When is it possible to predict a product’s success? How you answer this question may be the most important factor in how you design your product development process — and, ultimately, in whether your business succeeds or fails.
Is market performance predictable for a specific product or class of products? If so, we should use a set of processes when designing and launching businesses that are geared for prediction — ask the right questions, perform the right analyses, plan production and supply chain for predictable variations on projected sales. And, if prediction falls outside of an acceptable range, we should hold our teams accountable for predicting poorly.
In contrast, when performance cannot be predicted, a process geared for prediction makes no sense. Outcomes will usually be much worse than predicted, and we will waste design, production, and distribution resources. Worse still, we will punish employees even if they do everything right, and create incentives to avoid unpredictable products, even though these products are essential for growth. And products that are more successful than predicted can be even worse: we’re unable to deliver to customers, both losing potential revenue and compromising our reputation for other products as well.
For products that cannot be predicted, we should focus on recognition — on trying to identify as quickly and cheaply as possible whether a product is succeeding when it’s actually introduced in the market, and create production and distribution processes that are flexible enough to adapt to our recognition of success or failure. In this case, accountability is not for correct ex-ante prediction, but for fast learning and contingent execution.
Fundamentally, these are two different ways of doing business — processes that excel in one context fail miserably in the other — so the crux is trying to identify when prediction is possible. To do that, we must first understand how prediction works.
Though there are multiple types of prediction, the gold standard is the prediction of precise outcomes. Such predictions provide both magnitude and time horizon (or at least one of the two); finance tools like the CAPM and Black-Scholes formulas give us partially-specified predictions — value without time; Newtonian physics gives us a “law”, fully-specified predictive theory, completely deterministic and accurate for the directly observable world.
In the context of a new product, only this type of precise prediction yields enough information for us to plan on. In order for us to plan for a product launch, we need to know the sales and the time horizon — without both, our production and distribution schedules will go hopelessly awry.
The question is whether our ability to predict the success of new products reaches this threshold of precision. An easy way to begin addressing the problem is to examine the accuracy of your projections. One study, of an industrial conglomerate, found that estimates of cost-saving initiatives, line extensions, and new products had average accuracy of of 1.1, 0.6 and 0.1, respectively, suggesting that market risk is the major driver of our inability to predict.
Look at the variance of your new-market products. If you’re typically off by an order of magnitude but using processes that assume accuracy, scrap your current process and recreate it assuming unpredictability. First recognize if the product is solving a real problem and people will buy it, then figure out how to scale production and distribution in situations of success.
We have a set of processes created for recognizing success in inherently unpredictable projects — lean product development, customer discovery, discovery-driven planning — yet we consistently fail to apply them. As a result, our predictions for new products are either too high or too low, and we waste resources, customer goodwill, and employee motivation.
Rather than merely hoping our predictions will be better next time, we must accept that we can’t predict new product success, and refocus on recognizing as quickly as possible whether a newly introduced product is working.