ON USING GOOGLE SEARCH TO PREDICT THE FUTURE
About 15 years ago, three executives from Google visited me to ask about advertising. While they appeared to be intelligent engineering types, they seemed to know very little about advertising or marketing. They wanted to know all about it in one hour. I told them what I could in such a short time. They tapped away on their laptop keyboards like stenographers, trying to capture every word.
One of the things I told them was that Google could help companies to predict trends in their category. They tapped away. I told them that a machine learned algorithm could be tasked to identify rising and falling trends (predicting the future) in a category based on some simple rules:
- Track search word and phrase volumes that relate broadly to a category or to a topic on a daily/weekly/4-weekly basis.
- Set a minimum search volume within a country — perhaps a million searches per day or week, or a minimum per capita volume. Use this to exclude any high indexing ‘trends’ based on small volumes that are noise rather than signal.
- Set a minimum rise or fall level per day/week that must persist for a minimum number of days/weeks — perhaps for 12 weeks — before search patterns (up or down) can be determined to be a trend. The continuous change in the volume must continue in every week. Adjust the minimum rise/fall level so that it is not linear but a diminishing curve over time.
- Estimate search volume for the next day/week/4-weeks based on the persistent trend over the prior weeks. The algorithm should improve the prediction over time.
- Each day, retrospectively check the actual volume for the prior day/week/4 weeks against the estimated volume and adjust the algorithm to improve the prediction for the next day/week/4 weeks.
- If the prediction is accurate, publish the trends for companies in the category.
This machine learning method should result in an accurate prediction of a persistent trend (up or down) in a category. Publishing category trends for companies in the category would be immensely useful for marketers and business planners.
— — — — — — — — — — — — — — — — — — — — — — —