on payroll, "unscheduled downtime," and "standby." The hours that are in the category "scheduled downtime" would not be included in calculating demand, nor would they be included in the total hours available for the month. For example, assume that Vessel 3 had the mix of hours for the month of April shown in the second spreadsheet.

Table 2

So, in this case demand for the vessel in April was actually 75 percent, as compared to a monthly utilization rate of 43 percent. What was demand for Vessel 3 in April? I say 75 percent.

Now, if it just so happens that Vessel 3 always seems to incur unscheduled down time in April due to seasonal or other unexplained forces, then we might compensate for this scenario in our demand data figures by some "seasonal" factor, if it holds true over a number of years. As a rule of thumb, however, actual demand figures rather than utilization/purchase figures should be used for forecasting, assuming that the firm has reason to believe that the product should be readily available during a particular time period. Thus, analysts who take the time to track actual demand will find that using this data actually helps to solve many of the problems associated with unavailable internal data, mainly because it compensates for most of the internal factors causing discrepancies between demand and utilization. This is especially true in industries that are plagued by supply problems.

The realization of these factors has definitely changed the way I view demand forecasting for my firm's products and services. It has caused me to reword my forecast reports to begin with, "given an operational piece of equipment," or "assuming personnel availability." There may be a lot of analysts out there who seem to be missing the mark, but are actually dead-on accurate in regards to forecasting true demand. A little extra effort on the data-tracking side may alleviate this scenario, allowing for the correct strategy to be implemented given the firm's true situation.