← Work

Meta — 2026

Meta — 2026

Re-imagining AI-driven investigations

Re-imagining AI-driven investigations

Protecting bad actors at scale is vital for Meta's platform. While machine learning handles bulk enforcement, complex threats require specialized human expertise, law enforcement partnerships, and cross-functional collaboration. I reimagined AI-led investigations to empower and accelerate these high-impact safety operations.

About this case study

The product I led at Meta is confidential, so the screens, and flows here are a ground-up recreation I designed to illustrate the real work. My role, the problem, the process, the timeline, and the outcome metrics are real.

Role

Role

Lead Product Designer

Lead Product Designer

Tools

Tools

Figma, FigJam, FigmaMake, VSCode, Claude Code

Figma, FigJam, FigmaMake, VSCode, Claude Code

meta_case_study_hero

TL;DR

TL;DR

As Lead Designer for Specialized Human Review, I drove an AI-first vision to unify 16 fragmented internal tools into a single platform for 15,000 monthly active investigators. Working closely with Product, Research and Engineering, we delivered a live pilot in just three months, cutting handling time from 55 to 4 minutes while keeping human experts in charge of final decisions.

As Lead Designer for Specialized Human Review, I drove an AI-first vision to unify 16 fragmented internal tools into a single platform for 15,000 monthly active investigators. Working closely with Product, Research and Engineering, we delivered a live pilot in just three months, cutting handling time from 55 to 4 minutes while keeping human experts in charge of final decisions.

KEY IMPACT & SCALE

3 Months

3 Months

Taken from concept to a fully shipped live pilot product.

Taken from concept to a fully shipped live pilot product.

Unified Experience

Unified Experience

Consolidated 16 internal investigation tools into one workspace.

Consolidated 16 internal investigation tools into one workspace.

93% Reduction in Handling

93% Reduction in Handling

Handling time down from 55 minutes to 4 minutes, projected to save 25,057 hours annually, with zero critical issues in the initial pilot release.

Handling time down from 55 minutes to 4 minutes, projected to save 25,057 hours annually, with zero critical issues in the initial pilot release.

Context

Context

  • The Product Suite: Because complex investigations can span months, they are nearly 40 times more expensive than automated reviews. This investment is increasingly critical, as the volume of these high-stakes investigations surged to over 12 million in 2025—a nearly 4x increase since 2023. The goal of this project is to enhance the overall investigator experience, increase decision accuracy, and drastically reduce operational costs.

  • The Environment: We operated in weekly cross-functional sprints across product, engineering, design, and research. To accelerate velocity, Design owned both the user experience and the front-end implementation, freeing Engineering to focus on complex backend infrastructure and integrations.

  • The Product Suite: Because complex investigations can span months, they are nearly 40 times more expensive than automated reviews. This investment is increasingly critical, as the volume of these high-stakes investigations surged to over 12 million in 2025—a nearly 4x increase since 2023. The goal of this project is to enhance the overall investigator experience, increase decision accuracy, and drastically reduce operational costs.

  • The Environment: We operated in weekly cross-functional sprints across product, engineering, design, and research. To accelerate velocity, Design owned both the user experience and the front-end implementation, freeing Engineering to focus on complex backend infrastructure and integrations.

The problem

The problem

Meta's investigative team faces fragmented tools, manual workflows, and rapidly increasing case complexity.

Pain points:

  • Severe Tool Fragmentation: Investigators are forced to navigate 16 tools, with information isolated in each, which leads to conflicting data and lost trust.

  • High Cognitive Load & Ambiguity: Users struggle to know which tool to use and how to use it effectively.

  • Friction in Case Handoffs: When cases are triaged or transferred to a new team member, the receiving investigator lacks historical context, current status, and next steps.

Meta's investigative team faces fragmented tools, manual workflows, and rapidly increasing case complexity.

Pain points:

  • Severe Tool Fragmentation: Investigators are forced to navigate 16 tools, with information isolated in each, which leads to conflicting data and lost trust.

  • High Cognitive Load & Ambiguity: Users struggle to know which tool to use and how to use it effectively.

  • Friction in Case Handoffs: When cases are triaged or transferred to a new team member, the receiving investigator lacks historical context, current status, and next steps.

Discovery & Design Sprint

Discovery & Design Sprint

I structured the initiative into two phases: qualitative research, then a high-intensity design sprint with stakeholders to create solutions.

1. Foundational Research & Audit

  • 1:1 User Interviews: Partnered with a user researcher to interview investigators and observe their live workflows.

  • Feature Audit: Conducted a comprehensive audit across existing investigation tools.

2. 3-Day Design Sprint & Ideation Workshop

  • Journey Mapping: Led collaborative mapping exercises to capture end-to-end user journeys and highlight operational friction points.

  • Problem Framing & “How Might We” Prompts: Converted journey map insights into actionable HMW prompts.

  • Crazy Eights & Rapid Concepting: Facilitated rapid sketching sessions to quickly generate diverse interface concepts.

Strategy & Approach

Strategy & Approach

Through our research and workshop we validated that an AI-powered solution was essential to optimize the workflow. Rather than building a supplemental sidebar chat, we determined that a standalone, unified investigation platform was required. The goal is to 10x investigator impact by automating manual tasks and surfacing actionable insights.

Design principles:

  • Build for Trust and Confidence: Features should deliver clear, transparent results and empower investigators to guide and improve the system.

  • Prioritize Accuracy and Quality: Deliver accurate results from the beginning and continuously improve with user feedback.

  • Empower Through AI assistance: Help investigators overcome uncertainty and quickly get started.

Through our research and workshop we validated that an AI-powered solution was essential to optimize the workflow. Rather than building a supplemental sidebar chat, we determined that a standalone, unified investigation platform was required. The goal is to 10x investigator impact by automating manual tasks and surfacing actionable insights.

Design principles:

  • Build for Trust and Confidence: Features should deliver clear, transparent results and empower investigators to guide and improve the system.

  • Prioritize Accuracy and Quality: Deliver accurate results from the beginning and continuously improve with user feedback.

  • Empower Through AI assistance: Help investigators overcome uncertainty and quickly get started.

The Solution & Development

The Solution & Development

To eliminate tool fragmentation, I designed a unified, end-to-end workspace that guides investigators through the entire case lifecycle: detection, case creation, investigation, response, and enforcement. Because investigations vary widely in complexity, I anchored the design on a "hero use case"—a foundational, cross-functional workflow that is simple to track and analyze, yet robust enough to scale for massive, complex cases.

I integrated an AI investigation agent that instantly analyzes data, surfaces hidden patterns, and recommends optimal actions. To meet security, privacy, and regulatory requirements, the system operates on a human-in-the-loop model. Investigators guide the AI throughout the process and retain absolute authority over the final enforcement decision. This intuitive, stage-based design ensures seamless onboarding, allowing anyone to jump in, grasp the status of an investigation, and know what to do next.

To eliminate tool fragmentation, I designed a unified, end-to-end workspace that guides investigators through the entire case lifecycle: detection, case creation, investigation, response, and enforcement. Because investigations vary widely in complexity, I anchored the design on a "hero use case"—a foundational, cross-functional workflow that is simple to track and analyze, yet robust enough to scale for massive, complex cases.

I integrated an AI investigation agent that instantly analyzes data, surfaces hidden patterns, and recommends optimal actions. To meet security, privacy, and regulatory requirements, the system operates on a human-in-the-loop model. Investigators guide the AI throughout the process and retain absolute authority over the final enforcement decision. This intuitive, stage-based design ensures seamless onboarding, allowing anyone to jump in, grasp the status of an investigation, and know what to do next.

CAPTION

A new AI investigation tool that consolidates and automates investigative workflows.

CAPTION

AI agents automate the entire lifecycle from detection to investigation and action, enabling analysts to quickly review and validate findings.

Outcome & Business Impact

Outcome & Business Impact

Delivered the pilot experience within 3 months, driving immediate adoption across Meta's investigation teams, with ongoing expansion planned to support additional teams and use cases.

  • Focused the initial release on the account lockout use case, a high-value opportunity for AI optimization.

  • Empowered self-serve configurations, allowing operations and investigation teams to make workflow adjustments independently.

  • Automated entity object inspection, allowing human investigators to focus solely on final verification and validation.

Delivered the pilot experience within 3 months, driving immediate adoption across Meta's investigation teams, with ongoing expansion planned to support additional teams and use cases.

  • Focused the initial release on the account lockout use case, a high-value opportunity for AI optimization.

  • Empowered self-serve configurations, allowing operations and investigation teams to make workflow adjustments independently.

  • Automated entity object inspection, allowing human investigators to focus solely on final verification and validation.

Next project

Next project

Crafting a world-class design system

Crafting a world-class design system

Steve Chen — Product Designer

© 2026. Designed & built by me