How Liberty identified 15% optimization with MMM

Context

A team designed to deliver solutions. Liberty Latin America, the leading telecommunications and technology company in Central America and the Caribbean, has an internal team dedicated to developing tools and solutions to support their marketing teams. These include forecasting tools, dashboards, predictive models, statistical robustness, and more. In a typical sales process, Liberty generates online leads and activates users through their call center. However, they faced increasing challenges in accurately calibrating digital attribution with activated users, highlighting a clear need to fine-tune their media performance.

From spending more to spending smarter. While Liberty’s team was already implementing advanced optimizations across their digital channels, they chose to further enhance their advertising investment through Marketing Mix Modeling (MMM). This approach enabled them to identify the optimal media mix based on their sales data and reallocate their budget to maximize results.

Challenge

Comparing apples to oranges. The first major challenge in this project was normalizing data across online and offline channels to make meaningful comparisons—especially when trying to align daily metrics (online) with weekly metrics (TV, radio).

Understanding impacts beyond marketing. The second challenge was incorporating variables outside the marketing area to explain final activations. For instance, analyzing how many days passed between a lead registration and user activation.

Solutions in stages

Before working with us, Liberty’s marketing mix model had a statistical confidence level of an R² Adjusted of 0.39. To build a complete and statistically reliable marketing mix model capable of identifying where to optimize media investment, our Marketing Science team provided several recommendations to Liberty Latin America's data insights team. The suggested optimizations were implemented in stages:

STAGE 1: Interpolation → Data collection and normalization
Weekly data gaps in offline media were resolved by simulating normal distributions of the missing data. This allowed the standardization of data within those gaps, converting weekly groupings into daily distributions.

STAGE 2: Share of voice → Liberty vs. competitors
We modeled the effect of share of voice in the market by analyzing Liberty’s keyword bidding data and their search impression share to estimate their market presence.

STAGE 3: Delayed Index → From order variable to activation variable
The model initially struggled because it used conversions, whereas Liberty’s business hinges on product activations. To address this, we developed a Delayed Index, which integrated the average number of days it takes to transition from an order to an activation into the model.

STAGE 4: Back-casting → Simulating the future to complete the past
To address missing clean activation data for historical accuracy, we applied a back-casting approach. This method projected the data distribution forward and used the same pattern to simulate the missing data retrospectively.

Featured products

1. Marketing Science Consulting. Integrate science into your marketing approach and transition to a data-driven culture. Our multidisciplinary team provides a consultative approach to data science, following industry best practices.
2. Markting Mix Modeling. Design your ideal marketing mix and quantify the incremental impact on ROI and sales from all your marketing activities. This enables you to reallocate your media budget for better results.

Results

In the customer's voice

Modeling activations is not the same as modeling conversions. This MMM provides a clear optimization agenda and opens the door internally to a potential data-driven budget reallocation.

Alan Hasman

Data & Analytics Manager @ Liberty

Next steps

A definitive optimization agenda. With Liberty and Bunker DB working side by side, we developed a highly reliable model to create an optimization agenda aimed at improving the efficiency of the marketing budget. The next steps will involve a mix of budget reallocation and incremental studies, turning off channels in specific regions or moments to continue refining the model.

1. Intra-digital reallocation. The first step involves reallocating the budget across online channels (Google, Meta, etc.).
2. Calibration with lift studies. Next, incremental studies like Geo Lift and Conversion Lift will be conducted to adjust and fine-tune the MMM.
3. Reallocation from offline to online. Lastly, investments in TV, radio, and other offline media will be shifted to online channels.