“Harbingers of Failure”: When Early Adopters Spell Doom Rather than Boon for a New Product


There’s an interesting new perspective about certain early adopters of new products:  Rather than being a predictor of success, they could well be a harbinger of failure.

Four researchers – Eric Anderson of Northwestern University along with Duncan Simester, Song Lin and Catherine Tucker from MIT – have come to this conclusion after analyzing actual purchase transaction data collected from consumers.

Their findings were published in the January 2015 edition of the American Marketing Association’s Journal of Marketing Research.

Specifically, the researchers mined a comprehensive dataset of purchase transaction information collected by a large retail chain that sells consumer packaged goods.

What the four researchers discovered was that there are certain customers whose decisions to adopt a new product are a signal that the product will likely fail rather than succeed.

Moreover, their analysis revealed that because these early adopters have preferences that aren’t representative of other consumers in the market, these adoption patterns can be isolated from those of other customers, enabling a company to predict the propensity of a new product to succeed or fail.

These “harbingers of failure,” as the researchers dub them, are consumers who fall into two categories:

  • They purchase products that are “flops” – the ones that end up failing and being removed from the market.
  • They purchase products that, while remaining available in the market, are “niche” offerings that few other customers buy.

Either way, the consumers exhibit purchase behaviors that are an “unrepresentative” subset of purchasers.

The study suggests caution when looking at aggregate positive sales figures in product test markets. Instead of considering sales figures in the aggregate, companies should drill down and study the characteristics of the buyers – whether they are ones who typically back winners or losers.

The report draws ties to several “historical” brand introductions in which purchasers of the Swiffer® mop correlated with Arizona Iced Tea® – both winning product introductions – as compared to purchasers of Diet Crystal Pepsi® and Frito-LayTM Lemonade – both of which bombed.

According to the researchers, the success of the second product (Arizona Iced Tea) could have been foretold by analyzing the sales behavior of the first (Swiffer).

Similarly, the failure of Frito Lay Lemonade could have been foretold by looking at the disappointing sales behavior of the first (Diet Crystal Pepsi).

Because of the extensive database of transactions tied to individuals that is available today thanks to bar-code scanning, loyalty programs and the like, many large consumer product firms have access to a wealth of granular data. The study contends that more people should use these data to improve their share of product introduction successes.

The full report, including research methodology and statistical analysis, can be viewed here.

The End of Privacy

An article by technology author Steve Lohr published last week in The New York Times caught my eye. Titled “How Privacy Vanishes Online,” it explores how conventional notions of “privacy” have become obsolete over the past several years as more people engage in cyber/social interaction and web e-commerce.

What’s happening is that seemingly innocuous bits of information are being collected, “read” and reassembled by computers to build a person’s identity without requiring direct access to the information.

In effect, technology has provided the tools whereby massive amounts of information can be collected and crunched to establish patterns and discern all sorts of “private” information.

The proliferation of activity on social networking sites such as Flickr, Facebook and LinkedIn is making it easier than ever to assemble profiles that are uncanny in their accuracy.

Pulling together disparate bits of information helps computers establish a “social signature” for an individual, which can then be used to determine any number of characteristics such as marital status, relationship status, names and ages of children, shopping habits, brand preferences, personal hobbies and other interests, favorite causes (controversial or not), charitable contributions, legal citations, and so on.

One of the more controversial experiments was conducted by MIT researchers last year, dubbed “Project Gaydar.” In a review of ~4,000 Facebook profiles, computers were able to use the information to predict male sexual preference with nearly 80% accuracy – even when no explicit declaration of sexual orientation was made on the profiles.

Others, however, have pointed to positive benefits of data mining and how it can benefit consumers. For instance, chain grocery stores can utilize data collected about product purchases made by people who use store loyalty cards, enabling the chains to provide shoppers relevant, valuable coupon offers for future visits.

Last year, media company Netflix awarded a substantial prize to a team of computer specialists who were able to develop software capabilities to analyze the movie rental behavior of ~500,000 Netflix subscribers … and significantly improve the predictive accuracy of product recommendations made to them.

To some, the Netflix program is hardly controversial. To others, it smacks of the “big brother” snooping that occurred in an earlier time during the Supreme Court confirmation hearings for Robert Bork and Clarence Thomas, when over-zealous Senate staffers got their hands on movie store rental records to determine what kind of fare was being watched by the nominees and their families.

Indeed, last week Netflix announced that it will not be moving forward with a subsequent similar initiative. (In all likelihood, this decision was influenced by pending private litigation more than any sort of altruism.)

Perhaps the most startling development on the privacy front comes courtesy of Carnegie Mellon University, where two researchers have run an experiment wherein they have been able to correctly predict the Social Security numbers for nearly 10% of everyone born between 1989 and 2003 – almost 5 million people.

How did they do it? They started by accessing publicly available information from various sources including social networking sites to collect two critical pieces of information: birthdate, plus city or state of birth. This enabled the researchers to determine the first three digits of each Social Security number, which then provided the baseline for running repeat cycles of statistical correlation and inference to “crack” the Social Security Administration’s proprietary number assignment system.

So as it turns out, it’s not enough anymore merely to be concerned about what you might have revealed in cyberspace on a self-indulgent MySpace page or in an ill-advised newsgroup post.

Social Security numbers … passwords … account numbers … financial data. Today, they’re all fair game.