How I increased profit from scaled-up advertising campaigns?

Business Challenge

Company X, a dynamic distributor of impulse goods operating in domestic and international markets, faced a classic growth problem: how to scale advertising campaigns without incurring losses.

The marketing department was heavily investing in Meta Ads and Google Ads. However, the process for allocating budgets was chaotic and based on rigid, inefficient metrics. The consequences were clear:

• Suboptimal spending - some campaigns generated profit, but scaling them led to a decrease in profitability (profit diminishing as budget increased).
• Lack of prediction - there was no tool available to objectively verify whether increasing, maintaining, or decreasing the budget was truly optimal for a specific SKU and market.

Company X was spending money on advertising, but it didn't know exactly how much it could spend to maximize profit.

Business Goal

The goal of this project was to implement an advanced analytical tool capable of classifying and predicting the performance of advertising campaigns across Meta and Google platforms.

Instead of relying on intuition or fixed thresholds, I created a Decision Support System (DSS) designed to provide recommendations regarding the optimal budget level for every active campaign.

Measurable Success: Achieving an average increase in profit per unit sold (SKU).

Implementation

The project was conducted in four closely linked stages, combining data auditing, advanced statistics, and IT implementation.

Step 1: Data audit

Initially, the data was limited in scope (only available for a few months), which prevented seasonal analysis. I performed an in-depth audit of the available data, identified key variables, and structured the SKU database to understand Company X's full sales ecosystem.

Step 2: Building the predictive model using machine learning

I initially built one classification model based on predicting whether a single, arbitrarily set profit threshold would be exceeded. Testing showed that this approach was too general and lacked accuracy.

I changed my strategy: I developed a series of regression and classification models capable of forecasting the specific amount of profit per unit sold and the probability of achieving various profitability levels. The model did not just predict; it generated actionable recommendations for the marketing department on exactly how much to increase or decrease the daily budget.

Step 3: Performance testing and model optimization

I observed significant data heterogeneity—model accuracy varied depending on the traffic source (Meta vs. Google) and the target country. Instead of one universal model, I developed several dozen specialized model series (based on 5-7 variables), testing them over a full operational month. This allowed me to determine the most precise configuration for each country and traffic source.

Test Results: Achieved classification accuracy of around 92% (Meta Ads) and 88% (Google Ads).

Step 4: Implementing the Decision System

Finally, in collaboration with the IT team, I implemented the best model series into a dashboard for the marketing department. This enabled real-time monitoring of model performance and immediate campaign scaling based on predictions, rather than guesswork.

Results and Recommendations

The implementation of this advanced analytics system had a direct and measurable impact on Company X's finances:

• Loss Reduction - the models identified the precise daily budget ceiling above which campaign scaling was unprofitable.
• Spending Optimization - campaigns that should have been maintained or reduced were paused much earlier than before, immediately limiting spending on non-profitable ads.
• Profit Maximization - the most significant achievement was an average increase of 6 PLN (or about 1.5 EUR) in profit per sold SKU. Given the scale of hundreds of thousands of products sold monthly, this represents a substantial and repeatable increase in operating margin.

Recommendations for Your Business

If your marketing or sales team is struggling with the question, "We are spending money on advertising, but we don't know if it's optimal spending?", my solution can be critical.

It is not enough to know what was done in the past—you must predict what will work tomorrow. I specialize in building predictive systems that transform raw operational data (such as ad spend) into concrete, financially justified business recommendations.