Pre-launch Forecasting

From Market Research
Jump to: navigation, search

When we predict a product’s sales before it is launched we are doing a pre-launch forecast. As a general rule, pre-launch forecasts are wrong, often massively wrong. Nevertheless, the alternative to having no pre-launch forecast is to have nothing at all, so pre-launch forecasting is popular and useful.

As discussed in decompositions, there are two main decompositions used in prelaunch forecasting:

Sales = Market Share × Market Size

and

	Intention 
Sales =     to	    ×	Purchase  × Population 
	purchase       frequency      size

The first of these decompositions can be combined with choice modeling, as choice modeling is particularly popular in that the preference share estimate that is provided can be interpreted as an estimate of market share. Thus, as most companies have a good idea of the market size, it becomes easy to use this decomposition to predict sales.

The second of the decompositions seems sensible on face value. However, it tends to produce forecasts of sales that are much too high. This is because people are unable to gauge either their likelihood to purchase or their purchase frequency. If you think about it for a while, you will realize that the questions themselves are nearly impossible to answer. How can you rate your purchase intent, for example, if you do not know what other products will be available?

Some research companies have methods for calibrating the purchase intention and purchase frequency data with the goal of making the decomposition more accurate. These are the rules of six companies:[1]

  • Company A assumed that 100% of those who say they “Definitely will buy” actually will buy (and that nobody else will buy).
  • Company B assumed that 28% of those who say they “Definitely will buy” actually will buy.
  • Company C assumed that 80% of those who say they “Definitely will buy” actually will buy and that 20% of those who say “probably” will buy.
  • Company D assumed that 96% of those who say they “Definitely will buy” actually will buy and that 36% of those who say “probably” will buy.
  • Company E assumed that the buying proportions for each of the five categories are 70%, 54%, 35%, 24% and 20%.
  • Company F assumed that the buying proportions for each of the five categories are 75%, 25%, 10%, 5% and 2%.

A key point to appreciate about these different rules of thumb is the wide discrepancy in their predictions. This highlights the poor validity of purchase intention questions.[note 1] Furthermore, all of the methods down-weight the purchase intentions, which highlights that purchase intentions are systematically over-stated.

Another problem with the calibration of purchase intent is that the extent of calibration required will likely relate to the number of competitors and the uniqueness of the product, suggesting that any simple approach to calibration is unlikely to be successful.

Despite the inaccuracy of questions asking purchase intentions, concept tests are very popular. In part this is because the diagnostics they provide (e.g., likes and dislikes) can be highly informative and in part it is because they are simple. Furthermore, they can be used to produce forecasts that are more valid than those of the decomposition. The trick is to study the results of previous concept tests. If, for example, you find that a top box score of 20% results in sales of $10,000,000 on average while 30% results in sales of $15,000,000 then you can estimate that a score of 25% will lead to sales of $12,500,000.

Researchers can often use substantially more complex decompositions than these for producing pre-launch forecasts. For example, some take into account factors like distribution, advertising expenditure, shelf placement and word-of-mouth.

Once a pre-launch forecast has been computed it should then be substituted into an economic analysis which takes into account the extent of cannibalization and the likely profit margin of the product.

Notes

  1. Moreover, the observed differences between these approaches to calibration must be less than the true variability between studies, as each of the heuristics is itself only taking into account an average level of bias.

References

  1. Jamieson, Linda F. and Frank M. Bass (1989), "Adjusting Stated Intention Measures to Predict Trial Purchase of New Products: A Comparison of Models and Methods," Journal of Marketing Research, 26 (August), 336-45.