Background

In institutional investment banking, equity and fixed-income research reports must adhere to strict regulatory standards and internal compliance policies before publication. Supervisory Analysts (SAs) serve as the gatekeepers of quality, compliance, and accuracy.

As Principal Design Lead, I designed an AI-driven Compliance and Metric Validation Tool integrated directly into the research authoring and review environment. The system automates routine policy checks, explains rule violations with suggested replacements, and performs cross-metric validation between handwritten report content and underlying model source data.

Operational challenges

Supervisory Analysts operate under high pressure, serving as the final approval layer before research reaches institutional clients. Through contextual inquiry and workflow mapping with SAs, I identified three primary operational pain points SAs face on a daily basis.

Time-to-market pressure

Financial research loses value rapidly if delayed. Analysts must publish quickly following earnings calls or market-moving events. Approval delays directly impact the timeliness and client value of published research.

High workload and volume surges

Every research piece requires mandatory SA sign-off. During quarterly earnings seasons, report volume spikes exponentially, creating severe operational bottlenecks.

Repetitive manual verification

SAs spent hours manually checking repetitive elements.The review process was time-consuming and frequently led to tedious back-and-forth revision cycles with research analysts.

Product architecture & core features

To address these challenges, the platform introduces two core intelligence capabilities tailored to the SA review workflow:

01.  Intelligent risk scanning

Scans research content against internal compliance guidelines and regulatory frameworks, and identifies language that may carry risk or break internal policies. When content is flagged, the system provides rationale and suggested alternatives with instructions to bring the language into compliance.

02.  Metric discrepancy detection

Automatically extracts handwritten metrics from report text (e.g., Price Targets, Stock Ratings, Revenue Forecasts) and cross-references them against underlying financial model source data.Discrepancies are flagged where text numbers diverge from source figures, preventing costly publication errors.

Early explorations

To align product design with LLM architecture early, I worked at low fidelity to test how compliance risks and metric mismatches could be shown. Reviewing these early explorations directly with the prompt engineering team helped determine the structured JSON output.

Categorizing compliance findings

Tested a tab structure for grouping compliance findings into distinct categories.

Metric comparison

Tested showing numeric mentions by alongside their reference rating, and flagging inconsistencies.

Track changes approach

Tested pairing the AI’s compliance rationale with suggested edits and track changes.

Delivered interface

An AI-powered review workspace that highlights policy risks directly in the report, allowing reviewers to resolve language issues instantly and accelerate approval.

Types of flagged issues

Each card is standardized to provide immediate clarity for reviewers: a category badge and AI confidence score at the top, a clear rationale explaining the compliance risk, the exact flagged excerpt, and an actionable suggested resolution with one-click actions when appropriate.

Numeric mismatch

Detects discrepancies where handwritten metrics diverge from verified source data.

Rumor-related language

Flags unconfirmed market rumors or unsourced claims, prompting the author to provide sources or adjust the language.

Exaggerations

Identifies promotional phrasing, bias, or inflammatory claims, offering alternatives that maintain objective framing.