Confessions of a market forecaster – we might not be real wizards, but let us at least pretend we are!


This is a very odd confession, but I love market sizing and forecasting. Strong language, I know, but I just love it. I think it is the combination of attention to detail and precision calculation, coupled with the intellectual leaps of faith needed when assumption making. Starting with a blank sheet and a new market gives me the opportunity to spend hours on pure thought, and there are very few things that excite me that much (and only a couple of those are legal!) I just wanted to share a few of my thoughts about the market sizing and forecasting.

Sizing a Market – more about common sense than rocket science

There is sometimes a mystery that surrounds the market sizing process, that I don’t think is necessary. There are very few ways to size any market – and none of them are rocket science. The accuracy and quality of any market size is primarily dependant on the quality of sample being used, and the way you segment the market to create representative groups.

For example, if I were to size the market for business PCs in the US using a buy side model (sizing the market from companies actually buying the PCs), I would segment the market into company size, by industry and by geography – I am using PCs as an example because often product markets, particularly a ubiquitous one like the PC is more universally understood and has less problems. The sample in each segment would be interviewed (or use another source of primary data) to discover annual spend on PCs by each sample company. The sample portion would represent a quantifiable proportion of the segment, which if it were 10%, the market size for the segment would be 10 times the total spend of the sample. Then you add all the segments up to produce the market size. The main assumption you make is that the samples spending is representative of the segments, so it is crucial that you divide the market into segments where spending is more consistent, this is why segmenting by industry is often used as it is likely banks spend is more similar to each other than to, for example, an agricultural business.

The main source of inaccuracy is ensuring the responses are representative, which is why this method of sizing always needs a cross reference, usually with service provider revenues. This provides a check on any mis-sampling where by chance you sampled the only 30 utility companies that only upgraded their PCs once a decade. These problems appear when you have markets which are less saturated, like many services sectors, particularly outsourcing. For example, if only one utility company has signed up for an HR outsourcing deal, and your sample misses it, then your market size is shot, likewise if you stumble upon just the companies that outsourced in a market that has few deals, you’d skew the market upward.

Verifying signed deals and revenues with the suppliers in a market helps to rectify these types of errors, and can mean either using an estimation for the industry garnered from the demographics/service provider revenues, or enlarging the sample or looking for new data points to guide the market size.

This is why most market sizing models used by analysts start with supplier revenues – you build a list of known suppliers in the market, which for the PC market is fairly easy for the top 10. You estimate the specific market revenues from the financials, which you may cross check through interviews / feedback. Then, you build a probability model to fill in the gap, estimating the part of the market not covered by the providers that you know. Essentially saying what is the probability that I am missing a supplier in the top 5, top 10, top 25 and in the remainder. This is where the analyst needs to be honest about her or his knowledge of the market and  be realistic about the likely number of smaller firms. Which is why having good demographics to build out a buy-side model helps set the parameters for the market – and helps to double check the probably.

It is these things that provide us forecasters with the biggest challenge, especially building a market sizing model that works in a fragmented (ad often hard to define) market like outsourcing, managed services or process automation. However, the saving grace is deal data, this gives another way of building the market size. You can estimate the annual revenue flow from each contract, segment it into service type and produce a market size. The issue here is again one of sampling, no database of contracts is complete. So we compare the data produced from the contracts with the vendor revenues and with the survey data to complete the picture. Hopefully producing an accurate market size!

Forecasting a Market – the joy is finding that consensus of inputs to perform one trendline

Market forecasting is a similar task in many respects to market sizing, but, for me, it is where most of the joy in the process comes. I suspect this is because you can draw on more information and it is about bringing together and distilling contrasting data into a simple trend.

The simple forecast method that is typically used for technology markets, is euphemistically called an assumption-based or judgement-based forecast. Largely, I tend to rely on a mixture of methods and tend to combine them.

For an outsourcing market, I would typically use a model that predicted the sales of a product from past data, generally a time series model. This provides a base trend line for the market, typically the market would accelerate or decelerate in line with the established trend for the particular market. This essentially extrapolates the existing trends, part of this process would also include looking at economic growth and making assumptions for how a particular market is likely to grow given the economic outlook.

The next stage would be to alter the predicted growth based on likely forthcoming events in the market. This tests the analysts and my knowledge of the marketplace mapping out the likely market drivers and inhibitors, quantifying their effects, weighting them by probability/strength of impact and summing the drivers and inhibitors to produce an overall market impact. Part of this equation is also the adjusting the forecast based on survey work, providing the forecast with direct input from the customers and suppliers. In the outsourcing markets, we also sample specialist advisors and consultants which can add another perspective – broader than a single client and, typically, more realistic (or is that cynical) than suppliers.

It is the balancing of the different factors and the bringing all the data together to produce a single growth rate that I find the most satisfying. A forecast at its best is a distillation of survey work, market data and analyst expertise, the marriage of the objective and subjective in as precise a way as possible.

The Bottom-line: There’s nothing more beautiful that quantifying that view of the market at that particular point in history and occasionally getting it right

The thing I love most about sizing and forecasting a market is that balance between fact and foresight.  When people look back at historical forecasts, the size of the market at that particular point in time should always be solid, providing said forecast is doing their job properly.  The forecast was simply what we all thought the market was going to do based on the available inputs we had at the time.  It doesn’t matter if it was widely wrong, as long as the inputs were sound the the forecaster was not smoking something too dubious =)

However, if we really got (some of) those forecasts in line with what eventually happened, then we can die happy knowing we performed some genuine wizardry doing something that we loved to do. 

Posted in : Uncategorized


Leave a Reply

Your email address will not be published.