Research & Insights

By Seth Teicher, October 21, 2014

How Data Science Can Predict if a Controversial Ad Will Backfire

Watching Super Bowl commercials is a celebrated American tradition. With more than a hundred million viewers tuned in, the pressure to stand out on Super Bowl Sunday is sky high. That’s why many companies try to make their spots just-controversial-enough. But not going overboard is easier said than done. This tone deafness was on full display in 2011 when Groupon’s Super Bowl ad diminished the suffering of people in Tibet. The reactions from viewers were loud and clear: Groupon took it too far. Now, data scientists are finding ways to avoid this super costly faux pas.

 

On the other hand, aired during the same campaign, Groupon seems to strike the right balance between controversy and humor with this Elizabeth Hurley ad touting the benefits of a certan type of deforestation. 

 

 

Why did one ad delight while the other fall flat?

 

Diagnosing the controversialness of Super Bowl ads was the focus of the paper, ‘Race, Religion or Sex: What Makes a Super Bowl Ad Controversial?.’ Here, Rumi Ghosh and Sitaram Asur from HP Labs put forward an ingenious technique to analyze the essential ingredients of advertisements that could be used to gauge an ads’ controversialness early on in the release cycle. The lessons learned could prove invaluable to media planners when crafting marketing campaigns.

How Did They Do It?

The researchers dug into large volumes of YouTube comments for both controversial and non-controversial ads. After selecting the ads, they used CrowdFlower’s data enrichment platform to classify comments based on various factors. The enriched labels served as ground truth to train their algorithms and conduct a large scale, programmatic analysis of the comments to draw definitive conclusions.

As the researchers put it:

“We extract(ed) early YouTube comments on a collection of around 45 Superbowl advertisements. We generate(d) a comprehensive set of over 2500 semantic and lin-guistic features and evaluate(d) their efficacy in automatically detecting controversial comments.”

Controversy Gone Wrong. Analyzing Super Bowl Ad YouTube Comments

The conclusions drawn from the data were quite interesting. Here are a few of the notable findings:

  1. The number of comments: 67% of controversial ads had more than 300 comments vs. 37% of non-controversial ads. However, increased controversial content did not always mean increased conversation.

 

controversial_brand_comments
    1. The number of comments that were, themselves, controversial: With controversial commercials, the comments themselves tended to contain more controversial terms.

 

  1. The presence of certain themes: Violence issues were often detected as controversial, and so were gender, race, religion, economy, politics, and sexuality.
comment_classification_table

 

    1. The presence of religious topics: Religion triggered the most conversation out of all topics.

 

  1. The number of words in the comments: There were slightly more words used in the comments on controversial ads.
number_of_comments
  1. The type of emotions expressed: Surprise and embarrassment were more commonly used on controversial ad comments.
types_of_emoticons_used

Mitigating Risk: How Can Brands Learn from this Research?

The lesson to be learned is, yes, controversy does get more attention. If you want to stand out, alluding to race, religion, and sex could pay dividends. But it’s clear that controversial ads, if not done properly, can upset a lot of people and lead to substantial backlash. 

While analyzing YouTube comments can’t necessarily be done before an ad is public, the controversial ingredients of those comments, distilled by these researchers, could be used as a measuring stick in advance of an ads’ release.

Walking the line is inherently risky, but CMOs and ad agencies would be wise to tap data science and consider using this rubric next time they want to go down that road.