The Beauty of Data

Carrie Fowle, Data Scientist

Fashion's Coronavirus Playbook

How to talk to customers in a Global Pandemic?  

Recently, I took part in Vogue's first Zoom conference. In one session focused on e-commerce, the panelists Virgil Abloh, Stephanie Phair, and Remo Ruffini shared their views on the impact of coronavirus to brands online. They continued to revisit the question of how to speak to customers during this strange time, emphasizing the need to connect on a human level. Intuitively, this idea makes sense: amidst a global pandemic with unemployment mounting, very few people want a luxury brand to be hawking its wares online without any sensitivity to the crisis.                      

I left the call inspired, wondering about how brands could use this uncertain environment to evolve for the better. And as a data scientist, I wondered if I could validate the idea that brands needed to connect with their customers more intimately now than ever before. I turned to Instagram to explore this question, as it is a platform where many brands connect virtually with their consumers, and pulled the three hundred most recent posts for each of 45 different brands. Using this data, I was able to demonstrate that connecting with people drives engagement and identify which brands were better (and worse) at adapting.

Identifying Major Post Categories

The first step in understanding how brands reacted to the crisis is understanding the content brands produce. To do this, I used a method called Latent Dirichlet allocation, or LDA, which groups together posts which use similar words to one another. By looking at the most common words in each group, we can get a sense of the posts that each group contains and identify topics.

We see that type one is very focused on human ideas/brand building and often contains linked materials. Like type one, type two often has linked materials, but it is far more focused on sales. Type three falls somewhere in between and serves as an intermediate catch-all.

To further understand these groups, we look at their usage of inspirational and sales language. Inspirational language is defined as language about higher-level concepts (love, family, creativity, etc.); whereas, sales language is defined as language pertaining to product (dress, purse, sale, etc.) Below we see each post types' average usage of these words both before and after the coronavirus became major news in the western world:

We see that before the crisis, type one leaned toward inspirational language, type two toward selling, and three somewhere in the middle. After the crisis, these characteristics became even more pronounced with groups one and three pushing further into inspirational language, and group two pushing further into selling.

Measuring Consumer Engagement

While it is clear that there are different varieties of post, we must also ask the question: "what do consumers think of these different types of posts?" To measure this, we must develop a metric of engagement. In many ways, consumer engagement exists on two axes: quantity and quality, so we use a metric that will capture both.

To do this, we first measure the sentiment of each comment on a scale from negative one to one. In the context of social media, we use a specialized library, VADER. This tool is built to capture the particular vernacular of the internet. It knows that 😊 is postive and 🤬  is strongly negative. It also knows that "great!!!" is very positive, while "great..." isn't.

After getting this measure for every comment, we sum the measures for all comments on a post to get the score for the post itself. By doing this, we can both capture the quality (i.e. positivity) of the engagement and the amount of the engagement. Using this score, posts with lots of positive engagement have the highest values, and posts with lots of negative engagement have the lowest.

Posts that connect on a human level resonate

Using this score, we can now compare the different types of posts over time. In the chart below, we see that around the time of Paris Fashion Week, as the reality of the situation sunk in for most of the Western World, the types of posts which drove Instagram engagement dramatically shifted. In the booming economy of the first quarter, we see that the more sell-driven posts saw higher engagement than their more concept-driven peers.

But as lockdowns became the norm, media consumers' taste changed dramatically, favoring the more humanist posts in groups one and three.

Some brands adapted better than others

With this general sense of what works in these uncertain times, we wonder which brands have more effectively navigated this paradigm shift. To identify this, we look at the difference between consumer engagements before and after the pandemic impacted daily life in the US and Europe (set as March 10, 2020, when "coronavirus" spiked in GoogleTrends data). On average, we see that there is a slight increase in engagement.

We compare the five brands which saw the largest increase to the five that saw the largest decrease (excluding Gucci whose metric is unreliable due to a large spike in engagement during Milan Fashion Week). Generally, we find that brands that succeeded in navigating the change made bolder changes in their strategy, dramatically increasing the number of inspiring words used in their posts (Dior being a notable exception).

We note that all three brands comprising Capri Holdings (Jimmy Choo, Versace, and Michael Kors) are among the five brands with the largest increases, with its original brand, Michael Kors, seeing the largest boost. Additionally, where a brand showed during fashion month generally did not impact the difference measured. We see New York, Milan, and Paris represented in the top five and all four locations represented in the bottom five.

In Summary

  • We identify major post types from fashion brand Instagram accounts and find that posts that focus on humanist themes (and not selling) perform better during the pandemic.
  • We see that brands that took bold steps to adjust their strategies performed better post-lockdown.