JPMC Blended Search (AI-Assisted Unified Search)

Enterprise Systems
Written By B K

Role
Senior UX Designer

Domain
Enterprise Systems, Data Search, AI-assisted workflows

Skills
Information architecture, search UX, stakeholder alignment, complex logic mapping, prototyping, content strategy

Overview

Party Central originally offered three separate search tools—Generic Search, Advanced Search, and “Advanced New” (later clarified as ID Resolution). Each tool had different logic, behaviors, and mental models, forcing Ops users to jump between interfaces and interpret inconsistent results.

My goal was to unify all three into one cohesive, scalable search experience that supported both quick lookups and deep investigative workflows. As part of the redesign, we also introduced Party Central’s first AI-assisted search, enabling Ops teams to retrieve results using natural-language queries.

The Problem

Search was one of the most-used but least-aligned features in Party Central:

  • Users had to remember which tool to use for which task

  • Logic and filters were inconsistent across the three tools

  • ID Resolution searches required switching contexts entirely

  • No single entry point gave users confidence they were “searching correctly”

  • Results pages varied in structure, terminology, and table layout

  • No support existed for AI or guided-search behavior

The experience needed to be consolidated, clarified, and built for scale.

My Role

As the lead UX designer for Blended Search, I was responsible for:

  • Unifying the interaction model for all search modes

  • Creating a scalable framework that handled entity, company, and relationship queries

  • Designing for AI-assisted search and integrating LLM outputs into existing logic

  • Mapping eligibility rules and search behavior across dozens of scenarios

  • Aligning with Ops, Product, Engineering, Data Strategy, and the AI Search team

  • Redefining search categories, filters, and result hierarchies

  • Prototyping flows and detailed microinteractions

  • Supporting testing and business stakeholder demonstrations

This was a multi-month initiative requiring deep understanding of operational workflows, regulatory constraints, and record governance.

Research & Insights

Several insights shaped the final design:

  • Users don’t think in terms of which tool to use; they think in terms of “I need to find this record.”

  • Case sensitivity, partial matches, and relationship paths were not clearly explained in any existing UI.

  • Users required quick access to external authoritative sources (GLEIF, BVD, Bloomberg).

  • ID Resolution had strong logic behind it but felt disconnected from the rest of search.

  • Most users began with simple queries (“Apple”, “Acme Holdings”) but needed to drill down quickly.

  • AI-assisted search worked well for interpretation, but users needed to clearly understand what the AI was doing.

These insights guided a unified structure with clear modes and far less cognitive overhead.

Design Approach

1. A Single Search Bar for All Modes

Instead of forcing users to choose a tool first, the experience begins with one universal search bar, similar to modern enterprise and consumer search patterns.

From here, users can:

  • Perform a standard keyword search

  • Toggle on “ID Resolution Mode”

  • Use “Expert Mode” for advanced filters

  • Enable AI-Assisted Search

  • View previous searches and saved filters

This allowed the system, not the user, to interpret the appropriate approach.

2. Clear, Guided Search Modes

Based on user research, the redesign introduced:

  • Standard Mode: For fast, keyword-based lookups

  • Expert Mode: For structured, parametric searches

  • ID Resolution Mode: Specifically for ambiguous or conflicting records

  • AI Search: Natural-language interpretation + suggested queries

Each mode feels part of one system—not a separate tool.

3. Modernized Results Layout

I introduced a standardized table structure with:

  • Clear hierarchy of entity vs. company vs. relationship types

  • Row-level preview details

  • Flags, statuses, and data-quality indicators

  • Quick actions (View, Compare, Link, Resolve, etc.)

This reduced the need to open multiple windows and jump around the app.

4. External Source Integration

Because Ops often verifies records using external sources, the new search includes:

  • Quick links to GLEIF, BVD, Bloomberg

  • Inline “source comparison” guidance

  • A persistent panel for external verification tools

This dramatically reduced tab-switching fatigue.

5. AI-Assisted Search Integration

Working with the AI Search team, we designed:

  • A natural-language search bar

  • Suggested queries (“Did you mean…?”)

  • A results rationale panel (“AI searched for X because you asked for Y…”)

  • Guardrails to prevent hallucination or overreach

  • A clear distinction between AI-generated guidance and official records

AI became an assistant—not a decision-maker.

Key Design Decisions

  • Avoided a “wizard” approach; opted instead for inline progressive disclosure

  • Kept filters visible but lightweight

  • Standardized terminology across the entire app (“Party,” “Record,” “Relationship”)

  • Added search history and saved filters for repeat Ops tasks

  • Built the system to scale — additional search modes can be added without restructuring

  • Designed in a way that doesn’t disadvantage users who prefer keyboard or expert-level workflows

Solution

Blended Search Homepage

  • Universal search bar

  • Mode toggles (Standard / Expert / ID Resolution / AI Search)

  • Recently viewed entities

  • Suggested tasks (Search by name, by ID, resolve duplicates, etc.)

Results Page

  • Unified results table

  • Filters drawer

  • Rationale panel (for AI mode)

  • Inline actions (View, Compare, Resolve, Link)

  • Relationship visualizations where relevant

ID Resolution

  • Decision panels that show why two records are or aren’t matches

  • Confidence indicators

  • Inline document references

  • Clear Maker-Checker handoff compatibility

Impact & Results

  • Eliminated the need to switch between 3 separate tools

  • Simplified logic and reduced user confusion

  • Provided a single, scalable foundation for future search enhancements

  • Enabled Ops to complete tasks with fewer clicks and less context switching

  • Introduced AI capabilities without adding risk or complexity

  • Improved trust in search results through clearer rationale and logic

This work also accelerated JPMC’s broader goal of modernizing Party Central into a unified operational platform.

What I Learned

  • Search UX is only as good as the clarity of its mental models

  • Users need predictable behavior more than they need advanced features

  • AI assistance must be transparent and interpretable to be trusted

  • Enterprise search requires balancing flexibility with safety and compliance

  • Close engineering partnership is crucial when dealing with complex logic

Artifacts (Optional)

  • Early search-mode diagrams

  • Logic flow for ID Resolution

  • AI rationale panel explorations

  • Table design iterations

  • Filter drawer patterns

  • Before/after comparison of the three original search tools

B K
Previous

JPMC Bulk Upload (Maker–Checker Workflow Redesign)