Modeling Search Data For Predictive Insights
A great presentation last week by Bill Tancer of Hitwise Intelligence has broken me out of the winter posting doldrums.
His topic on levering search data for consumer insights is one of my key interest areas on the bleeding edge of marketing research, and he did a great job of demonstrating the predictive power of modeling search data trends.
One of his most compelling talking points, reflecting the current state of the economy, showed an updated chart from this older analysis on the correlation of unemployment website visits with actual unemployment claims.
Google has gone here before as well, with their demonstration of the Google Flu Trends application, and how search data can be predictive of CDC confirmation of regional flu outbreaks by a couple of weeks.
Both of these examples illustrate how the modeling of aggregated search queries can be an incredible source of insights into consumer intent.
There are a couple of white space areas for marketing research with search data, and all worthy of further pursuit. For me these include:
- Analyzing search terms associated with digital marketing campaigns at the metro area level in order to link digital behavior to a store level or DMA based marketing mix model.
- Identifying the most predictive search terms (“grocery coupons”) that best correlate with widely tracked consumer attitude and behavior metrics (Conference Board’s Consumer Confidence Index) in order to understand where consumer sentiment is heading before the competition does.
For a better understanding of the modeling technique behind Google Flu Trends, download the PDF and Excel files that illustrate their method as it appeared in the February 19th issue of Nature.