Inspire Brands Ad Creative

Which parts of an ad go with higher scores

Optimizing ad creative matters for how a restaurant company spends on advertising and how its campaigns perform. The key is knowing which specific attributes of an ad go with success on the core ad metrics. During my summer at Inspire Brands, I studied that question for quick-service restaurant TV ads.

The scores came from the ACE Metrix database. Once a spot is on air, 500 or more people watch it and complete the same standardized survey. The survey yields the Overall ACE Score and seven components: Watchability, Attention, Likeability, Desire, Change, Relevance and Information.

Defining the attributes before watching

The study covered 548 TV ads from the past year, across the quick-service category. Before I watched any of them, I sat down with the head of Demand Gen Analytics to identify and define the attributes that mattered. We arrived at 21, each written as a yes-or-no question about what the ad showed or said, such as whether it featured a close-up of the product.

Some attributes needed a rule as well as a definition. For jump cuts and split screens, we agreed how many an ad of each length needed before it counted. Then I coded all 548 ads on all 21 attributes.

Choosing the model

I asked a member of the Data Science team for guidance on the statistical approach, then chose linear regression: in R, I modeled each of the eight scores against the coded attributes, to find the attributes that were both most strongly associated with the scores and statistically significant. A separate regression of the overall score on its seven components ranked how much each one mattered.

The analysis assumed that brand does not affect the scores, and brand was blinded, so the study could say nothing about any single brand.

What the readout found

Watchability was the most influential component of the Overall ACE Score by a wide margin, with Change and Relevance tied for second. In the readout’s terms, a close-up focus on the product, and action set at the restaurant, were the biggest positive drivers of the overall score and of every component. Real people and testimonials were the attribute most strongly associated with lower values in almost every score, and humor was associated with lower values in many scores. Thirty-second ads were associated with higher scores than fifteen-second ones.

For each score, a chart set every attribute’s estimate against its statistical significance and highlighted the attributes most strongly associated with that score, and each score got its own recommendation. For the overall score, it was a 30-second ad focused on the product, with the action at the restaurant. I presented the readout in August, and it was used to inform Inspire’s creative guidance.

Cautions in the readout

I raised two cautions in the readout itself. Attributes associated with lower scores tended to appear in fewer ads, because brands are more likely to use what performs well, so those estimates rest on smaller samples. And humor is subjective and polarizing, so a yes-or-no code may not capture how viewers experienced it. I also named the brand assumption as a limitation, since anecdotal evidence suggests brand does affect scores.

The ads were coded as they aired rather than varied in an experiment, so the results describe association, not cause.

Survey response quality

Alongside the ad study, I wrote a Python script to preprocess responses to a consumer survey and identify submissions that looked automated.