Skip to content

Work / 01Shipped work

Role
Experience Designer II
Worked with
Designers, product managers and engineers
Year
2025
Status
Plan selection live on Total Wireless · activation prototyped

A plan in 2.8 minutes, not 15.

An agentic assistant for Verizon’s Total Wireless. The plan-selection conversation was A/B tested and is now live on the Total Wireless site. The activation flow — one conversation instead of a 12+ screen form, reading the IMEI, SIM and PIN off the box with the camera — was prototyped, not tested.

tello.ai screens: the assistant home, the guided conversation, camera scanning (prototype) and the account view

In 60 seconds

Problem

Activation meant hunting for a 15-digit IMEI across 12+ form screens; choosing a plan took about 15 minutes.

What I did

Proposed a guided agent instead of an open chat; with the team, designed its plan cards (A/B tested, now live) and a camera activation flow (prototype).

Result

+7%

conversion, with plan selection 5× faster (15 → 2.8 min)

A/B test · 2,400 users · 4 weeks

My call

I proposed a guided agent instead of an open chat: one question at a time, plans as cards, every choice pinned — trading a free-form first turn for a path nobody gets lost on.

My part

I wrote the conversation before any screen existed, then designed the stepper, rich cards and three scan flows (20+ states) that cut the steps a customer takes.

On this page

  1. 01The problem
  2. 02My role and scope
  3. 03How I led the work
  4. 04Key decisions
  5. 05Process
  6. 06The product, screen by screen
  7. 07Outcome
  8. 08What I learned
  9. 09What I’d do next

01

The problem

People dropped out at exactly the steps that asked them to copy numbers. The brief was an AI assistant — the risk was an open chat that leaves customers guessing what to ask.

02

My role and scope

  • Part of a team of designers, PMs and engineers. My focus: cutting the steps a customer takes and simplifying the experience — from auditing today’s flows to high-fidelity UI.
  • Proposed the guided-agent model — one question at a time, plans as cards, every choice pinned — over an open chat.
  • Wrote the conversation before drawing a screen, then designed the stepper, rich cards and three scan flows (20+ states).
  • Extended the assistant from plan selection into an agentic activation flow.

03

How I led the work

01

Set the direction

Proposed the guided-agent model over an open chat. It became the plan-selection flow the team A/B tested and shipped.

02

Wrote the conversation first

Scripted every turn before any screen existed, then designed the stepper, plan cards and 20+ scan states from that script.

03

Made the case to leadership

Took activation from a 12+ screen form to a working conversational prototype and presented it to leadership.

04

Lead on the team

Own UX and design-system work on the team and guide two junior designers.

04

Key decisions

Decision 01 · my proposal

A guided agent, not an open chat.

Customers arrive to do one job: pick a plan and a phone, then activate. Most don’t know what to ask an assistant.

Rejected

✕ Open chat — “Ask me anything”

People have to guess what to type, and the happy path gets lost in back-and-forth.

Chosen

✓ Guided agent with cards and pinned choices

One question at a time; plans arrive as cards and every choice stays pinned, while “ask anything” stays available.

Trade-off we accepted

Less free-form on the first turn, in exchange for a path nobody gets lost on.

Plans arrive as cards
Plans arrive as cards
The chosen phone stays pinned
The chosen phone stays pinned

Decision 02

The camera does the copying.

The slowest steps asked people to find and type a 15-digit IMEI, a SIM number and a plan PIN.

Rejected

✕ Keep typed form fields

Today’s flow: long numbers copied by hand, where people stall and drop out.

Chosen

✓ Scan the numbers off the box

The agent reads IMEI, SIM and PIN with the camera and fills them in.

Trade-off we accepted

Scans can fail, so every scan offers rescan, type it in, or reach a human — one tap each.

Scanning the IMEI
Scanning the IMEI
Entering the plan PIN
Entering the plan PIN

05

Process

  1. 01AuditMapped today’s flows: 12+ form screens
  2. 02ScriptWrote the conversation before the UI
  3. 03DesignScanning, stepper, rich cards · 20+ states
  4. 04Tell the storyPrototype for leadership

06

The product, screen by screen

From the first question to an activated phone: plan cards, phone selection, cart and checkout, then camera scans for the IMEI, SIM and PIN.

Home: one question, quick replies and four clear starting points
Home: one question, quick replies and four clear starting points
Plans arrive as cards inside the conversation
Plans arrive as cards inside the conversation
Choose lines with a stepper; the plan stays pinned
Choose lines with a stepper; the plan stays pinned
Phones that come with the plan, as cards
Phones that come with the plan, as cards
Bring your number with a keypad in the chat
Bring your number with a keypad in the chat
Review the cart without leaving the conversation
Review the cart without leaving the conversation
Checkout: log in or continue as a guest
Checkout: log in or continue as a guest
Activation: five steps, one question at a time
Activation: five steps, one question at a time
Scan the IMEI off the box
Scan the IMEI off the box
Scan the SIM card number
Scan the SIM card number
Enter the plan PIN
Enter the plan PIN
Set up the eSIM, step by step
Set up the eSIM, step by step
Activated: welcome to Total Wireless
Activated: welcome to Total Wireless

07

Outcome

A/B test · 2,400 users · 4 weeks

Time to choose a plan

before15 min
after2.8 min
+7%conversionA/B test vs existing flow
+12%customer satisfactionPost-purchase CSAT survey
−6%funnel drop-offA/B test vs existing flow

Method and source

Random split against the existing flow over 4 weeks; CSAT from a post-purchase survey. Four weeks can include novelty effects. The activation flow was prototyped, not A/B tested. These are team results; my part is under “My role and scope”.

08

What I learned

Agents still need structure
Steppers, pinned facts and cards are what make an AI flow feel trustworthy — not the chat itself.

09

What I’d do next

Check the lift holds
Four weeks can include novelty effects. I would re-read conversion after the novelty window before scaling the pattern.
Test activation
Put the camera activation prototype through the same A/B rigour as plan selection.
Next · 02 — Shipped workBang & Olufsen →Shaping in-car sound by feel, not equaliser.