Creative Strategies with Topic Data

Think about your News Feed. Whether it’s excitement over remakes of childhood classics or a solar eclipse or topics that make you go “huh,” Facebook has become the new watercooler. With over 1.5 billion people across the globe using Facebook every month to connect about topics that matter to them, it is no wonder marketers are interested in the insights that can be gleaned from these online interactions.

To further explore the opportunities that topic data can provide marketers, we recently spoke with Nikhil Nawathe, Researcher for Facebook’s Creative Shop and Marketing Science teams. Nikhil recently presented research at the Print and Digital Research Forum in London, where he shared how topic data can help inform marketers’ creative strategies. Edited excerpts of our conversation follow.

We’ve seen that people are using Facebook to connect to moments and topics that matter to them. How are marketers and how is Facebook thinking about topic data?

Nikhil: Marketers are beginning to view people’s online interactions as an untapped source that can help them more fully dimensionalize consumers beyond typical demographic segmentations.

We are interested in understanding how people connect around the moments and topics that matter. To help marketers contextualize these interactions on Facebook, we first look at the publicly available information for a topic and then layer our data sets that include public posts, Page interactions and hashtags.

How have you been focusing your analysis?

Nikhil: We have been exploring several use cases for topic data. Our research ranged from topic-level analysis for a subject like food to whether brand mentions could be used as a signal for existing brand measurement approaches. For example, could a campaign brand mention be compared to a Nielsen Brand Effect test?

What did you learn?

Nikhil: Overall, we saw that interactions on Facebook were similar to what we know happens offline. However, understanding online topic data is much more complicated than just counting interactions.

What do you mean “complicated”?

Nikhil: Think about how your own friends interact on Facebook. Some friends are active posters while others only post occasionally. We have been able to observe that some audience segments on Facebook were consistently more active than others in uploading photos and video, messaging, providing status updates, posting on Timeline, commenting and liking posts.

For example, when we analyzed interactions about food on Facebook, women made up a larger share of its volume. Because of audience biases, it was important for us to understand how much more this audience segment was talking about food compared to other topics. We call this normalization. We normalized for these biases in our analysis.

What other insights have you uncovered in your research?

Nikhil: Continuing with my food example, during the summer, we looked at food trends as part of an analysis with Creative Shop.

Our hypothesis was that, although there were many summer food trends in the US, different audience segments would talk differently about them.

We used a 2 step-process to look at the interactions about food. First, to understand broad food trends and topics, we looked at recent food topic data to uncover naturally occurring behaviors around food in the summer. In other words, we did not use a pre-determined or dictated list of keywords. Second, we looked at the pattern of interactions that occurred across ages and genders.

Food_Trends_Graphic1

Food_Trends_Graphic2

And was the hypothesis true?

Nikhil: Yes. When we split the volume of interactions by age and gender, we saw that the audience composition skewed more heavily toward Millennial women. Taking the audience skew into account, we saw that there were certain topics around which women interacted more, such as nutrition and wellness. Meanwhile, men interacted more about chefs and recipes. We also saw trends in how different age groups discussed these topics. For example, there were more interactions on the topic of chefs and recipes for younger people, while older people interacted about farm-to-table. Nutrition and wellness was a common subject across all age groups. While these are useful insights for a marketer, we can dig even deeper into the data. For example, when we look at chefs, recipes and ingredients, we can see the age pattern was consistent between women and men from 18 to 65 with a spike for mid-20- and mid-30-year-olds. But if we look at the age patterns for women and men separately, the topic of nutrition and wellness peaks in volume for young women and is fairly consistent for men across all ages.

Based on what you have observed, how would you suggest a marketer act against this insight?

Nikhil: Let’s use nutrition and wellness as an example. A marketer could use insights about this topic to inform their creative messaging and expand reach to not only men of all ages but also young women.

You mentioned that you had also examined the possibility of using topic data to measure campaign performance. What did you mean by that?

Nikhil: Marketers are always looking for ways to understand the effectiveness of their campaigns. To understand if topic data could be a plausible solution, we worked with a CPG (consumer packaged goods) advertiser to examine brand topic volumes that happened during the campaign for that advertiser.

How did you evaluate if it was an effective measurement solution?

Nikhil: We split the advertiser’s target audience into 2 groups. One half of the target audience was a part of a treatment group, which were people who were exposed to the advertiser’s ads on Facebook. The other half was included in the control group, people who were intentionally not exposed to any of the advertiser’s ads on Facebook.

The effectiveness of each of the test groups was measured by Nielsen Brand Effect, a brand polling measurement solution. We also captured related interaction volumes in the US for this analysis. Brand- or product-related interactions that included the CPG advertiser were captured 3 weeks before the campaign ran and throughout the duration of the campaign. We analyzed the interactions by first capturing posts to the Facebook product Page, second surfacing relevant public posts about the brand and third identifying mentions of the hashtag used by the campaign.

Lastly the data from these 3 steps were combined to calculate the weekly volumes of interactions related to this advertiser and its campaign.

What did you find?

Nikhil: The Nielsen Brand Effect results were statistically significant and positive for the test group as opposed to the control group. But when we looked at the brand and campaign hashtag mentions we found that the data was noisy and had low volume. In particular, there wasn’t a statistically significant difference in volume for brand mentions after the campaign was over for the test group.

Based on this case study analysis, we did not see the value in using topic data as an indicator of effectiveness. However, based on previously published research, what we have seen work successfully for advertisers looking to more accurately measure campaign effectiveness goals is to focus on business objectives like offline sales or other purchase signals.

Based on the range of analyses that Facebook has conducted with topic data, how then should marketers act on topic data insights?

Nikhil:  Keeping the nuances of the data in mind, I see that there is an opportunity for topic data to help inform how marketers develop their creative strategy. It can be used to help explore and expand their creative process. A marketer could use the insights from topic data to determine if there are audience segments outside of the core target that should also be exposed to the campaign. The insights from the data also allow marketers to develop a deeper understanding of their consumers, which can help them tailor their creative assets and messaging to better resonate with their target consumer.