What’s In Your Ad? A Machine Learning Analysis of Bank Advertisements
A PhD candidate at the University of Chicago employed machine learning techniques to analyze 10 years' worth of bank and mortgage company commercials to understand the composition of the ads. The analysis involved nearly 22,000 commercials from 2,370 distinct bank and mortgage companies.
The number of banks advertising on TV grew to more than 500 in the run up to the financial crisis in 2008, dropped significantly in 2009, then rebounded in 2010. Since then, however, the number of banks advertising declined year over year as digital advertising became more popular. The number of mortgage companies advertising exploded post-crisis, with the average number of commercials per company growing from two to three between 2009 and 2015 to equal the average number of ads run per bank.
With nearly 1 million advertising occurrences over the 10-year period, Chase generated roughly 39 billion impressions running spot TV ads. Ally was the leader in national ads over that time period, running 56 ads 122,000 times, generating 40.6 billion impressions.
Not surprisingly, Quicken Loans was the leader among mortgage companies, running more than 170 ads 1.3 million times though both spot and national ads, generating 65 billion impressions.
Machine Learning Analysis of the Ad Composition
Using machine learning techniques, three types of features were extracted from the videos: 1) text; 2) images; and 3) audio. The analysis discovered that, of the 97% of commercials with spoken words (3% were just images), the average word count was 320.
It shouldn't come as a surprise that mortgage companies mentioned mortgages, loans, and refinancing most often. But nearly seven in 10 spot bank ads talked about the bank itself--not necessarily products.
From a keyword perspective, "local" was the most-frequently mentioned keyword in bank spot ads, found in 25% of the ads. Among mortgage company ads, "cheap," "safe," and "goals" were the most-frequently mentioned keywords.
Over the 10-year period, the frequency of "checking" mentions declined (a trend I would bet has reversed since 2015) as has the use of the word "cheap" (as free checking accounts have declined in number). The use of the word "mobile," however, has risen, and is now used almost as frequently as the word "online."
The machine learning analysis was also able to determine the racial composition of actors in bank ads, and found very little change in that composition over the 10-year timeframe.
Director of Research