How and why we created 3 LinkedIn dashboards

Social networking is a useful tool that helps me share my experience and talk about analytics, cases, and insights with my colleagues and potential clients. And, like with any other tool, you need to manage its efficiency. This is what I keep my eye on:

• Audience growth and observe the dynamics.?
• Trends, popular posts, and topics. There are a lot of subjects I’m interested in: I’m ready to talk about data, analytics, life hacks, and compelling cases from my experience. Still, it’s important to consider my followers’ interests rather than just mine.
• Staying on top of the numbers so that I can react swiftly to drops in engagement rates and reach.
• Getting the opportunity to formulate and test hypotheses. How often should I post? Which time is optimal for posting? What kind of design is the most engaging??

I have three LinkedIn accounts: my personal profile, the Valiotti Analytics page, and the Cyprus Data Community page. Of course, I could collect all the metrics manually, but that would take a ridiculously inefficient amount of time and effort. The more social networks there are, the more it takes to gather the data. Plus, many metrics crucial to my goals are much more convenient to track through visually clear charts rather than numbers under posts, or simple tables.?

And after all, aren’t we analysts? Why would we do something manually if we know how to automate this stuff?

So our team faced this task: to build a user-friendly, fully automated tool for monitoring the necessary metrics from 3 LinkedIn profiles. That’s when we decided to create a dashboard.?

Our approach

To address this, we adopted the workflow we’d already established while completing similar tasks for our clients.?

• We gathered a team of a BI expert and data engineers. Usually, dashboards like this are created by just one person. But…
• We had an opportunity to assign several employees to the task to ensure speedy completion.??
• The team familiarized themselves with available data, the LinkedIn API, and the statistics the service provides. Never underestimate this step: before the first interview with a client, always agree on what you’re gonna be dealing with. It will help you to clearly understand which of the requirements are feasible.
• After this, our BI expert interviewed the main users — our content team leader and I. During this stage, they established the context of dashboard usage. I’ll elaborate on that later.
• We worked iteratively: the interviews — first dashboard prototype — suggesting and coordinating the edits — next version — repeat until the final, actionable vision is achieved.?

How to tell a good dashboard from a bad one??

The main component of creating a dashboard is understanding the usage context.?

• Who is the end user??
• What is its main functionality??
• Which metrics will the end user track? For what purposes??
• How often will the user access the dashboard??
• Will they use the dashboard internally for work purposes or present it externally?
• Which rates, filters, time periods, and granularity will they need?

It’s important to speak the client’s language and gain a deep understanding of their issues and goals. This is the only way to create a workable and effective tool and not just a beautiful but inconvenient, cluttered dashboard.?

Engineering: two solutions for one project

Project stack:

• Python,
• Airflow (workflow management),
• AWS (cloud platform for hosting the project),
• PostgreSQL (database),
• Tableau (BI tool),
• Selenium (we needed a web driver for parsing the data and landed on Firefox; the scripts were written in Python)
• Fivetran.

Since LinkedIn didn’t provide us with API access, we decided to use specialized data collection tools. This is the stage where we encountered our main obstacle. The issue stems from us having two corporate accounts and a personal one. Depending on the account type, LinkedIn “shares” metrics differently, requiring different mining tools and approaches. While we had no issues with the solution for corporate accounts, figuring out the way to the personal account got complicated.?

We use Fivetran to collect data from our corporate Valiotti Analytics and Cyprus Data Community profiles.

Fivetran is a tool for gathering data from multiple sources and uploading it to one database automatically. Moreover — they’re our partner, and we’ve been collaborating on a number of client projects for some time now. So, there was no question about which tool we were going to use for our dashboard.?

To set up the data flow, we needed to connect Fivetran with our accounts and database. After this, it would automatically pull the data from the social network, transforming and transferring it to the database.?

My personal profile was a more difficult case. LinkedIn isn’t too enthusiastic about sharing the data for this account type.?

1. Connecting it to Fivetran wasn’t an option — that only works for corporate accounts.?
2. A specialized service for LinkedIn personal account data collection, inlytics.io, required solving a captcha after a while. Full automatization was our priority, so this option was in no way suitable for us.
3. LinkedIn didn’t provide the token for their official API — they stopped giving them out because of increasing demand.?

Lesson learned: even with all the necessary automation tools at hand, it takes some work to set up automatic data ingestion. We also needed to configure a solution from the ground up, using Excel sheets with statistics formed by LinkedIn. Those contain data for audience growth, demographics and engagement, as well as the top 50 popular posts. This amount of information is less than ideal, but even that much can be used with maximum efficiency if you know what you’re doing.

What was our solution?

• We set up Airflow + Selenium connection to collect the Excel files.?
• We set up the content plan in Notion, containing a heading, a topic and a link for each publication. This data was pulled via (Notion?) API.
• The data was then transformed via Airflow and merged with LinkedIn statistics. To match the data from two sources, we used links — they were the same for both tables.

As a result, we extracted maximum value from limited available data.

Both approaches allow us to track trends, engagement and subscriber growth. The initial plan was to also collect data about page visitors, private messages number, and follow our competitors’ progress. Unfortunately, LinkedIn doesn’t allow that.?

Visualization in Tableau

Well, now we established the usage context, gathered the data, finalized the mockups… time to finally build the dashboard!?

The first reason we picked Tableau is its robust functionality. Its features allow for implementing almost every conceivable idea. In a dashboard for another one of our projects, we even created a calendar using Tableau! This invention is ours — normally the software doesn’t provide this feature.

The second reason was that Tableau’s visualization solutions are very appealing. Even with no specialized design or dataviz knowledge, you can create a visually compelling, clearly readable dashboard.?

But we do have this knowledge. More than that — we have a style guide we use in all of our dashboards. It can be seen in the screenshots above: our dashboards in corporate colors aren’t just pleasing to the eye, they’re also very intuitive — which means they do their job.

Conclusions

The task we set for ourselves might look trivial — we only needed to visualize data for a couple of social network profiles. However, it presented an array of challenges and made us carefully craft the optimal solution.?

What do these dashboards offer in practice?

• Audience analysis. The dashboard collects information about my subscribers’ place of residence, their companies and positions. It allows us to hypothesize about the most interesting subjects and formats for our publications.?
• Regular tracking of the accounts’ central metrics: audience growth, subscriber engagement, reach with detailed statistics for every post. If I notice a drop or an increase in any of these indicators, I can trace when it happened and adjust my content strategy accordingly. For example, experiment with publications’ tone or topics.??
• Identifying trending subjects. I conduct a long-time analysis of 10 post topics and figure out what’s most compelling for my audience — neural networks, analytics, and data or business and entrepreneurship. After this, I narrow down the list to the 5 most popular topics, which I can then concentrate on. I’ve already concluded that my audience is interested in data analytics life hacks, personal accounts and business cases.
• AB-testing of visual design — I experiment with the look of my posts and assess the reaction.