My Framer Site

Scalable Cross-business-unit Search Design

Designed a data-driven search framework that adapts to Cross-business-unit needs to drive conversions

Business Context

Context

JD.com's Search Results Page is a critical product module:

Major traffic entry point: Most users arrive with clear shopping intent

Primary conversion path: Majority of purchases happen through search

High business impact: Significant contributor to total GMV, Serves millions of daily active users

The Opportunity

Business units (Auto Parts, Pharmaceuticals) requested search improvements to support their growing operations. Combined with platform-wide efforts to optimize conversion rates, this presented a clear opportunity to enhance the search experience for specific categories and peak shopping periods.

The Challenge

JD.com's search results page serves 150M+ daily users across 20+ business units — but struggled to balance:

Unified UX

VS.

Category-specific needs

Data-driven design

VS.

Stakeholder preferences

Design consistency

VS.

Seasonal flexibility

Team

Product Manager

Lead UX Designer

UX Designer

UI Designer

Business Units

Skills

Strategic Problem-Solving

Data-driven Design

Cross-Functional Communication

Designing Scalable System

Outcome

Validate the positive impact of iterations through UCTR and UCVR values

Timeline

2023 Q3

Module

Tool

Figma

Framework & Constraints

JD's search page follows a "intent progression" model (developed based on extensive data analysis and behavioral insights) — as users scroll, content becomes more targeted based on behavior data.

My design mandate: Work within this existing framework and optimize at component level—not rebuild the page structure—in order to maintain consistency across experiences.

This constraint shaped my approach by focusing on modular, flexible components that could adapt to different needs while respecting the proven framework architecture.

With this foundation established, the next step was determining where to focus design efforts for maximum impact.

Scope & Design Approach

Product team identified two high-impact opportunity areas based on business unit requests and search volume data. Working within the framework constraint, I structured the design approach into two parallel tracks—each requiring different methodologies due to their distinct problem characteristics:

Scope

Track 1: Category-Specific

Auto Parts

Pharmaceuticals

Track 2: Promotion-Specific

Holidays & Festivals

E-commerce Promos

Design Approach

Cross-functional qualitative research

Method:

1. Stakeholder Alignment

2. Category Deep-Dive

3. User Validation

→ Goal:

Design component variants tailored to category-specific needs

Data-driven behavior analysis

Method:

1. Historical Data Review

2. Behavioral Insight Mining

3. Content Strategy Alignment

→ Goal:

Create dynamic systems that adapt to temporal contexts

Design Principles

Modular

Component-level flexibility

Consistent

Shared design language

Data-backed

Every decision validated

Testable

A/B test before full launch

Design Solutions

Auto Parts

INSIGHTS

Pre-Purchase

Finding Compatible Parts

Challenge:

To search for compatible parts, users must first input their vehicle model. Current flow requires selecting through multiple pages: Brand → Series → Year → Model

What we found:

Users find this tedious and give up
"Too many steps, I’d rather browse first."
High drop-off at vehicle selection stage

Mid-Purchase

Understanding Product Options

Challenge:

Data analysis revealed two distinct user types with different needs:

Type 1: OEM-only buyers (brand-conscious)

Only trust original factory parts
Problem: Hard to identify genuine OEM products
Often confused by aftermarket alternatives

Type 2: Spec-matching buyers (price-conscious)

Want cheaper options that fit their vehicle
Problem: Don't understand specification numbers e.g., tire spec "215/60R16 95V"

Post-Purchase

Service Integration

Challenge:

After purchase, users need installation services:

Most auto parts require professional installation

Users don't know where to go

Business opportunity:

JD Auto BU operates offline service center chains
These centers needed more exposure
Opportunity to create one-stop experience

Solutions

License Plate Auto-Detect

Users enter their license plate → the system auto-fills the vehicle model
One tap to filter compatible parts
Save vehicle for future searches

Filter Compatible Parts Only

Original Brand and Sepc

Spec decoder

“OEM” and “Compatible” Label

OEM Labels & Specification Decoder

For OEM buyers:

Clear "Original" tags
"Original only" filter option

For Spec-matching buyers:

Spec decoder
"Compatible" tags

O2O Bundled offers

O2O Service Integration

"Product + Installation" bundled offers
Nearby JD service center locator
Service promotions in search results

Pharmaceuticals

INSIGHTS

Unclear Needs

Finding the Right Medication

Challenge:

Search data revealed users rarely search by drug names. Instead, they search by their health needs:

Scenario 1: Symptom-based searches

"arthritis"
User doesn't know specific diagnosis
Needs guidance on which medication to choose

Scenario 2: Condition-based searches

"fever medicine"
Clearer diagnosis, comparing options
Needs to understand different treatments

Scenario 3: Chronic medication needs

"blood pressure medicine"
Regular purchaser, knows what they need
Focuses on price and repurchase convenience

Unclear Needs

Needing Expert Guidance

Challenge:

Many users are uncertain about:
What's causing their symptoms
Which medication is appropriate
Whether they need professional advice

Business opportunity:

JD Pharma partners with hospitals
Offers online doctor consultation services
Service had low awareness and utilization

User quote:

"Usually when I go to the pharmacy, the pharmacist will recommend one or two medicines based on my symptoms, weight, and health conditions. I rarely go in already knowing which specific drug name to buy."

Clear Needs

Delivery Speed & Quality Concerns

Challenge:

Two very different purchase behaviors:

Urgent, one-time needs:

Acute illness (fever, pain, cold)
"I need it NOW"
Primary concern: delivery speed
Less price sensitive

Chronic, repeated purchases:

Daily medication (blood pressure, diabetes)
"I need to stock up"
Primary concerns: unit price, bulk discounts
Will wait for better deals

Solutions

1: Symptom-based searches

Subcategories based on symptoms + “Ask a Pharmacist” support.

Doctor Q&A

2: Condition-based searches

Medications for common illnesses + product effects and best-sellers.

3: Chronic medication needs

Targeted meds + dosage guidance + ingredients + long-term treatment options

Context-Aware Recommendations & Integrated Doctor Q&A

For acute medications

For chronic medications

Adaptive Information Hierarchy

Nearby Pharmacies

Holidays & Festivals

Surge Demand

Spike in Contextual Keywords Search

During public holidays and seasonal festivals, search behavior shifts—users often enter contextual keywords (e.g., “Mid-Autumn gifts”) instead of item names or categories. These searches are not well supported today, leaving untapped opportunities for conversion.

Unmet User Needs

Keywords Only Match Product Titles

During holidays and festivals, users search with contextual keywords (e.g., "Mid-Autumn gifts") instead of product names.

"I want health and outfit ideas for Winter Solstice, but search result shows random winter titled children books..."

What we heard

Insight

High demand + unmet needs = opportunity

Users have strong cultural shopping habits tied to specific holidays,

but our search doesn't align with these contextual search needs.

Strategy A

Public Holidays with Heavy Promotions

Mid-Autumn, Spring Festival

BU curates special gift collections
Heavy marketing investment
→ Strategy: Surface promotional features and use-case driven shopping lists

Strategy B

Seasonal Celebrations without Promotions

Winter Solstice, seasonal health themes

Users want themed recommendations
No specific BU promotions
→ Strategy: Editorial-driven exploration

Solution1

For promoted holidays

Use-case driven shopping lists
Featured gift collections
Category-specific promotions

Solution2

For seasonal themes

Editorial recommendations
Themed product curation
Lifestyle-driven discovery

landing page

landing page

Promotional Events

Unmet User Need

High Search demand for features and event pages

During major promotional events (6.18, 11.11, Black Friday), users search for event pages and features—often influenced by offline ads or livestreams.

Common searches:

"618 home appliance sale page"
"11.11 Luoyong Hao’s Live Stream Room"
Specific feature names e.g. customer service

Problem:

Current search logic treats these as product queries, returning generic product results instead of the curated event pages and features users are seeking.

Poor Marketing-to-Search Conversion

Broken Funnel from Offline Ads to Online Purchase

Context:

All business units + platform invest heavily in offline advertising (subway, billboards, TV)
Significant spend on online marketing (social media, livestreams, influencer partnerships)
Ads primarily promote event landing pages and livestream rooms

The gap:

Users see offline/online ads → Come to platform → Search for advertised content → Can't find it → Drop off

We're failing to capture marketing-driven traffic because search doesn't connect ads to destinations.

Insight

User demand + marketing investment both pointing to the same problem:

Search needs to surface event pages and features, not just products.

Solution

Direct search access to event pages and promotional features

Direct links to event landing pages
Feature shortcuts in search results
Campaign-specific entry points
Integration with livestream events

search for Live Stream page

search for function

Results After Launch

We launched the feature with A/B test

UCTR uplift

1.5

UCTR = Click PV / Search UV

diff = A/B - 1

UCVR uplift

1.8

UCVR = Total Order Lines / Search UV diff = A/B - 1

Insights: Users in the treatment group showed strong engagement with model-based filtering, confirming demand for personalized recommendations. Interestingly, some users initially disabled the “recommend accessories by vehicle model” option, only to re-enable it later—indicating a trust-validation step before committing to system suggestions.

Next Steps: We plan to run qualitative research to uncover the motivations behind toggling behaviors and refine the recommendation logic to build stronger user confidence.

Filter car accessories by vehicle model