NEW YORK: Marketers must be careful not to over-emphasise the results drawn from data mining on the web, which can yield a biased or misleading view of actual shopper perceptions, a study has suggested.

The Boston Consulting Group, the consultancy, and Leo Burnett, the agency, compared online insights with those supplied by in-person qualitative interviews and quantitative market research.

"Although web mining is a valuable and fast-improving tool, its specific nature limits its ability to generate truly useful consumer insight. It is a complement to, rather than a replacement for, traditional research methodologies," they said.

This is because the 90% of individuals with online access in developed markets display widely divergent behaviour when it comes to discussing brands, companies and products via the net.

For example, 79% of people had posted or browsed such content in the past, while 7% read it but made no comments of their own, and 14% had never engaged in either pastime.

In the week before the study, only 27% of web users had registered such views online in the last week, 37% had perused the opinions of others but not added to the debate, and 37% did neither.

The Boston Consulting Group and Leo Burnett further revealed that 14% of the internet audience were responsible for 85% of all such posts and read 52% of this material.

By contrast, a 61% majority of consumers delivered a modest 5% of buzz about goods and services, and read 10% of the same information.

Research on healthy eating also showed the 6% of "heavy posters" uploading 80% of content were significantly younger than other active netizens. Similarly, looking "trim" was important for 70% of this group, versus 30% of the whole population.

BCG and Leo Burnett thus suggested data mining is useful for understanding topics and themes, but fell behind traditional research for determining how this translates into purchase behaviour and forming a broadly applicable customer profile.

Equally, the poor quality of much online information requires up to 75% to be stripped out prior to analysis. As such, monitoring brand buzz and the real-time response to marketing campaigns may currently be the best use of data mining tools.

"Product categories that consumers find intrinsically rewarding to think and talk about - for example, travel, fashion, and video games - may lend themselves to web mining better than, say, cleaning products or frozen food," the BCG/Leo Burnett research added.

Data sourced from Boston Consulting Group; additional content by Warc staff