Ads Personalisation

In today’s digital era, advertisements are omnipresent but if there’s one thing about online advertising that people hate, it’s seeing irrelevant ads that don’t suit their interests. In contrast to this, personalised ads are more relevant to users’ interests, preferences, and behaviours, increasing the likelihood of engagement. Also, from a brand point of view that pays for advertising, one wants to be sure that ad spend results in sales.

Ads Personalisation, the practice of tailoring advertisements to individual users based on their interests, demographics, behavioural and contextual attributes, has emerged as a powerful tool for marketers to enhance engagement, relevance, and ultimately, conversions. Ads personalization leverages advanced data analytics and machine learning algorithms to deliver highly relevant and personalized ads to individual users. Within the Airtel ecosystem itself, platforms like Airtel Thanks App, Wynk Music App and Airtel Xstream expose various forms of advertisements in accordance to users’ interests, preferences, and behaviours. This personalised approach to advertising plays a crucial role in shaping the way advertisers target and engage with their audiences.

Technologies Powering Ads Personalisation

Several technologies enable the implementation of ads personalisation:

1. Data Analytics and Machine Learning: Advanced analytics and machine learning algorithms analyse vast amounts of data to derive insights into user behaviour and preferences, facilitating targeted ad delivery.
2. Cookies and Tracking Mechanisms: Cookies and tracking pixels collect user data across websites, enabling advertisers to create detailed user profiles and deliver personalised ads across different platforms.
3. Real-Time Bidding (RTB) Platforms: RTB platforms facilitate the automated buying and selling of ad inventory in real-time auctions, allowing advertisers to target specific audience segments based on predefined criteria.

Ads Personalisation in Airtel using Data Science/ML

Ads personalisation relies on a combination of data collection, analysis, and targeting techniques to deliver tailored advertisements to users. Here’s a simplified overview of how ads personalisation works:

1. Data Collection: Ads personalisation begins with the collection of user cohorts from various sources.
2. Data Analysis: Once the data is collected, it is analysed to gain insights into user preferences, interests, and behaviours. Advanced analytics techniques, such as machine learning and predictive modeling, are used to identify patterns, trends, and correlations in the data.
3. Audience Segmentation: Based on the insights obtained from data analysis, users are segmented into different audience groups or segments.
4. Ad Targeting: Advertisers use the insights gained from audience segmentation to target their ads to specific audience segments. Ad targeting may involve selecting relevant keywords, interests, demographics, or behavioural attributes to ensure that ads are displayed to the most relevant audience.
5. Ad Delivery: Finally, targeted ads are delivered to users across various digital channels, including websites, social media platforms, mobile apps, search engines, and more. Ads may be displayed in various formats, such as display ads, video ads, sponsored content, and native ads.

Technical Implementation

• Airtel’s Data Lake stores 350 million user cohorts which are refreshed. regularly. The Data Science team fetches user cohorts from data lake, processes it such that it is viable to be fed to a DS model and stores the refined data in one of the tables at the data lake.

• A scheduler reads data from the DS output table, stores all the users’ cohorts in a compressed json format (*.json.gz files) and pushes it to one of the Airtel S3 buckets. This activity is performed every 3 days to push refreshed data.
• A ‘Segmentation Service’ picks these files from S3 which are in sizes in 100s of GBs, processes all the users’ data (~350 million) and stores it in aerospike for fast retrieval when required at run time.
• These files are processed in multiple chunks of smaller size with custom byte ranges. The successfully processed files are deleted while if any issue occurs during processing of any files, error % is recorded and the remaining records are processed moving the file to partially processed state.

• The ‘Ads Personalisation Service’ is created fungible enough to support multiple data models like Tensorflow, LightGBM etc. The system can support multiple models parallely and recommendation strategy could be switched between them seamlessly without affecting the existing workflow.
• The user’s affinity is calculated for every ads campaign by feeding the user cohorts and the related campaign data to the DS model. The model returns a confidence score (predicted CTR) of that user for that particular campaign. From these confidence scores, eCPM is calculated based on the campaign’s category i.e. whether it is a CPM or CPC campaign.
• Now, whenever a ‘Get Ad’ call is made by any client app to the Ads Server system, the qualified ads campaigns for all adspots are fetched and a call is made to the Ads Personalisation Service to rank all these ads campaigns in order of their relevance for that particular user.

• For every ad spot, the Ads Personalisation Service sorts the campaigns on the basis of their decreasing eCPM values and returns the response.

High Level Diagram

Conclusion

Feeding the Data Science model with the refreshed user and campaign features received at run time improved the recommendation engine significantly, thereby uplifting CTR and increased revenues. In this manner, the user is shown the most relevant ads according to the confidence score in real time and helps the publisher to deliver the expected KPIs for the brand campaigns.

As further optimisations, dynamic attributes like client IP address, time of the day can also be fed to the model to better predict the confidence score and deliver the expected CTR.

Benefits of Ads Personalisation

1. Improved Relevance : Personalised ads are more relevant to users’ interests (with right balance between privacy and utility), increasing the likelihood of engagement and conversion.
2. Enhanced User Experience : Tailored advertisements provide users with valuable and meaningful content, leading to a more positive overall experience.
3. Higher ROI for Advertisers : Targeted ads result in better conversion rates and lower acquisition costs, maximising the return on investment for advertisers.
4. Increased Ad Revenue : Publishers and ad networks can command higher ad rates for personalised ads due to their increased relevance and effectiveness. As a result, personalised ads contribute to higher ad revenue for publishers and help sustain the viability of online content and services.

Acknowledgment

I hope you’ve found this article insightful and valuable. Your feedback is important to us, so please don’t hesitate to leave your thoughts or questions in the comments below.

Special thanks to Airtel Ads team and all stakeholders who played a crucial role in bringing this feature to life.

Airtel Ads Team :

Ajit Singh, Sanket Arora, Parshant Verma, Shubhi Garg, Rajat Kumar, Harpreet Kaur, Ahana Nath, Alok Mathur, Surya Tunuguntla, Ayush Gupta