Back to Success Stories
Media
Data & AI TeamsData & AI Strategy
Problem
The client’s main goal was to build an automated system that could send specific ads to a targeted user and increase conversions for their customers. Prior to working with Fusemachines, the client’s data science team was manually categorizing specific ads and clustering users to achieve specific conversion goals for their customers. The client was facing a scaling issue as new customers were being onboarded faster than they were able to perform these manual tasks.
The data science team created an initial machine learning model to automate the manual clustering and ad segmenting tasks leveraging historical data collected at the user level. The first version of the model the internal team created had not performed at a level their data science team had expected. The first machine learning model produced results with lower conversion rates, compared to segments created through the manual process.
The client needed to quickly bring on machine learning engineers to their team that could elevate the performance of their existing model or build a new model that would outperform the manual process. The model would need to understand user interests, cluster users into segments, predict ad copy, and select a device and time for clustered users to push notifications in order to improve conversion rates.
The data science team created an initial machine learning model to automate the manual clustering and ad segmenting tasks leveraging historical data collected at the user level. The first version of the model the internal team created had not performed at a level their data science team had expected. The first machine learning model produced results with lower conversion rates, compared to segments created through the manual process.
The client needed to quickly bring on machine learning engineers to their team that could elevate the performance of their existing model or build a new model that would outperform the manual process. The model would need to understand user interests, cluster users into segments, predict ad copy, and select a device and time for clustered users to push notifications in order to improve conversion rates.
Challenge
- The organization wanted to beat manual work performed by the data science team by 20%.
- The data they were able to collect didn’t allow them to collect all of the features needed and didn’t process it correctly to predict the best outcomes.
- No off the shelf product or API could solve their problem.
- They required adding a machine learning model to be trained that would successfully send out a targeted ad to a specific user.
Solution
Within 6 months of working with this company, Fusemachines created and deployed a neural network based factorization model that automatically generated multiple segments for the push notifications the client sent out for their customers. Data was collected on a daily basis to train the model using a user’s activity (geostat, time of the day the user clicked offers, etc). When training the model, precise ads would be sent out at peak times likely to generate conversions.
Fusemachines was also able to fulfill the client’s desire to eliminate the need for a manual process by optimizing the model to train itself on live data and tune itself automatically multiple times a day based on live performance. The model created by the Fusemachines team sent approximately 100M push notifications daily (with an average of 10-20 million push notifications during peak hour).
Fusemachines was also able to fulfill the client’s desire to eliminate the need for a manual process by optimizing the model to train itself on live data and tune itself automatically multiple times a day based on live performance. The model created by the Fusemachines team sent approximately 100M push notifications daily (with an average of 10-20 million push notifications during peak hour).
Results
The model greatly improved the client’s bottom line by boosting campaign performance by 15-45% and beating human performance of created segments by over 30%.
Since creating this model, approximately 20 billion push notifications have been sent overall. The amount of push notifications being sent out has increased 4 to 5 times since Fusemachines began working with this client.
Fusemachines delivered a customized Machine Learning model, developed critical data infrastructure for their ML pipeline, and eliminated the manual process altogether greatly reducing the time and resources needed to train models, iterate segments and improve ROI. Fusemachines AI talent continues to further develop ways for the client to send push notifications using advanced AI techniques to cluster users together and provide more personalized push notifications. The models the Fusemachines team created remain in production today.
Accelerate your organization’s data to AI journey.
Tell us how we can help and we’ll take care of the rest.
Tell us how we can help and we’ll take care of the rest.