Creative Impact Analysis and Optimization for Instagram

Opinion: Working with data partner Unmetric, L2 processed more than 40,000 Instagram posts from 15 beauty brands

Brands spend time and money carefully building their Instagram presences to cultivate an engaged audience. As part of this process, brands use the standard analytics provided by Instagram and often leverage third parties to optimize the time and day of placement. They may occasionally look at other measures such as the relationship between spend and engagement to try to better understand return on investment.

None of these measures, however, helps optimize the most impactful part of the Instagram user experience: the creative.

Working with data partner Unmetric, L2 processed more than 40,000 Instagram posts from 15 beauty brands. We captured:

  • Spectrum, segmented by 25 colors and shades
  • 450 different objects or environments
  • Time and date of posting
  • Engagement rate (calculated as interactions divided by follower count)
  • Three primary types of imagery: presence of logo, presence of text, presence of human face

Note: Based on a randomized quality-assurance process, we believe that this analysis produces correct classification in 71 percent of cases.

To test a range of strategies, the brands analyzed vary in price point, product offering and digital sophistication, and the findings are, therefore, not necessarily representative of the beauty category overall.

Key findings:

  • Brighter colors tend to drive higher levels of engagement.
  • Those brands with higher levels of palette discipline, regardless of palette, tend to have higher engagement levels.
  • Text and logos within images tend to drive higher levels of engagement, while models (human faces) appear to impact engagement negatively.
  • Seasonal and product-related imagery tend to have higher engagement rates.

Looking at all posts across the 15 beauty accounts, a few key trends emerge.

First, brighter colors tend to be most impactful in driving engagement. Using color saturations as independent variables and engagement as the dependent, the “driver” of engagement represents the normalized betas from a linear regression.

Among the brands investigated, bright red, bright pink, bright magenta, bright yellow and bright blue drove engagement at the highest levels. However, many brands focus their Instagram color palettes around shades that do not drive high engagement levels. For example, while beauty brands exhibit high use of dark red, orange and dark blue, these colors typically result in low to negative levels of engagement.

Inclusion of text or logos in the image produced significantly positive results in post-level engagement. The presence of models in an image, however, produced negative results, which we believe may be a category-specific issue. The vast majority of Instagram posts in the beauty category have close-up images of human faces, and these posts may simply blend together.

To analyze a broader range of image types, we compared post engagement rates with a specific image subject and the standard deviation of those engagement rates with the image being analyzed. This allows us to assess not only whether specific images are related to higher engagement, but also how consistently they are related (a lower standard deviation representing greater consistency).

Since most posts include imagery of multiple subjects in one frame (i.e., one post can include both eyeshadow and eyelashes), we focused on the interactions between two or more image types.

The data set includes more than one year’s worth of posts and, interestingly, those related to winter and holidays did well in terms of driving engagement. In fact, posts that include a home or other type of residential property performed best. This may be a relic of the category—the beauty sector sees a consistent uptick in sales in the run-up to the holidays and marketing collateral likely follows suit.

We believe this initial analysis shows promising capabilities for deeper auditing and analysis within a category and will continue experimentation. We will specifically look to:

  • Incorporate post frequency and hashtag text analysis
  • Create a scorecard for competitive benchmarking
  • Dive deeper into outlier analysis
  • Expand brand and category list
  • Develop an optimization model

Jon Gibs is chief data officer at business intelligence provider L2.