# Proving Alpha: How a Hedge Fund Measures AI's ROI - Hari Kumar, Contour Asset Management

## Executive summary

Hari Kumar, Head of AI at Contour Asset Management, details the challenge of quantifying AI's Return on Investment (ROI) within a fundamental hedge fund setting. Instead of relying solely on Profit & Loss (P&L), Contour focuses on process-based metrics, such as 'cost per useful insight.' The firm employs live A/B testing of analyst research workflows (with and without AI agents) and is building unified platforms to integrate disparate data sources (e.g., Yipit credit card data, AlphaSense) and operational functions (trading, ops, compliance) to minimize repetitive work and maximize the value of existing data spend.

## Key takeaways

- Process-Based ROI Measurement: Measuring AI impact requires moving beyond P&L. Contour tracks metrics like 'cost per useful insight' and analyzes the time spent by analysts on specific tasks to determine if AI is fundamentally changing the workflow or merely adding 'thrust' to the existing process. (10:04)
- Live A/B Testing of Workflows: The firm conducts live A/B tests by documenting analysts' research processes both in a pre-AI world and when utilizing new AI tools, such as agents that synthesize expert call notes or field consumer surveys. This validates the efficiency gain of agentic workflows. (02:25)
- Agentic Idea Tracking and Timing: AI agents are used not only for idea generation but also for tracking ideas over long periods, ensuring that previously dormant or low-priority theses are resurfaced when market conditions make them relevant, thereby optimizing timing. (06:24)
- Platform Consolidation and Data Integration: A major focus is building unified internal platforms to integrate disparate operational workflows (e.g., order reconciliation, compliance) and combine multiple data sources (e.g., AlphaSense, Yipit) into a single, learnable system. (14:35)

## Technical details

- AI Workflow Testing: The firm uses live A/B tests to compare analyst output and efficiency when performing tasks (like expert call synthesis or consumer surveys) with and without AI tools/agents.
- Data Integration & Platforms: The goal is to build an internal platform that ingests and correlates data from multiple providers (e.g., AlphaSense, Yipit) and operational inputs (expert call transcripts, internal notes) to provide a holistic view of an investment thesis.
- Metrics and Observability: Key metrics include 'cost per useful insight' and tracking the time spent by analysts per name/task, linking time investment to eventual P&L outcomes.
- Operational Automation: Efforts are underway to consolidate fragmented operational workflows (e.g., order keeping, compliance) into a single platform to minimize repetitive manual work.

## Practical implications

- For build engineers, the concept of 'cost per useful insight' suggests implementing deep observability metrics that link resource consumption (time, compute, data access) directly to business outcomes (ideas generated, trades executed).
- The need to consolidate disparate systems (Bloomberg, AlphaSense, Yipit) highlights the necessity of building unified, API-driven middleware layers that abstract away vendor-specific limitations.
- The emphasis on A/B testing workflows provides a framework for measuring the incremental value of new features or AI models in a production environment before full deployment.

## Topics

AI ROI Measurement, Hedge Fund Technology, Agentic Workflows, Alternative Data Integration, Process Optimization, AIE CODE summit

Source: https://www.youtube.com/watch?v=ZMZlAn6Pyfg