INTRODUCTION
In today’s digital age, personalization is a crucial factor in capturing and retaining customer attention, especially in highly competitive industries like telecommunications. Customers expect tailored experiences that align with their interests and usage behaviors. Traditional marketing campaigns, however, often fail to meet these expectations due to their one-size-fits-all approach. Our client, a leading telecommunications provider, sought to transform their marketing strategy by adopting a more personalized, data-driven approach. Their goal was to create highly relevant marketing messages for customers, delivered through SMS, to promote campaigns effectively. The challenge lies in personalizing communication at scale, leveraging both customer data and emerging technologies like Generative AI. In addition to this challenge, various factors such as personal identifiable information (PII) and selection of relevant customer data were also considered to ensure that data privacy protocols were followed.
CHALLENGE
While our client had already implemented a machine learning (ML) model to identify the next best offer for their customers, they wanted to leverage customer-level data to further increase revenue. At that time, all customers were receiving the same promotional SMS messages along with the next best offer, but without any personalization that could make the spiels of these messages more relatable to the customers. Additionally, the client also wanted to relate to the customers more by using spiels written in Taglish, which is a mix of Tagalog and English; this code-mixing is widely used and accepted as a conversational manner of communication across different socio-economic classes in the Philippines. To address these concerns, the client needed a more refined method of delivering content; they wanted to leverage the wealth of mobile usage data available to them, which included information about the apps customers frequently used and how much data they consumed. The challenge is to use this data to segment customers effectively and create personalized SMS messages that would resonate with each customer’s unique preferences.
OneByZero saw the lack of personalization in these messages as an opportunity to boost the take-up rates for the promoted products through Customer Segmentation and Generative AI. Although the main purpose of the segmentation was to group the customers based on interests and preferences, it also served as a way to eliminate any PII of the customers. Lastly, Generative AI helped solving the client’s challenge of creating personalized and localized communication at scale.
SOLUTION
To solve this problem, we developed a marketing framework centered around Customer Segmentation (CS) and Generative AI (GenAI) which enabled us to create personalized spiels of the promotional messages for each customer segment and to achieve the following:
- Identified the relevant clusters of interests using customer segmentation. We were able to identify customer's interest and usage based micro segments at granular levels (e.g. weekend social media scrollers).
- Utilized our ML-powered Next Best Offer (NBO) engine to identify suitable promos and freebies for the customers.
- Leveraged GenAI to generate the personalized messages on basis of the customer interest based segments and recommended promo.
Implementation
The implementation of this solution followed a structured, four-step process: the results of the Interest Segmentation Analysis and the Predictive Product Recommender were used as inputs for the Gen AI Model Development to generate the personalized messages. These messages were then validated and approved before finally being broadcasted live.
.png%3FspaceId%3D16ebcd74-563a-48a2-9ce7-686f30d5c337?table=block&id=1282cc79-102c-80b2-b808-f8dd79f8f687&cache=v2)
1. Interest and Customer Segmentation Analysis
We began by segmenting the customers based on their mobile data usage patterns — grouping them according to the volume of data they consumed and the types of apps they most frequently used. To ensure that we could extract the customers’ particular interests, we removed the most common apps used by the customers and excluded the background apps as well. This allowed us to segment the customers properly; for example one segment might be focused on shopping while others could be more related to gaming. This segmentation provided a solid foundation for tailoring marketing messages to specific user groups.
The diagram below shows how the different applications used by the customers were grouped together into a category based on what the applications do. Then the different categories were further grouped together based on the similarities of the activities within each category. These different domains were now considered customer interests. The combination of different domains and the degree of the customers’ engagement in the domains described where their interests lie.
2. Machine Learning for NBO Recommendations
The primary objective of the existing ML model was to increase the client’s revenue and their products’ take-up rate by analyzing customer data and predicting the most appealing offers to the customers, which would encourage them to purchase a product in the next few days. This model analyzed a wide range of customer data points, including past purchases, app usage, and demographic information, to identify which products were most likely to appeal to each customer. By analyzing customer behavior, the model determined the product a customer was likely to buy and then upsold it to them to boost the client’s revenue and that specific product’s take-up rate. For example, if a customer’s purchase history indicated a preference for products priced around 20 pesos, the model could recommend a similar item, such as Product A, which was priced at 30 pesos. Using this approach to identify the next best offer resulted in a higher net take-up rate and revenue.
3. Generative AI Model Development and Validation
With the interest segments established and the next best offers identified, we then built the Generative AI model to produce personalized SMS messages using these as inputs. The model was trained on existing marketing spiels, learning the structure and language needed for each segment. By using AI-generated content, we automated the process of message creation, allowing for the rapid production of thousands of unique messages, each tailored to the interests of specific customer groups. During this process, we also instructed the Generative AI to generate the messages with specific placeholders. These placeholders would then allow the messages to be utilized by the broadcast platform by dynamically filling them out with the identified next best offer for each particular customer.
The messages were validated by randomly sampling the generated batch of messages, which were reviewed by domain experts familiar with marketing and campaign messages. We also devised ethical and technical guardrails to automatically check these messages, which gave an overview of the quality of the generated messages. Specifically, we created a spiel checker in Python which ensured that the spiels follow the specified format and exclude the use of profanity, potentially offensive language, and unwanted terminologies. We also utilized bigrams in order to determine common phrases related to the interests and to gauge whether the spiels of the messages represented these interests. The whole process of building and tuning the Generative AI was executed in repeated iterations, with each iteration improving the Generated AI’s performance and accuracy. The iterations ended when all messages passed the guardrails and were approved by the domain experts.
4. SMS Broadcast to Live
After the messages had been dynamically filled with the product recommendations, the outputs were then passed into a database, from which the telecommunications campaign platform ingeststhe messages for broadcast to all the target customers. This Generative AI campaign became part of the live broadcasts running alongside other traditional campaigns. This ensured that the Generative AI campaign ran with preset campaign broadcast schedules to maximize conversion potential.
RESULTS & BENEFITS
By implementing this personalized marketing strategy, the client achieved significant improvements in customer engagement and product take-up rates:
- Boosted Take-up Rate by 37%: By leveraging Gen AI to create personalized messages on top of the ML-based product recommendations, we were able to increase the overall product take-up rate by 37% compared to the rate when only using the ML model.
- Efficient Scalability: The combination of segmentation, Generative AI, and machine learning allowed the client to scale their marketing efforts with minimal manual input. The system could generate thousands of personalized messages in a fraction of the time it would take using traditional methods.
- Localized Communication: The inclusion of Taglish, which is a fusion of Tagalog and English and a commonly spoken language by the masses in the Philippines, made the messages more relatable to the masses, fostering stronger connections with customers.
WE ARE ONEBYZERO
Headquartered in Singapore with local presence in Asean nations, we are a modern data & AI consulting firm. We focus on transforming enterprises with cutting edge solutions to generate value from data. We specialize in serving the Telecommunications, Banking & Financial Services, Retail and Ecommerce industries.
We focus exclusively on AI/ML, Data & Martech
AI & ML, including Generative AI
We help organizations define their AI/ML strategy, develop and operationalize AI/ML & generative AI models, and implement MLOps to streamline operations. We have experience doing a variety of work in classical data science and cutting edge GenAI for Telcos
Modern Data Platforms
Our team of experts build robust data pipelines, design data warehouses and lakes with strong focus on data quality and lineage, to build C360, marts, support reporting, dashboards to help organizations uncover insights in their data using a modern data stack.
Martech & Personalization
We enable customers to modernize their digital platforms to deliver omni channel personalization use-cases. We have deep experience with marketing use-cases & personalized content generation, automated campaign grid optimization and next-best offers & actions.
.png?table=block&id=1732cc79-102c-801d-b8e2-ff2d69f2ab9f&cache=v2)
.png?table=block&id=1732cc79-102c-80a0-8bb7-f6df62c32cb9&cache=v2)