Making More Powerful Lookalike Audiences With Machine Learning

Not known for their technology, advertising agencies often struggle in using all their clients’ data to its fullest potential. That’s where LaunchLabs comes in. LaunchLabs helps ad agencies make data-driven decisions that lead to more effective and successful marketing campaigns. LaunchLabs wanted to test using AI innovations to make its lookalike audiences perform better. They turned to NineTwoThree to test its data’s predictive power based on locations and brands.
launchlabs-hero
launchlabs-concept

Concept

LaunchLabs helps ad agencies and car dealerships maximize lead generation through data-driven solutions and harnessing the power of first-party data.

Since ad agencies often do not know how to use their clients’ data, LaunchLabs was founded to help car dealerships get more targeted traffic to their websites.

LaunchLabs wanted to enhance its product’s effectiveness with the predictive capabilities of machine learning. Knowing the latest innovations in machine learning led the company to wanting to test this method to create even better audiences for its customers.
‍
This is where they needed a machine learning partner, namely NineTwoThree Studio.
lau-challenge-screenlau-challenge-screen-mob

Challenge

There were two major challenges that NineTwoThree faced in the course of this project, and they both related to data.
Assessing
Predictive Power
The LaunchLabs dataset presented challenges in determining its ability to predict outcomes effectively. Incomplete information in many columns made assessing their predictive value difficult. Some columns showed high correlations with each other, making it hard to understand. Certain columns seemed unrelated to the target variable (website visit flag), and missing data for non-responders posed additional challenges.
lau-challenge-nulber2
Database Schema for Machine Learning
The quality and correlations of columns in the dataset raised concerns about whether the database could effectively support machine learning. Highly related columns could impact the dataset's usability for ML, and the presence of columns with zeroed-out data, especially related to personal interests, questioned its suitability for ML purposes.
lau-challenge-nulber1

Solution

To evaluate LaunchLab's predictive power, a project was initiated to build machine learning prototypes. This involved building lookalike audiences based on location as well as brands.
lau-solution-arrow-down
Phase 1
Model for Lookalike Audiences
(Car Dealerships in Virginia)
To evaluate LaunchLab's predictive power, a project was initiated to build machine learning prototypes. This involved building lookalike audiences based on location as well as brands.
Phase 2
Model for Specific Brand
(e.g., Honda)
Building on the success of Phase 1, a model was developed for crafting lookalike audiences for a specific brand, such as Honda, following similar data collection and model development processes.

Impact

LaunchLabs' efforts have resulted in providing more precisely targeted audiences for ad agencies, enhancing their understanding of effective data utilization and product offerings.
‍
The dataset demonstrates promising predictive power, as seen in the model results.
lau-impact-shadow lau-impact-screen
As these findings are implemented, LaunchLabs can expect significant impacts on their business objectives:
target
Enhanced lead conversion rates through precise targeting of website visitors.
money-send
Improved sales and marketing ROI by delivering personalized and precise advertisements.
magnifying
Reduced customer acquisition costs by focusing on high-potential leads.
NineTwoThree Studio collaborated with LaunchLabs to address critical challenges related to predictive power and database schema for machine learning. This collaboration resulted in more precisely targeted audiences, improved lead conversion rates, and enhanced sales and marketing ROI overall.
923-footer-icon

Thanks for reading!

NineTwoThree. 2024