The Prompt I Needed: Using Copilot Analyst for Social Media Analytics

When AI was first introduced to the world, I was hesitant to dive into this mystical realm. What are the best practices? How do I avoid plagiarism? Is this the rise of machines?  In those early days, I certainly didn’t see how it could help with data analytics and reporting.

Over my career, social media platforms have drastically changed the way they share analytics (a rant for another day). So, when I set out to analyze a relatively large data set, I asked myself what type of metric I should focus on and wondered whether engagement was still the right fit for the company’s needs.

Regardless of what metric I was going to use — spoiler: I decided engagement was still the right fit for our goals — I knew I needed help from a tool to compile data into a reportable format that told a story. I remembered a training I attended that briefly touched on one of the newer features of Copilot: the Analyst tool, designed specifically for data analysis. This was just the prompt I needed.

Determine the Objective

First, I needed to determine my objective so I could write clear, productive prompts. My goal was to create an informal internal report showcasing each social media platform separately (Facebook, Instagram, and LinkedIn). I wanted monthly, quarterly, and yearly breakdowns.

While I wanted to see other metrics like impressions (reported already by the platforms), my goal was to find a way to import large amounts of data by post and have it calculate my engagement rates. This was calculated by the Analyst tool using the data I provided and the equations I instructed it to use. (Example: total interactions divided by total impressions, etc.).

I also played around with different types of engagement rates and equations you can calculate. This was not necessary for my project, but it provided an added element to my presentation.

Consider your Data Sources

I considered how each platform reports different types of data. One challenge I noticed was that data reported through the social media platform, also known as native data, varies slightly from our scheduling tool’s reported data. I chose either native or third-party data and stuck with it for each platform, to allow for consistency in my reporting.

Remember Your Expectations

Since perfection wasn’t critical, I had freedom to experiment. I found Analyst helpful for informal reports. I have not tried it for formal reporting, so I would proceed with caution if that’s your goal. Perhaps consider using it for a smaller analysis that allows you to perform a quality control check by hand until you can ensure accuracy would be helpful.

Prompt It Well

Use detailed prompts that explicitly state what you’re providing. Here are some of the types of items I include in my prompts:

• What you are providing it: A spreadsheet that contains data.
• What you want it to do: Calculate engagement rate by taking reach divided by total interactions.
• What you don’t want it to do: Exclude data from irrelevant years. (It may be helpful to delete data you don’t want included before attaching to avoid confusion and to avoid noting which what years to exclude.)
• How you want it: Organize results into an Excel spreadsheet.
• Revise: Don’t be afraid to tell it what was wrong and what you want it to revise. AI does have a certain proclivity for agreeing with you and modifying itself to your specific needs. Don’t be afraid to point out its errors and correct it with feedback in your next prompt. I tend to be very straightforward and have even said “No, that is not correct. Please modify to…” or said “That is close, please modify…”

I recommend asking it to do one thing at a time. I learned that Copilot gets confused if you ask too much at once. However, don’t confuse this with detailed prompts, which are important for getting the most accurate results from your chosen AI platform. You can provide a detailed prompt and still ask it to do one specific task.

Sample prompt:

Organize types of data by each channel type and calculate engagement rate with only the data available for each of these content types individually. Calculate engagement by taking the following formula: total interactions by impression. Create a table that shows the post and the engagement rates. Organize it into an Excel spreadsheet where it is organized by channel type and rank data in order of top to bottom engagement rates. Please include the date it was posted in a column as well.

You can then follow this prompt up with something like: “Take this same information and use the formula reach divided by total interactions to develop your engagement number.”

Results

After many experiments, research, and more than a few moments of remembering why I wasn’t a statistics major, I was able to provide a helpful report that showed our engagement rates by post overall for the year, month, etc., and glean a decent idea of the top posts for the year.

Having all of this in Copilot and easily accessible made it helpful for deepening my report with follow-up prompts, such as asking it to show me the top five posts by engagement, impressions, and even reach. While each of these metrics don’t tell a complete story on their own, all of them together reveal areas to improve depending on your goals.

Using a tool like Analyst helped jumpstart what would have usually been a cumbersome manual task and ultimately helped me evaluate our team’s social media approach. Results were well received and prompted further discussions on ways to improve our process going forward, engagement rates and what metrics were right for us. These discussions are ongoing, and they provide a good starting point. I’ve continued to utilize the Analyst tool over the past year since this initial experiment, significantly saving time and resources on data tracking.