Counter-Memo: measure and break analysts' anchoring on AI-drafted credit and investment memos

AI drafting of credit and investment memos is now crowded: S&P, Moody's, Evalueserve, Reply, Arya and many generic memo generators all do it. Bear-case generation exists as a feature (Suprmind's red-team and debate modes, Reply's worst-case scenarios). What I could not find is a product that treats the reviewer, not the memo, as the thing being tested. Two pieces look open: planting known-random leans in shadow cases to measure each team's follow rate, and logging whether a reviewer opened the opposing case and changed anything, packaged as evidence of effective challenge for model-risk and credit-risk functions. Practitioner writing from Kovrr, Corporate Compliance Insights and others argues that agreement and override rates should be measured, but I found no shipped product doing it for credit memos. The timing helps. SR 26-2 (April 2026) reportedly places generative AI outside the core model-risk guidance and tells banks to build a separate framework. That leaves model-risk teams looking for concrete controls that show effective challenge, and the hardest part to evidence is human review of outputs that can't be traced. There are risks. The drafting incumbents could add a "challenge" tab cheaply, so a new entrant should win on independence (it sits outside the drafting vendor, which is itself a model-risk argument) and on rigorous seeded-lean experiments the drafting vendors have no reason to run against their own product. I could not verify the "Anthropic 86%" follow-rate figure, so don't build marketing on it until the primary source is confirmed. RIA compliance is a weaker buyer than bank model-risk teams.

• Suprmind

  The closest product match. It runs an investment thesis through several models and has a bull-vs-bear debate mode plus a red-team mode that flags deal-breakers. It is a tool for the analyst, though. It does not log whether reviewers engaged with the challenge, and it is not sold to model-risk or compliance teams as evidence.

• Reply AI Credit Memo

  A prebuilt credit memo app whose forward-looking agent runs best-case and worst-case scenarios on the memo's key pillars. That is partial overlap with an opposing-case feature built into the drafting tool. I found no reviewer-behavior logging.

• S&P Global Credit Memo Builder

  An agentic credit memo platform launched June 2026. It is a drafting incumbent that could add a challenge step. For Counter-Memo it is the workflow to plug into, and possibly an acquirer, more than a direct competitor today.

• Moody's Credit Memo

  An automated credit memo product used at large European and APAC banks. Moody's stresses that analysts stay in control. It is another drafting incumbent with no evident challenge or anchoring measurement.

• Evalueserve credit memo agents

  A custom agent pipeline for drafting credit memos, including validation and source-tracing agents. The validation checks facts, not reviewer independence.

• Kovrr (human-in-the-loop review measurement)

  A blog post, not a product I could confirm. It argues for measuring override rates, how they vary between reviewers, and how they drift over time as signs of rubber-stamping. That is the measurement idea behind Lean Test, written up by an AI risk vendor.

• Lumenova AI

  An AI governance vendor publishing frameworks for governing generative and agentic AI after SR 26-2, which reuse effective challenge and ongoing monitoring. A likely adjacent competitor for the model-risk budget, but it governs at the model level, not the level of each memo's review.

• Stack AI / V7 / Taskade memo generators

  Generic tools for building investment memos. Some include a risk-flag section, but none generates an adversarial counter-memo or measures reviewer behavior.

Checked 2026-10-28