Social Network

Predictive analytics in SMM, or how to evaluate the effectiveness of advertising campaigns at the planning stage

It just so happened that for the fourth year I have been working as an SMM specialist and engaged in conversion marketing. I confess to you: I hate to communicate with people, especially those who do not understand me. But the irony of life is that most of the time I have to communicate with clients who do not understand what exactly I am doing. And at about the tenth or twelfth meeting, when another customer asked me to give an accurate forecast and a guarantee of conversions from the social media channel even before I know the name of the project, I thought that I had not met a single analogue on the market that could calculate a similar forecast based on the specifics of my channel.

And having developed a formula for predictive analytics, I would kill two birds with one stone: I gave guarantees to the client and reduced the communication procedure to a minimum.

Of course, I understood that I could use approximate data from the advertising office, dance with a tambourine around the decomposition, or parse BI analytics tables from SEO optimizers or PPC specialists, but this somehow did not fit into all the intricacies of social networks … But here I was lucky to meet, probably, with one of the best analysts on the CIS market, Anton Lipsky. He helped to collect and implement a predictive analytics formula based on the tasks that my real and potential customers set for me.

Therefore, let’s talk about what metrics we can predict and what not, what is the specificity and fundamental difference of our formula from everything that was on the market, and how predictive analytics behaves in practice.

What indicators can we predict?

  • Absolutely all metrics that can be digitized (we can calculate the conversion rate and the number of sales from Instagram – great, so we can make a forecast of these indicators).

  • Data based on historical indicators (we have data from previous advertising campaigns, which means there are statistics, the rest is a matter of formulas and patterns).

  • Homogeneous indicators (if you work with homogeneous indicators, then we can work with them).

What indicators can we not predict?

  • Subjective and situational performance criteria (alas and oh, we cannot predict such metrics as “creativity” or “good and bad”, because they, at least, are difficult to digitize, and as a maximum – to evaluate objectively).

  • Data that does not have a pattern (working with a random set of information, we will not be able to identify the statistics of certain indicators).

  • Indicators are out of the context of time (and if our data is taken out of the context of time, situation, region and audience, nothing will come of it).

Having decided what data we can or cannot take into work, we proceeded to create the very structure of the formula, which we will try to parse.

The formula consists of 4 large blocks

The first block is the “Base” tab , in which we enter all the historical indicators for previous advertising campaigns that we have. The formula will then analyze certain patterns, narrowing the extreme points and making a statistical prediction.

In this case, it is important to understand that campaigns should have the same goal, be homogeneous and similar in terms of conditions and results, so that the formula does not go crazy when you first sell socks for USD 1,000, and then the latest iPhone model for 0.99 USD.

It is also important to consider that the more input data we enter, the more accurate our results will be. We enter all the data we need into the formula and proceed to the second stage of working with it.

Forecast Tab is the one that interests us the most.

An actual forecast that will give you and your client an understanding of what results we will get from the worst, the best and the best outcome. We can start from the required number of transactions, understanding approximately what coverage and budgets we need, and from the budget, immediately showing the client what results he will get within his budget.

In addition to all this beauty, the formula calculates both general data, which will be taken from all parameters of the Base tab, and narrow data, within the 40% range of budget indicators.

Thus, already at the negotiation stage, we can show our customer what he will receive for his budget and how much budget he needs to invest for the coveted 100,500 sales per minute from Instagram.

But the fun doesn’t end there, because we have two more tabs.

The third is the Analyze clause of the formula, where a brief analysis of advertising campaigns is carried out in terms of quantitative and qualitative indicators, as well as relative and absolute deviations from the general base of inputs are indicated.

And in addition to everything on this tab, we can see the potential reason for our deviations: from illiterate audience settings to the technical side of the issue.

And for dessert we have the Comments tab, where we need to enter data on the positive and negative aspects of the campaign. For example, it worked well because there were good creatives, or it worked badly because the region was limited.

And after that we can understand in the previous tab what is wrong or, on the contrary, very much the same in our campaigns.

A small case on the practical application of our formula

We tested a new algorithm on one of my longest-running projects of the Ukrainian chain eatery FreshLine. Based on the previous 10 advertising campaigns, we made a forecast and received the following actual results:







Number of conversions

Forecast (max / narrow data)

458 700

$ 1.41

$ 0.82

$ 0.83




370 650

$ 0.051

$ 0.37

3 $ 0.53



Despite the insignificant scatter of the data and the deviation from the maximum result, we received data that are quite close to the forecast. We could get more accurate data on condition of an increased number of input data, for example. Nevertheless, we tested the formula in the field and we can confidently say that it worked in favor of the client and specialists, everyone was satisfied.

To summarize, we can say that the formula has its pros and cons:

Formula advantages


  • Multifunctionality and flexibility of the formula;

  • unlike decomposition, it does not just read data, but does predictive analysis;

  • adapts to the tools and functionality of SMM, and not some other channels;

  • individual and adaptable for different projects / regions / mechanics.

  • The need for historical and homogeneous data;

  • the formula may not work if there are fundamental qualitative changes in the campaign conditions;

  • An accurate forecast requires a lot of input data, there are indicators that we cannot predict.

The application of this formula proves that predictive analytics in SMM is a reality, and now we can give forecasts and guarantees for our clients. The predictive analytics formula is universal and can be adapted to various business tasks. And of course, it is important to remember that the actual results may differ from the predicted ones for objective and subjective reasons, since we always work with three parties to the project: the specialist, the customer and the audience, and it is not known how this or that variable will affect the results of the project.

And if you have any questions about the functionality of the formula and the implementation of predictive analytics in your project – write in the comments, and we will try to figure out your problem together.

From the Editor. Don’t miss Nikita’s other articles on our blog:
Case: how to make 1960% ROI in SMM without a website and CRM system
Case: 600+ sales in SMM in 2 weeks with a budget of $ 100
How to track KPI and performance in SMM
Why there is no SMM in Ukraine

If you find an error, please select a piece of text and press Ctrl + Enter

Leave a Reply

Your email address will not be published. Required fields are marked *