The next in a series of blog posts for my Decision Modeling class examining the use of the tools learned in class as applied to the outside world. This one is one how some of the techniques learned in class might apply to my current job.
One of the potential uses of the tools learned in this class that I’ve been mulling over in my current job is the potential usefulness of Monte Carlo simulation to improve Revenue Forecasting.
But first, a little background: I’m a financial analyst for a teleconferencing and webcoferencing company. I work on the expense-forecasting side of the equation, but the revenue forecasts sit right next to me, so I overhear a lot about their challenges in developing an accurate forecast. There are a couple different ways to categorize our largest revenue sources. First there is the difference between “attended” and “unattended” calls. “Attended” calls require an operator to be on the line during the conference call to provide assistance and help handle Q & As. A typical example of this kind of call would be an Investor Relations call, in which the a client company’s CFO, CEO, or other executives present information to stock analysts. “Unattended” calls don’t require an operator, and tend to be shorter calls. Another method of dividing the revenue up is based on the size of the customer: Small and Medium Businesses (SMBs) are more numerous, but each individually generates a relatively low volume of traffic. But much larger corporate accounts can each generate a very large volume of traffic. There is a lot of empirical, historical data on all of these factors that we could use to build a more accurate revenue forecasting model.
There are several uncertainties involved in forecasting revenues from these sources. The first is the total number of SMB clients currently using our services. On a monthly basis we may cycle through SMB clients as the sales team signs new clients up and old clients fall off or switch to competing services. The second is the total number of minutes each SMB client might be expected to use in a month. Without having looked at the data, I can only guess that both of these will likely form some sort of normal distribution. But other analyses would need to be done on the empirical data to look for trends over time and other clues that might help us refine our model. Finally, there seems to be some variability in the rates that SMBs may be charged for these services.
The second set of uncertainties concern the large corporate client. In any given month, there is a risk that we may lose any of our corporate clients to a competitor. The exact risk might be uncertain, but some idea of that risk might be gleaned from examining the historical data to get an idea of the turnover. These are usually rare because long-term relationships with large clients are bound by contracts that may have penalties for breaking the contract. The next uncertainty concerns the probability of whether we will “win” new large corporate clients. Usually our revenue forecasting team is apprised of current deals in the hopper, but often are given unrealistic assessments of the probability of a deal being signed, with the sales group often reporting “We are definitely getting this sale” only to find, all too often, that we will not be getting the sale after the forecast including that sale has already been finalized. The final uncertainty with large clients is the volume they will generate. Usually, these contracts include a minimum obligation of minutes to be used by the clients, but large clients don’t always meet these minimum monthly usage obligations. Sometimes, they also may greatly exceed these minimum obligations.
With all of these uncertainties playing a role in our revenue forecasts, it starts to seem absurd that we are using point estimates. Clearly, a Monte Carlo simulation would be the superior method of providing useful data for Revenue Forecasts. (Somewhat less so for most of the cost forecasting that I do. Some variable costs are driven by our traffic volume, but the Revenue Forecasting team handles these costs. Most of the rest of the Sales, General, and Administrative costs are theoretically controllable, though experience has shown this not to really be the case either.)
But then I started to consider what value this would really add, based on how this superior forecast might influence decision making. To some degree, a better long-term revenue forecast might improve long-term capacity planning, but generally our forecasts are focused on short-term fluctuations and short-term accuracy more than for the long term. Which lead me to wonder what decisions, if any, our forecasts are being used to inform. Other investment and R&D decisions are primarily made based on the CEO’s “vision”, and these forecasts seem to have very little to do with that. In fact, the most prominent use of our regular monthly forecasts that I have seen are two: monthly departmental meetings with the Finance group in which the CFO or Finance Directors rail and bemoan the lack of control various departments have on their finances, and point to dismal revenue forecasts as a sign that these departments need to bring costs under control, and our own investor relations calls, in which the executives attempt the opposite spin on the numbers to try to talk up our long-term profitability to stock analysts. In that case, the decision that is being driven being driven in part by our forecasts, ultimately, is whether stock analysts will recommend to investors either to buy, sell, or hold our stock.
Given the uncertainties inherent in what value can be gained from a more textured and nuanced revenue forecast (or expense forecast, for that matter) that is more accurate than a simple point estimate, it would be hard to recommend to the company that they invest in purchasing licenses for a risk analysis module such as @Risk. It then becomes clear that what the company (particularly the finance department) really needs to do is invest some time and effort in analyzing what are the objectives of our regular monthly forecast, what we will be trying to achieve with them, and why – and to communicate those objectives clearly throughout the company and especially throughout the Finance department. Once that quality discussion has been had, the question of the accuracy of those forecasts can be re-addressed.