← All rulings

ANS-100/P — BFSI Sector Calibration Note

Drafted 30 August 2026 · Ratified 30 August 2026, before the first banking, financial services and insurance company was scored · Rater 1 · Companion to ANS-100/P v2.1 (frozen anchor set)

0. Decision record — edition one proceeds single-rater (author decision, 30 Aug 2026)

The author has elected to skip the blind second rating for edition one. Recorded consequences and mitigations:

  1. The methodology page states plainly: edition one is a single-rater edition scored against published anchors with a full audit trail. No inter-rater reliability claim is made.
  2. Boundary markers persist on HCLTech, TCS, and Persistent — a second rating was the mechanism that could have resolved them; without it, the markers and stated ranges are the honest representation.
  3. Both statistical ties (HCLTech/TCS; TechM/Mphasis) publish alphabetically within band with ranges disclosed; no internal ordering claim is published.
  4. The rater-2 kit becomes the public replication kit: the blinded evidence packs and the clean v2.1 rule text publish under CC BY-SA with the standing line — "any reader can re-run these companies from the packs alone; divergences are invited to the corrections contact." The untaken test converts into an open invitation, which is the strongest available substitute.
  5. Recommendation preserved: run the inter-rater test before edition two; the 8-point tolerance stays in the methodology as the standard replicators are invited to apply.

1. Why this note exists

The LTTS file demonstrated that /P proxies are tuned to IT-services disclosure norms; scoring a new sector without pre-ratified application rulings would mean writing anchors after the scores they govern. BFSI (9 companies: HDFC Bank, ICICI Bank, SBI, Axis, Kotak, Bajaj Finance, HDFC Life, ICICI Lombard, Paytm) runs on a fundamentally different revenue model — income from spreads, fees, and premiums, not billed effort — so D6/D7 need explicit translation. These rulings apply v2.1 to BFSI; they do not amend it.

2. Proposed rulings (ratify individually: B1 ☐ … B10 ☐)

B1 — Q25 economic pair. RPE translates to total income per employee, read alongside the cost-to-income ratio trajectory. The A1 attribution test is unchanged: level 3 requires a disclosed number linking named AI systems to the income-per-employee or cost-to-income outcome. Credit-cycle and regulatory effects are named in the file as confounds, mirroring the M&A-confound treatment.

B2 — Q27 becomes genuinely scoreable. Banks and insurers disclose process-automation rates. Rule: digital-channel share (e.g., % of transactions digital) is NOT an automation rate — it measures customer channel choice, not no-touch processing. Q27 requires process-level rates: loans sanctioned without manual intervention, claims auto-adjudication %, straight-through onboarding/KYC rates. A5 grading (representative vs showcase) applies unchanged. Expect Q27's revival here after being nearly dead in IT services.

B3 — Q30 re-anchored. Bank revenue is not effort-linked, so the IT-services test passes trivially and would give free points. Re-anchored question: is revenue being re-founded on AI-era products and pricing? Level 2 = digital-first products material and disclosed; level 3 = AI-native products or AI-mediated pricing disclosed as revenue or volume lines (AI underwriting products, usage-priced APIs/BaaS, AI-personalised pricing). Undisclosed shares resolve downward (A2 logic).

B4 — Q31 exposure re-anchored. BFSI's AI exposure is disintermediation (AI agents and aggregators owning the customer interface), margin compression from AI-cheapened competitors, and adversarial AI/fraud. Mitigation = proprietary data moats, distribution lock-in, AI-native product lines, disclosed fraud-AI capability. Level 2 = exposure acknowledged with stated mitigation; level 3 = quantified mitigation.

B5 — Q33 translation. Couldn't-exist-pre-AI offerings for BFSI: AI underwriting/collections products with disclosed volumes, conversational banking at disclosed scale, real-time AI fraud prevention as a product, AI advisory. Distribution partnerships with AI firms are evidence of intent (level 2), not shipped offerings.

B6 — Security ceiling goes live. Production agent autonomy in regulated, money-touching flows with zero disclosed posture is exactly the condition the security ceiling was designed for. Ruling: where D2 evidence shows autonomy at level 3+ in payment/lending/claims flows and Q17–Q20 are all undisclosed, the file must explicitly evaluate the conditional ceiling rather than default to "partially assessable" — and the evaluation is shown in the file. Expect the ceiling's first live tests in this sector.

B7 — Q18 and the RBI baseline. RBI-mandated IT/cyber governance (board IT strategy committees, cyber frameworks) is a regulatory baseline = level 2 maximum, regardless of sophistication — it evidences compliance, not AI-specific posture. Level 3 under A3 requires AI-specific audited/certified posture (e.g., ISO/IEC 42001), AI surfaces in a public bug-bounty scope, or published AI-specific enforced controls. Stated alignment with the RBI's AI framework (FREE-AI) counts as level-2 specifics; audited implementation counts toward 3.

B8 — Q24 denominator. The A4 ≥5%-of-TTM-revenue bright line is measured against total income. Disclosed technology spend ratios count toward level 3 only where an AI-specific component is identifiable and recurring; generic "tech spend" is not an AI line item.

B9 — Q21 via BRSR. Banks disclose workforce training in BRSR filings; the A7 countable test applies unchanged. Generic digital-literacy training ≠ AI training; the file states which figure it used.

B10 — Entity and classification rulings. Banks are scored on the standalone bank with subsidiaries noted (HDFC Bank excludes HDFC Life/AMC/HDB — HDFC Life is separately in the universe). Paytm (One97) is scored as a BFSI payments company per the universe classification. Insurers' claims-automation disclosures are first-class Q27 evidence. SBI's scale and PSU disclosure norms get a file note, not a rule change.

Signal library additions for BFSI: 20-F/6-K filings (HDFC Bank, ICICI — SEC registrants), RBI regulatory disclosures, BRSR, investor-day technology sections, CISO/CTO conference talks, app-release evidence, bank-tech analyst coverage.

3. What does not change

D1/D3/D5 anchors; all P-caps; the evidence floor; the 18-month window; A1–A10 as ratified; ceilings' arithmetic; the language rule (evidence, never character). This note is application, not amendment — the freeze holds.

4. On ratification

Score BFSI in universe order (HDFC Bank first — pack already assembled, companion file). Each BFSI file cites the B-rulings it used. Sector synthesis after all nine, mirroring the IT slice.

Published 9 September 2026 under CC BY-SA 4.0 as part of the Edition One replication material. Status changed from "proposed" to "ratified" on publication: the ruling was ratified on 30 August 2026 before the sector was scored, and the draft filename had not been updated. No ruling text is changed by this publication.