# A Hedge Fund Tests AI With AB Tests, Not Hype _Contour Asset Management runs live AB tests and tracks cost per useful insight to see if AI actually improves hedge fund research._ **Published:** 2026-10-24 **Source:** https://www.startuphub.ai/ai-news/artificial-intelligence/2026/a-hedge-fund-tests-ai-with-ab-tests-not-hype --- Hari Kumar told [AI Engineer](https://www.youtube.com/watch?v=ZMZlAn6Pyfg) that his 6 billion dollar fund is running live AB tests to see if AI actually improves research, not just speeds it up. Contour Asset Management, Kumar said, has 15 to 20 people and about 10 analysts covering 30 to 40 names each, so any gain has to show in process before it shows in P&L. The test is deliberately simple. Kumar said the team documented a full three month cycle from one earnings report to the next without any AI tools, logging data sources and conclusions, then ran the same name again with new workflows added one by one. Examples include agent run expert calls with and without agent synthesis, and agent fielded consumer surveys versus human ones, to measure how long sourcing takes and whether conclusions change. The fund tracks a metric it calls cost per useful insight. Kumar said that metric combines insights generated per analyst per day, time spent per name, and eventual P&L, which at a fundamental long short shop arrives over months not days. To feed it, Contour is layering fragmented data inside an internal platform. He described piping in credit card panel data from Yipit and call transcript data from [AlphaSense](https://www.startuphub.ai/startups/alphasense) via MCP and API access, so agents can combine those datasets with expert notes, rather than leaving each in its own portal. He framed that as an alternative to buying a bundled system like Rogo or Kensho, with the internal build learning each analyst's model pattern. The same consolidation idea extends to trading, where agents track entry and exit timing and cross industry exposure, and to operations and compliance, where he said the goal is to replace piecemeal [Bloomberg](https://www.startuphub.ai/ai-news/artificial-intelligence/2026/ai-trade-us-iran-peace-talks-on-bloomberg-surveillance) and order management tools with one reconciled workflow. What the numbers do not show yet is an outcome. Kumar was explicit that idea generation is slow to validate and that timing often matters more than the idea itself, so agents also watch stale notes and flag when a shelved thesis becomes interesting again. He also flagged a cost the ROI sheet often misses, the time spent fixing model or connector breaks every few days, which he said is falling but still offsets gains. Without a disclosed change in hit rate, time saved per insight, or cost per insight over quarters, the improvement is process described not profit proven. That gap matters because Contour is not a small allocator. The firm disclosed a 4,710,570 share position in BILL Holdings worth 256.9 million dollars after adding 1,472,132 shares last quarter, according to Markets Daily. A fund with that scale, Kumar argued, needs a unified layer that learns continuously across research, trading and ops. What would change the read is a time series for cost per useful insight by analyst and by tool, or a before and after view of entry timing. Until then the open question is whether building a per analyst platform beats renting one, and whether faster synthesis actually moves the three to twelve month alpha Kumar is chasing. --- Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory. © StartupHub.ai. All rights reserved. You may not republish this article in full without a license. Search engines and AI research tools may crawl and summarize for reference. Bulk reproduction or model training on this content requires a license. See https://www.startuphub.ai/terms.