Aiming to extract more value from its current customer base and increase its conversion capacity, Vero sought to optimize its outreach strategies through personalized, individual offers. Project Aurora was designed to validate the effectiveness of a predictive model, focusing on up-sell, increased profitability and the sustainability of the revenue-cost relationship.
Vero faced the challenge of not having the capacity to analyze its customer base in real time to personalize product offers. Recommendations were updated only once a month, which ended up creating a misalignment between the recommended offer and the customer's real needs. There was difficulty in accurately understanding behaviors and preferences, resulting in offers that were not granular, and the operation dealt with a large base fragmented across different CRMs (Adapter, Simetra and NG), encompassing more than 1.2 million customers.
MadeinWeb, in partnership with AWS, developed Project Aurora, a predictive best-offer recommendation model focused on generating value and monetizing Vero's user base. The solution was based on a supervised regression model (CatBoost), trained on behavioral profiles and contract variables, including internet speed ranges, TV tiers, number of SIM cards, contract value and geography.
Clustering of the base was used as an exploratory step to understand segments, while the predictive model generates the personalized offer recommendation for each customer.
The ingestion and integration of data from the three sources (Adapter, Simetra and NG) was automated with AWS Glue, with governed datasets stored in Amazon S3. The ML pipeline is orchestrated by Amazon SageMaker Pipelines, which coordinates processing, training, validation and model registration.
Features are managed in Amazon SageMaker Feature Store, ensuring consistency between training and inference. Each candidate model is versioned in the SageMaker Model Registry with attached validation metrics, and batch inference is executed via SageMaker Batch Transform.
Drift and predictive quality monitoring is performed by Amazon SageMaker Model Monitor, with operational alerts via Amazon CloudWatch. Analytical queries on results and campaigns use Amazon Athena.
The model combined analytical results with the updated plan portfolio, enabling granular recommendations within the user's same contract value range. The pilot project was executed with a geographic focus on customer bases in Brasília, São Paulo, Minas Gerais, Paraná, Rio Grande do Sul and Santa Catarina.
The strategic execution involved personalized messaging (WhatsApp) for up-sell offers, such as migrating “Naked” customers to packages with Mobile or SDP products. To ensure a fast and efficient acceptance process, the team rigorously monitored every stage of the communication flow, evaluating indicators such as sends, deliveries, views, acceptances, final conversion and average ticket delta.
“The pilot results demonstrated that an assertive approach, built from machine training and the application of algorithms on the customer base, when combined with appropriate product attributes — speed, mobile or SDP — and an average ticket delta without relevant price variation steps, generates greater effectiveness in the offer acceptance rate.”
— Maysa Santos - innovation specialist at Vero“The project showed that, when we combine data, artificial intelligence and a structured open innovation approach, it is possible to significantly increase the accuracy of commercial offers and capture concrete revenue and monetization results.”
— Maximiliano Carlomagno - partner at Innoscience
Amazon Web Services (AWS) is the world's most broadly adopted and comprehensive cloud platform, offering over 200 fully featured services from data centers globally. Millions of customers, including the fastest-growing startups, largest enterprises and leading government agencies, are using AWS to lower costs, become more agile and innovate faster.
With AI agents applied to the development cycle, DM saved 337 hours, boosted Product and Quality productivity by over 40% and standardized deliveries in just a few months.
ONR created IARI, a generative AI platform that automates the extraction and standardization of property registration data across real estate notary offices, with plans to reach more than 3,000 registries.
Tenda automates contract generation and eliminates 90% of the manual process
We use cookies to improve your experience on our site. Essential cookies are necessary for the site to function. You can manage your preferences or accept all cookies. Learn more
