Warning: this content is older than 365 days. It may be out of date and no longer relevant.

How to Build a Data Driven Social Media Strategy Plan

Emily asks, "What are your favorite resources for social media strategy planning?"

The overall strategy for social media planning is and should be tightly coupled to your overall digital marketing goals. What are you trying to achieve as a department or organizational unit? Demand generation? Lead generation? Ecommerce sales? That’s the starting point.

On my website, I care about four things – consulting opportunities, public speaking opportunities, book sales, and newsletter subscriptions.

Determine Channel Effectiveness

Once you have a clear idea of what goal you’re measuring towards, build and run an attribution model. If you’re dealing with mostly digital data, I recommend a machine learning model based on techniques like Markov chain modeling or Shapley values (I prefer Markov). What this achieves is an analysis of which channels already help to drive those conversions.

For example, I recently changed how I publish content – I’m back to traditional blogging. (the reasons are detailed in this post) When I look at my attribution model for how my site is doing, it’s definitely the right choice. Here’s my model for January 2020:

2020 Attribution Model

We see that organic search was the primary driver of conversions, impacting 58.9% of conversions and yielding 184 conversions, followed by email at 16.5%, yielding 52 conversions.

Now, let’s look at how January went, with my new strategy:

2021 Attribution Model

That’s a big difference. Organic search is now 51.2% of conversions impacted, but the absolute number is more than double at 387 conversions. That’s a huge jump, to double the number of conversions from organic search year over year.

Email continues to bring up second place with 36.2% of conversions impacted, with 273 conversions, also a nice gain over the previous year.

What about social media? After all, we’re trying to build a social media strategy plan here. In both models, Twitter has driven the most social conversions; Facebook was doing okay in 2020 but has dropped down to the middle of the pack in 2021.

So, I have a good sense of which social media channel is working. Now the question is, how do I get more out of that channel?

Determine Channel Tactics

Once we know what channel’s working, it’s time to figure out WHY it’s working. If we can crack the code about what makes a certain channel tick, then we can do more of what works. To accomplish this, we first need to know what data is available.

For this exercise, I’m looking at Twitter. It’s consistently a winner, year-over-year, in terms of social media channels for me. Why? What am I doing on Twitter that’s resonating?

To answer this question, I’ve got to get Twtiter data. Right from the platform interface, I can export tweet-level data. What kinds of data?

Twitter platform data

While this spreadsheet is good – with all the major features like likes, retweets, impressions, etc. it’s not very good for understanding what’s working and what’s not. A big part of a social media strategy plan is understanding the content itself, and that’s not included in here (and never will be).

What is included is the text of my tweets. Using basic text statistics, I can look for the words and phrases I use the most, then engineer new numeric fields based on those topics. For example, I’d count how many times I use the phrase "data science" or "social media".

Once I have that information, it’s time to figure out if any of these metrics have a mathematical relationship to the outcome I care about, which in this case is clicks on the URLs in my tweets. Why? In order for someone to do business with my website, they need to get to my website. How do they do that? By clicking on links in the content I share that goes to my website.

To build a model of what predicts my desired outcome, I use a machine learning technique called XGBoost to identify which participating variables have the strongest mathematical relationship to my target outcome, clicks. Bear in mind, I have to remove things like linear dependencies – metrics like engagements include clicks, so those variables shouldn’t be used in this kind of prediction. I also have to remove things like zero values – metrics which are always zero, like all the promoted tweets metrics (because I don’t promote my tweets).

What’s the result?

Twitter Channel Tactics

Impressions is the greatest predictor of whether a tweet will get URL clicks, followed by retweets and detail expands. This is where domain expertise now comes into play. As a social media marketer, what things can you do to generate more impressions, more retweets, more detail expands?

Impressions is pretty straightforward – you could pay to promote tweets. You could tag people in your tweet. You could use more popular or more niche hashtags. You could cross-promote your tweets on other social networks. You could include tweets in email marketing.

Retweets would follow similar tactics, plus the very obvious asking people to retweet more important/valuable tweets.

Detail expands is even easier – this feature isn’t available on tweets that don’t require it, like links and basic text tweets. Images and video do – so this would be encouragement to upload more media tweets.

Is this the answer? Not quite. This is the testing plan, because regression analysis of any kind (which is what XGBoost is) only ever proves correlation and you learn in stats 101 that correlation is not causation. So the next step in our process would be to build a testing plan, testing each of these predictors – impressions, retweets, detail expands – individually and seeing if URL clicks increase. I’d set up a three week plan and tweet the same amount about the same topics, to the same audience, but with a focus on the tactics above for each of the predictors, one per week.

After three weeks, I’d examine my results and identify whether or not the test was successful in generating more URL clicks than previously, and for the winning tactics, I’d now have a recipe for my social media strategy on that channel.

Finally, I look at the analysis in terms of content and topics above. Why aren’t the topics I’ve identified a good predictor of URL clicks? Why aren’t people clicking on a specific topic more than others? The lack of predictive value for specific topics hints that I might not be sharing the kind of content my audience wants from me, so I should consider collecting data on a very common marketing topic and seeing what topics are most popular, then test that as well.

Assemble the Final Plan

After doing the above evaluation for as many channels as my attribution model suggested, I’ve got a social media strategy plan fully baked:

  • I’ve identified whether social media is working for me or not
  • I’ve identified what channels are working for me
  • I’ve identified what tactics on those channels get results
  • I’ve built a testing and action plan for those tactics to prove their effectiveness

The final piece of the plan is something known as perturbation – introduction of randomness. One of the dangers of rigidly following a data-driven plan is over-optimizing, focusing only on what works with no testing of new things, even if those new things deliver less-than-great results. Anywhere from 5-20% of your marketing should always be perturbation, always be testing something new and different for a period of time to find out whether it’s appropriate or not.

Perturbation testing could be within a set of tactics or channels, such as trying out Tiktok or Clubhouse, or it could be very granular, such as testing out new content or themes. Whatever it is, you always want to be testing new things to make sure you don’t miss the boat on an important sea change in your audience.

So, that’s the data-driven answer on how to build a social media strategy plan. It will take time, resources, and expertise to put one together, but once you have it, you will stand behind it confidently.

Shameless Plug

If this sort of planning is something you’re interested in but you don’t have the time or knowledge base to do it, my company, Trust Insights, can build a data-driven social media strategy plan for you.


You might also enjoy:


Want to read more like this from Christopher Penn? Get updates here:

subscribe to my newsletter here


AI for Marketers Book
Take my Generative AI for Marketers course!

Analytics for Marketers Discussion Group
Join my Analytics for Marketers Slack Group!



You might also enjoy:


Want to read more like this from Christopher Penn? Get updates here:

subscribe to my newsletter here


AI for Marketers Book
Take my Generative AI for Marketers course!

Analytics for Marketers Discussion Group
Join my Analytics for Marketers Slack Group!