Notes from a pilot's kid

Who are brands really talking to on social media?


  1. How sure are you that your target audiences are those that are reading and interacting with your content on social media?
  2. You probably don’t know.
  3. We applied a network analysis methodology to find out who a brand is actually talking to online.
  4. In an example, we find that ~43% of Zipcar’s audience follow the brand not as customers, but as environmentalists, avid travelers, entrepreneurs, or other niche interests.
  5. With this information, Zipcar can better decide what content to share that its audience will care about.

Don’t assume that brands’ fans or followers on social media are also the target audience of the brands’ products and services.

The whole point of marketing on social media is the ability to:

  • reach the right people at the right time with the right content
  • have two-way conversations with a community of loyal fans of your brand

You start a social media account for your company or brand, hire a community manager, (maybe) put some money into fan or follower acquisition, and voila, you have tens of thousands of people to talk to.

Marketers usually assume that these people are all customers or fans of whatever the brand sells. It’s also logical to assume that you should create content for your ‘target audience(s)’, because that’s who would check the brand’s social media page, right?

Maybe. Probably not, in our experience. On my research team at Hill Holliday, we have questioned these assumptions and tried to figure out a way to better know the people who follow or talk to a brand. So we developed a network analysis methodology to make smarter generalizations about these important groups of people.

An example: here’s who really follows @Zipcar on Twitter

We looked at Zipcar’s ~20,000 followers on Twitter. After collecting the data and identifying communities using Gephi, here’s what the network of followers looks like:

Full Network of Zipcar's Followers

Some initial details:

  • Each dot is a Twitter account that follows Zipcar. We’re looking at 20,944 Zipcar followers as of February 2013.
  • The 498,000 lines between the dots indicate that one account follows another account, a ‘connection’.
  • The different colors are highly-connected communities identified by a clustering algorithm.
  • Bigger dots have more connections and might be influencers in general or within their community.
  • If dots are closer to each other, it means they have more connections in common than with dots that are distant.

An initial finding: most of those who follow Zipcar know someone else who follows Zipcar.

From this part of the analysis, we already know that beneath our original number – a follower count, used as a metric to measure success in social media marketing – a lot of those people ‘know’ each other. Over 70% of Zipcar’s followers have a connection with at least three other people that also follow Zipcar.

We also know that there are seven large and several medium-sized groups exist (the communities colored on the graph), and there may be something that the people in these groups have in common that explains why they have so many connections in common.

Groups of people that follow a brand and all know each other have something in common that brought them together.

One-by-one, we explored the communities to find patterns that may help us get from a context-less number – 20,000 followers – to some smart generalizations about 12 groups of a few thousand followers each.

To understand the communities better and make the generalizations, we do three things with the metadata of the accounts in each community:

  • plot each Twitter account’s location on a map to see if the community is defined by any strong geographic patterns (for this version, using the Google Geocoding API and Basemap)
  • extract keywords from the Twitter accounts’ profile descriptions to find patterns in the interests and hobbies that people often specify on their profiles (using the Twitter API and the Natural Language Toolkit)
  • read recent tweets from a subset of accounts from a community to identify patterns in Tweet content (manually)

Six communities have a location in common.

For Zipcar, we found that geography defined six of the twelve large communities; that is, these six communities showed a strong pattern in the location posted in user profiles in the communities. Their common geographies, all major cities where Zipcar has a large presence, are plotted below. Bigger dots on the plots indicate more community members in those locations.

Here they are, isolated in the network graph with their locations plotted on a map. (Bigger dots mean more accounts are from that spot on the map.)


Boston Cluster

Boston Map

Number of Accounts: 23, Average # of Connections 2129


Portland Cluster

Portland Map

Number of Accounts: 493 ,Average # of Connections: 21


Philadelphia Cluster

Philadelphia Map

Number of Accounts: 309 ,Average # of Connections: 9


Canada Cluster

Canada Map

Number of Accounts: 493, Average # of Connections: 8


Chicago Cluster

Chicago Map

Number of Accounts: 701, Average # of Connections: 15


Seattle Cluster

Seattle Map

Number of Accounts: 466, Average # of Connections 12

We can now make a generalization that many users in each community are either customers or local supporters of Zipcar in some of Zipcar’s major markets. With this assumption, we now know that there are six locations where Zipcar customers and supporters interested in the company on Twitter have hit a critical mass such that they exist as their own community.

Note: Why aren’t all of the users in the Portland, Oregon community actually from Portland? First, recall that these groups were first grouped by their connections with each other. Examining their posted locations is a secondary task to understand whether geography plays a role in why this group of people is highly connected. For example, a Portland State University graduate friend who moved to New York City may still be part of the “Portland” community in the graph, rather than the New York City community, if he or she is more highly connected with college friends than new friends in New York. Second, the ‘Location’ field on Twitter is unstructured; users can opt to provide their true location (e.g. “Portland”, or “PDX”), but it’s common for users provide something ambiguous (e.g. “on a plane”). This leads to some inaccuracies with geocoding tools, which we used to determine location coordinates.

Six communities are defined by their interests.

The remaining six large communities have some special interest in common, which we discovered by examining common terms in users’ Twitter profile descriptions, and also by reading users’ recent tweets. The interests and top keywords for the groups are shown below. Looking at the top keywords in each community’s corpus of Twitter profile descriptions, it’s easy to see what they have in common that may have caused them to be so highly connected and defined as a group by our algorithm.

Here they are, isolated in the network graph along with top keywords from their profile descriptions:

Car Enthusiasts

Car Enthusiasts

Top Keywords: car, auto, cars, news, automotive, London

Number of Accounts: 411, Average # of Connections: 7

Technology, startups, social media marketing


Top Keywords: love, media, social, marketing, life, music

Number of Accounts: 3797, Average # of Connections: 10

People who care about ride sharing, urban transportation

Urban Transportation

Top Keywords: transportation, urban, car, community, sharing, city

Number of Accounts: 1295, Average # of Connections: 16

Avid travelers


Top Keywords: travel, car, world, rental, life, lover

Number of Accounts: 621, Average # of Connections: 15



Top Keywords: green, energy, sustainability, sustainable, social, environmental

Number of Accounts: 1281, Average # of Connections: 33

People who use social networking for business


Top Keywords: social, media, marketing, business, love, life

Number of Accounts: 1639, Average # of Connections: 63

We can now make a generalization that many users follow Zipcar on Twitter for an aligned interest in improving the environment, ride-sharing, traveling, etc, and they may react positively to seeing content about these topics.

Use this type of research to know a brand’s audience better and direct content strategy at specific communities.

So, beyond this simply being nice to know, what can we do with this information?

  1. Influencers: If there is a community that the brand cares about, we can identify influential people within that community that already follow Zipcar and may be open advocates for the brand in future marketing efforts.
  2. Smarter Follower Acquisition: We can compare the audience sizes in these places with the markets that Zipcar cares about. Do the largest geography-defined communities correspond with the largest markets for Zipcar? If not, or if there are focus areas for the company that do not have a community of followers in the graph, we can work on a content strategy or more targeted advertising approach on Twitter to acquire more followers in high-priority markets.
  3. Better engagement expectations for location or interest-specific content: With the geography-defined communities we’ve identified, when Zipcar plans to publish location-specific content, such as a suggestion for a weekend trip, Zipcar can now have an expectation of how many of its followers would find such content relevant, and in the future, could re-prioritize location-specific content based on the sizes of these geography-defined communities.
  4. Smarter measurement: Measurements of engagement, click-through rates of links in tweet content, and other key performance indicators now have some added context, and can be more precise. Instead of reporting a measurement that 0.5% of Zipcar’s audience engaged with a Tweet about discounted rentals in Portland, we can now provide a conditional measurement: of those who follow Zipcar because they’re in DC, X% of users engaged with Portland-specific content (to get more precise, since we know which followers belong to each community, we could also determine if those who interacted with a DC-specific Tweet are actually part of the DC community, or if they are part of a different community).
  5. Smarter growth goals: We now have a benchmark for growing market-specific communities of followers. Depending on marketing goals, running this analysis on a quarterly basis would inform Zipcar how topic or location-specific content is growing audiences interested in these areas.