CASE STUDY · Telecom · AI
Conversions via Conversation Design
A conversation design and AI personalization project that turned Verizon's deflection bot into a conversion driver.
Conversation design
AI Personalization
B2C
Move slider to see before vs after
Challenge
A context-aware conversational experience that personalizes every journey, educates users on bundled services, and creates seamless paths to conversion.
ROLE
As the Conversational & UX owner, I closely collaborated with CMI, Content, product and tech partners, came up with UX vision and drove alignment across team.
TEAM
UX owner (Myself) · Product manager · Lead UX · VDS & content · Legal · Tech & QA
Impact
· Scalable conversation design framework
· Personalized customer journeys
· Updated product road map
· Reduced churn
The problem
Verizon's chatbot was optimized to deflect support tickets - not convert sales.
Massive Top-of-Funnel Drop-off
Up to 95% drop rate across early digital conversion steps (availability checks, plan matching, and initial qualification), leaving conversion rates in the single digits.
Over-Reliance on Call Centers
Up to 70% of total sales still occur through non-digital or assisted channels, with 25% of inbound calls focused purely on sales inquiries (plans, deals, pricing clarity).
High Post-Sale Care Burden
Human agents spent hours resolving highly repetitive, structured tasks like Fios return labels and simple bill checks.
The goal
Our North Star
Success meant fewer users bailing out to a live agent and more prospects and existing customers converting on chatbot.
Reduce unnecessary agent escalation on chatbot.
HOW I got there
Approach
Given the complexity, scale and vested interests in the project, I approached it from a holistic as well as collaborative mind-set, whilst keeping the user at the core.
What I Specifically Did
Framework Optimization
After leading a cross-functional workshop and auditing the live flow, I designed a five-step structure that replaced the one-size-fits-all menu.
AI system design
How AI does the work
AI isn’t a feature here - it’s the layer that makes one conversation feel different for every user. Three specific moments where I designed the AI’s role:
01
Intent Detection
NLP classifier routes the first message into one of 12 intent buckets — bot responds to what was said, not what a menu predicted.
I defined the 12 buckets and what each one unlocks
02
Session Context
For signed-in users, the bot reads plan tier, device age, and open tickets. Responses reference their actual situation.
I designed the response templates and signal hierarchy
03
Adaptive Flow
Conversation paths branch on purchase readiness, history, and open cases. I defined the decision tree; AI fills the branches. They can handle deflections equally.
I owned the branch logic end-to-end
The Human Layer Behind the AI
ML TEAM
Built the intelligence that identifies user intent.
Trained the intent classifier
Selected and optimized the NLP model
Maintained inference infrastructure
Conversation UX (My Role)
Designed how the product responds once intent is identified.
Defined all 12 intent categories
Designed decision logic and branching
Wrote contextual response patterns
Prioritized account signals for personalization
AI predicts the user's intent.
I designed what happens next.
System flow: how intent maps to action.
proposal
Secured Alignment from Partners
The biggest challenge was aligning business, product, and engineering around a new direction.
Some UI elements that research identified as low-value were considered critical by partner teams. Instead of debating individual components, I reframed the conversation around user goals and demonstrated how the redesigned flow could better support both customer needs and business outcomes.
By visualizing the future experience through iterative prototypes, we reached alignment and unblocked the project.
Methods: Low-fidelity wireframes · High-fidelity prototypes · Cross-functional design critiques
Highlights
3 rounds of presentations and multiple iterations.
Outcome
Key partners aligned,
Extended partners directionally aligned,
Shared partner goals.
Learnings
It is key to convey potential opportunities by creating a scalable UX framework and advocating to users all through-out.
Outcome
Shipped a Context-Aware AI Assistant at Scale
We successfully deployed the baseline conversational engine across both the Verizon App and Dotcom ecosystems, replacing thousands of static FAQ pages with interactive, guided
Metrics represent project outcomes shared by stakeholders and are presented at a high level due to confidentiality.
Continuously Enhancing Chatbot Journeys Through Research
We use a continuous cycle of data analysis, customer feedback, and usability testing to identify friction points and optimize every step of the chatbot experience.
