Managing new products is important for many firms. One challenging problem is how to predict the success of a new product, especially in their early life cycle. Using large-scale retailer data, we find the novel prediction that the early success of a new product among certain customers may have negative—rather than positive—implications for the long-term success of the product. The work identified that a significant fraction of customers systematically purchases new products that eventually flop. We term these customers “harbingers of failure”.  Their early adoption of a new product is a strong signal that a product will fail—the more they buy, the less likely the product will succeed.

The findings challenge the conventional wisdom that positive customer feedback is always a signal of future success, and the fundamental assumption in most new product forecasting models that more product sales indicate a greater likelihood of long-term success. The research suggests that not all early adopters of new products are the same. Further, it develops a pioneering approach, based on feature engineering, which extracts important behavioral metrics from a massive dataset in a parsimonious manner. The practical implication is that firms can use this approach to efficiently identify the harbinger customers through past purchases.