Learning from using facial coding on 30,000 ads

by Nigel Hollis | May 07, 2018

One of the principles applied to innovation at Kantar Millward Brown is that any new technique has to be scalable, ideally on a global basis. This was why we were the first major consumer insight firm to partner with Affectiva to apply the automated coding of facial expressions into our Link pre-test. 30,000 TV ads later and that learning has scaled well beyond providing feedback on individual executions.


The facial coding employed in Link is a true System 1 measure of people’s instinctive emotional response. Changes in facial expression are recorded in real time as people watch a test video (recordings are made with the respondent’s approval) and coded automatically into 7 discrete emotions. At an overall level we find that the more an ad energizes people the more effective it is likely to be in driving sales (particularly for established brands).

Beyond this overall finding, facial coding has confirmed many of the principles learned from testing a total of over 150,000 ads around the world. The following are three findings by Anna Lakomy, Senior Business Analyst at Kantar Millward Brown, are worth keeping in mind when you develop new content for TV.

Make it easy to understand

By all means use intrigue and mystery to engage people but make sure there is a clear resolution to the story because if people do not easily understand what is going on then the ad is less likely to be effective. Videos that people report are difficult to understand evoke a lot more frowns and less attention. Holding people’s attention is particularly important for TV ads given that they rely on a narrative or story that evolves over time.

Brand status matters

Established brands elicit more smiles and as a consequence higher valence and expressiveness compared to new brands, which instead elicit more disgust, frowns, and sadness. This suggests it is harder for new brands to breakthrough as they do not have the same emotional foundation that established brands do. Users are more likely to recognize and respond to a brand with which they have a pre-existing affinity.

Testing rough executions works well

Just as with comparisons of self-reported data facial coding finds virtually no difference between testing rough and finished executions. Slightly higher valence is observed for finished films, which reflects our original finding that a finished film can improve overall enjoyment due to production values.

If you are interested in learning more about what facial coding can tell us why not listen into the webinar on May 17 in which Rana el Kaliouby, Co-Founder and CEO of Affectiva, and Graham Page, Managing Director of Offer & Innovation at Kantar Millward Brown, explore how representations of gender and ethnicity can influence advertising effectiveness and stereotypical portrayals can damage a brand's reputation. Click here for more details.

Meanwhile, what do you think of the findings from facial coding? Please share your thoughts. 

Leave a comment