I recently wrote on the topic of Forecasting.  The subject of this post is to differentiate between building an accounting oriented forecast with all the requisite statements properly footed versus creating a model that is readily usable by you and your investors to test basic cost and revenue assumptions, gauge capital needs, and guide your instinctive planning for the business.  I obviously favor the latter.

I like to build these models with one sheet that has all the input cells highlighted and shows a summary of the big picture results.  It’s easy then to manipulate the key variables and see instantly the effects on the plan.  I’m not interested in year-by-year balance sheet changes, depreciation schedules, etc.   I just want to know how much cash is required to get from here to sustainability.  I want to test out various theories regarding major items like customer acquisition cost to see what those do to my EBITDA and cash numbers over time.  And, I want to be able to show with clarity what return an investor can expect based on the desired valuation going in.

But, first let’s take a look at the really big picture:

On August 2, Dr. J. Tinsley Oden, Director, Institute for Computational Engineering and Sciences at UT Austin, gave an interesting presentation at the Austin Forum on “The Emerging Age of Predictive Computational Science.”  Dr. Oden is pictured above from the Forum site.

Dr. Oden talked about some “grand challenges” of prediction, e.g. nuclear weapon effectiveness when actual explosions are banned by treaty, the affects of new medical regimens, climate and weather changes, consequences of natural and manmade hazards, and even nanomanufacturing outputs.  In general the methods used to make such predictions have not been accurate enough and have cost plenty.

He walked us through a history of scientific methods from Plato through Aristotle, Bacon, Hume, Popper, Bayes, and Laplace.  I took good notes and will be happy to answer your questions about any of these, but I will spare you more detail in this post.  Or you can order Dr. Oden’s new book on mathematical modeling in mechanics if you’ve got an extra $101 and can wait until October.

I thought his punch lines, however, applied to the subject at hand:

• Are we trying to solve the right equations?
• Are we able to solve them correctly?
• Do we realize that all inputs are imperfect and have issues of validation and verification?
• Even hard science is subjective due to uncertainties in parameters, models, and even experimental data.
• Therefore life’s most important questions are subject to probabilities.

So, what does this mean for one who is trying to forecast a startup?  Certainly it implies that assumptions need to be clearly stated and even perhaps stress tested across a range of values.  For example, what if a salesperson can only sell 3 deals per day instead of 4?  Does it make sense to assign some probabilities to either end of a range of assumptions and see what those suggest as to the range of outcomes?

Are we paying too much attention to the easy parts like expenses and not to the difficult parts like revenue and its components of selling costs, pricing, and volume estimates?  Do we really have enough information to solve any of our equations without just compounding one guess on top of another?

Do we as entrepreneurs and investors have the ability to operate in the world of startup uncertainties?  In an ancient time (not that long after Popper), I ran a chain of hardware stores in Atlanta.   (And yes, I did go from “hardware” to “software” in the most literal sense of both, but that’s another story.)  I could forecast one year’s revenue by month for each of eight stores within a 1% margin of error.  There wasn’t much one could do to make a dramatic change in a decades old company with pretty established neighborhood markets.  In high tech I’ve never come close, but I believe I’ve missed on the good side more often than the bad side, which is about the best that can be expected.  

So, flip that coin, after making sure it’s perfectly symmetrical, and get moving on your model.  Just save those “supporting schedules” for your accountant when there are some real numbers to report.