Boosting ROAS with spend optimization

Ketto, a healthcare crowdsource site, helps people raise money for healthcare, surgery, and other medical expenses. It spends heavily on its branding efforts and relies on a multichannel marketing strategy to maximize donations for the causes it supports.

The Challenges

Marketing performance data was diffused among multiple ad tech platforms, making it difficult to see which channels were truly performing and their true return on ad spend (ROAS). The two main challenges that were hampering their decision making were:

Unknown effectiveness of each marketing channel and their respective return on ad spend (ROAS).

Geographic and target segment saturation hampered spend allocation.

Data Complexity

It is vital to acknowledge the role data plays in the complexity of Ketto’s issues. The marketing performance data was spread out among multiple marketing agencies and ad tech platforms. Further, these platforms provide performance data at an aggregated level using campaign taxonomy and definitions that they define. Ketto had to undertake the herculean task of maintaining an enterprise-grade, universal marketing taxonomy. Before getting started, taxonomy mapping and the subsequent data harmonization was required.

Our Response

Data Cleanse

As part of solution development, Brainlabs first cleansed, standardized, and harmonized Ketto’s campaign taxonomy and performance data.

Channel Performance

Once the performance of each channel was understood we could determine the impact of reach and frequency saturation on ROAS.

Spend Optimizer

Channel performance insights were used to build a recommender model for an optimized marketing spending plan. These model recommendations are delivered to Ketto through Spend Optimizer, a web-based application hosted by Amazon Web Services. The Spend Optimizer use a machine learning layer that employs Market Mix Modeling (MMM) and Forecasting to quantify the impact of marketing inputs on sales or market share. Other key features of the Spend Optimizer include:

  • Scenario planning handled by a programming-based optimization layer.
  • Web-based dashboards allow for user management and access to information.
  • Front-end UX allows users to see optimization recommendations and more design scenarios.

On AWS, Brainlabs deployed this solution using the following components:

  • Raw Campaign Data: AWS Redshift
  • Analytical Data Set: AWS Redshift
  • ML objects Storage: AWS S3
  • Compute: AWS EC2
  • ML Pipeline: Apache Airflow
  • ML Results: RDS MySQL DB
  • Dashboard: Tableau
  • Planner: EC2 and S3 based web frontend and java scripting

The Outcome


For four weeks, Ketto tested the Spend Optimizer across multiple ad campaigns. After testing, the solution was deployed across all the geographies and campaigns. During the testing, it was observed that for the scenarios, spend optimization lifted revenue by approximately 10%. Also, in those scenarios, by avoiding instances of low revenue, ROAS was 20% higher than the past observed averages.