Hari Kumar, head of AI at Contour Asset Management, explains how the investment firm measures AI adoption in this 18-minute AI Engineer talk.

Compare the work before claiming the return

  • Contour documents analysts’ research with and without AI tools, including the sources found, conclusions reached and time spent.
  • One comparison follows a company between quarterly earnings reports; tools are introduced separately to see which steps change.
  • An agent can conduct an expert interview while a person writes the synthesis, or produce the synthesis for the analyst to assess.
  • Investment returns can take months to emerge. Faster research is an intermediate measure, not proof of better investment performance.

Track cost per useful insight

  • Kumar connects analyst time, ideas generated, tools used and eventual profit or loss.
  • His proposed unit is cost per useful insight, with follow-up on which ideas mattered and where they led.
  • The talk describes a measurement process without publishing a numerical return on AI spending or a formal definition of “useful.”

Bring old ideas back at the right time

  • Agents revisit research notes, compare dormant ideas with market changes and alert analysts, portfolio managers and traders when conditions change.
  • The goal includes challenging an analyst’s initial thesis with new evidence, alongside generating additional ideas.
  • Combining interview notes with spending and other purchased datasets can expose connections that isolated subscriptions miss.

Include operations and maintenance

  • Trading tools track entry and exit timing and positions across companies; Kumar wants traders to contribute research ideas too.
  • Operations and compliance work spans separate systems for orders, reconciliation and vendors. Contour is working toward a shared workflow.
  • An internal platform can adapt existing data subscriptions to individual analysts’ habits and financial models.
  • Broken model calls and connectors consume staff time. Count repairs, interruptions and learning costs alongside the time an AI tool saves.

Caveat: these are Contour’s own practices and plans. The talk reports no controlled performance results establishing investment gains from AI.

“You can’t show ROI immediately.” — Hari Kumar, 5:43