Verizon
A data aggregation tool that reads real-time public sentiment and outage data to identify churn-risk customers, pairing each signal with a mitigation strategy.
2025
Software Engineer, UX Designer
Full Stack, UX
Aahil Nishaad, Afnan Tuffaha, Luis Sarmiento, Vichu Selveraju, Kai Tjia
The Problem
Verizon reads churn from its own data (support tickets, billing, feedback forms), but most of that surfaces only after someone's already gone. The public signals that surface prior to churn, like reviews, social posts, complaint threads, and outage reports, go unreported. At a 1.13% quarterly churn rate on postpaid connections, that blind spot works out to more than $150M lost in a single quarter.
The Solution
The dashboard pulls from five public sources, runs NLP over them for sentiment, keywords, and topics, and ties every finding to a specific, definable action. The point was to stop reacting to churn and start treating it pre-emptively.
Sorry! Some of this work is proprietary. Please reach out for further details
Key Design Decisions
→ The only product voice on an engineering team. I was the only designer and product-oriented contributor who could also write code. Verizon's stakeholders described what they wanted in pretty abstract terms, and my job was turning that into features a very technical, non-product team could actually scope and build.
→ Designing for the role, not just the user. A two-week async questionnaire turned up something useful: regional and central analysts care about completely different numbers. That pushed the whole product toward role-specific workflows, the same regional-interface need Verizon later flagged as a priority.
→ Full pipeline under a two-week clock. Abstract client needs → async research → personas and scenarios → lo-fi → hi-fi in Verizon's brand system, presented to execs. Tight timelines meant parallelizing this process alongside development.
Outcomes
Pulled together 170,000+ public data points and surfaced churn drivers that Verizon's own reporting had never caught.
Takeaways
→ Being the only design and product voice on an engineering team was an exercise in translation. Reducing stakeholder requests into features and condensing tens of thousands of raw data points into a scannable screen became a critical skill.
→ I owned the NLP and analysis layer, sentiment, keywords, and topic modeling, with contributions to data sourcing, collection, and cleaning. Constructing the full pipeline across five messy public sources in four months required streamlined engineering to hit deployment deadlines.
Retrospective
→ Company sentiment is averaged on each post's overall label, so a post that criticizes Verizon while praising a competitor still counts as positive for Verizon. Per-entity sentiment was deferred to future work, and deprioritized to accommodate deadlines.
→ Topic modeling relies on a hand-built mapping from raw clusters to business categories. It works today, but as carriers rename plans and new complaints surface, that mapping needs constant manual upkeep.