Upwork · Talent Scout · Design Lead

Turning no-shows into
engaged clients.

40% of scheduled recruiter meetings ended in no-shows. Clients signed up, booked a call — then ghosted. The intake flow was supposed to be Upwork's premium on-ramp, but it was bleeding qualified leads before a single conversation happened. I redesigned the flow end to end in 5 weeks.

Role
Design Lead
Team
PMs, Engineering, Data Analysts, Content Designer
Duration
5 weeks
Tools
Figma, UserTesting.com, Miro
Process
Research → Ideation → Prototype → Test → Iterate → Ship
01 — The Problem

The revenue leak nobody owned

Talent Scout is Upwork's premium recruiting service — pre-screened, skill-certified candidates for high-value roles. The intake flow was the front door: clients describe their needs, book a recruiter call, get matched. Simple in theory.

In practice: 40% of booked meetings ended in no-shows. Another ~20% dropped off on the sign-up page itself. Three hypotheses emerged: the sign-up asks too much too soon, the wait between sign-up and first meeting kills momentum, and the intake experience doesn't earn the client's investment.

40%
no-show rate on booked meetings
~20%
drop-off on the sign-up page
02 — Research

Two hiring mindsets, one broken flow

I interviewed 8 potential clients and found the intake flow was failing both groups — but for opposite reasons:

The Scrappy Hirer — budget-tight, time-poor
  • Hard to hire talent outside their area of expertise
  • Hard to hire for a time-bound project
  • Hard to find someone meeting their budget as a small business
  • Hard to quickly find the best match
The Vetter — skills exist, but can't evaluate fit
  • Easy to find talent with specific hard skills, but can't evaluate soft skills
  • Not easy to hire in a reasonable time frame
  • Needs someone to screen and pre-vet talent on both soft and hard skills
03 — Usability Testing

Six screens, six friction points

Usability testing on UserTesting.com revealed something worse than a single broken step — the friction was systemic. Every screen in the flow had its own failure mode:

Category selection screen
Category Selection
  • Left and right columns increase cognitive load
  • Low-domain clients unsure of role categories
  • No back button to return to Talent Scout landing page
Role selection screen
Role Selection
  • Non-tech-savvy clients unsure of needed roles/skills
  • Clients wanting a website may not know they need an HTML developer
  • No search corrections for misspellings
Skills selection screen
Skills Selection
  • Clients unaware of remaining steps in the process
  • Willing to provide more info if it helps the recruiter
  • Need visual confirmation — selected skills vanish into a "cloud"
Sign up screen
Sign Up
  • Too much info required upfront
  • Clients unclear on why info is needed
  • Illustration doesn't match non-development categories
Confirmation screen
Confirmation
  • Users lose interest/patience with excessive info requests
  • Some drop off due to phone number requirement
Meeting scheduler screen
Meeting Scheduler
  • Small businesses worry TS favors big companies
  • Want clarity on first meeting benefits
  • Mobile users missed date/time picker initially
  • Clients want to share more about the role
04 — Ideation

One question that reframed the entire flow

I mapped the full journey with stakeholders — every touchpoint, every drop-off, every emotional beat. From that map, the team voted on HMW questions. The winner became our north star:

"How might we collect as much information from the client as possible before the recruiter call, without it feeling too cumbersome?"

The journey map surfaced a tension: recruiters need detailed info to prepare good matches, but every extra question pushed clients closer to abandoning. The design had to thread that needle — gather more while feeling like less.

User journey map
05 — Problem Decomposition

Three problems, decomposed

I brought engineering into the room early — before any wireframes. Together, we decomposed the friction into three solvable problems:

Problem 1: Cumbersome sign-up
  • Sign-up page is the biggest drop-off point
  • Simplify the sign-up process
  • Provide auto-fill for country
  • Pre-fill email and name
Problem 2: Role confusion
  • Low-domain clients don't know what role to hire
  • Let users type anything, use data science to match relevant roles
  • Guide to another service if role doesn't fit our categories
Problem 3: Pricing uncertainty
  • Clients want to know how and what they should pay
  • Provide min-max hourly rate range
  • Notify if budget is below TS range
  • Show estimated rate based on role, skills, location
Sprintathon session
06 — Architecture

An IA with no dead ends

I rebuilt the information architecture with engineering, ensuring every path led somewhere useful. The key moves: budget/timezone questions upfront (so recruiters arrive prepared), a "skip all" escape hatch for low-domain clients, and real talent previews to sustain curiosity mid-flow.

Information architecture Low-fidelity prototype
High-fidelity prototype
07 — Iteration

The test that flipped the flow order

Think-aloud testing revealed something we hadn't anticipated: clients' mental model was backwards from our flow. They wanted to talk to a recruiter first, then provide details — not the other way around. That single insight triggered four strategic iterations:

Iteration 1 — user flow reorder
Iteration 1
User Flow Reorder
Changed from category → intake → create account → scheduler to category → scheduler → intake → create account — matching clients' mental model of getting help from a recruiter first.
Iteration 2 — merged input fields
Iteration 2
Required Info Merge
Merged input fields (name, company, email) into the meeting scheduler. This helps clients understand why the info is needed and auto-fills the sign-up form afterward.
Iteration 3 — summary page
Iteration 3
Info Confirmation
Added a summary page as a "waiting room" for optional information — giving clients control without pressure.
Iteration 4 — cognitive load reduction
Iteration 4
Cognitive Load Reduction
Removed the interactive map and profile samples during intake questions to keep users focused. These elements appear later in the journey instead.
08 — Shipped Product

The shipped flow

Final design overview
Final design — flow 1 Final design — flow 2 Final design — flow 3
09 — Outcomes

What changed

The flow reorder was the unlock. Once clients booked a recruiter call before the detailed intake, every downstream metric moved:

73%
improvement in client conversion
40→16%
no-show rate reduction
3d→1d
recruiter talent-match time

The real lesson: We assumed the problem was content — too much, too confusing. It wasn't. The problem was sequence. The same questions that felt burdensome before booking a call felt reasonable after — because now the client had a reason to invest. Order creates meaning.

What's next

Let's close
a gap together.

I'm looking for teams where research drives the roadmap and design ships — not just specs.

lin@linzhao.design