Welcome Note:

“Today I’ll show one practical AI use case that makes Ind AS 109 reporting faster and more reliable—Expected Credit Loss (ECL)—with a quick numeric example and the control points auditors care about.”

Slide 2 — Ind AS 109 (what changes in reporting)

Why it matters in financial reporting

Ind AS 109 impairment is forward-looking: recognize expected losses, not only incurred losses.

3-stage approach drives provisioning:

Stage 1: 12‑month ECL (no significant credit risk increase)

Stage 2/3: lifetime ECL (credit risk increased / credit-impaired)

Measurement expects: probability‑weighted outcomes, time value of money, and reasonable + supportable forward-looking info.

Summary:

“The reporting challenge is not the formula—it's the data + staging + scenarios + documentation every quarter.”

Slide 3 — Where AI fits in financial reporting (auditors will accept)

AI can support (not replace) judgment

Data extraction & standardization (from ERPs, loan systems, treasury, GL) → fewer manual errors

Risk segmentation and pattern recognition across large portfolios

Anomaly detection in GL / movements / overrides to flag unusual patterns for review

Narrative + disclosure assist (drafting, consistency checks) — with human validation

Summary:

“AI is most useful where volume is high and logic is repeatable: data prep, segmentation, early warning flags, and drafting support. Final decisions remain with management.”

Slide 4 — The chosen use case: “AI‑assisted ECL pipeline”

Problem (today)

Manual collation of PD/LGD/EAD, ratings, delinquency, collateral, macro-overlays

Staging rules applied inconsistently, heavy reliance on spreadsheets

Target state (AI‑assisted)

Automated data pipeline + rules engine for staging

AI flags: unusual transitions, outliers, override justifications needed

Reporting layer: ECL summary + what‑if/sensitivity

Key data typically needed

Internal/external ratings, PD/LGD/EAD, transition info, default history, macro inputs

Configurable staging and reporting/what‑if concept is commonly built into ECL tool designs

Summary:

“Ind AS 109 is data hungry. The use case is AI helps collect/validate inputs and highlight exceptions, while the ECL engine applies the accounting logic consistently.”

Slide 5 — Worked example (simple numbers for clarity)

Portfolio: 3 exposures (simplified ECL = EAD × PD × LGD)

(Note: real models discount cash shortfalls; this is a teaching simplification.)

Exposure

Stage

EAD

PD

LGD

ECL

A

Stage 1

₹1.00 cr

1.5% (12m)

45%

₹0.675 lakh

B

Stage 2

₹0.50 cr

12% (LT)

50%

₹3.00 lakh

C

Stage 3

₹0.25 cr

60% (LT)

65%

₹9.75 lakh

Total

₹13.425 lakh

Summary:

“Stage drives whether PD is 12‑month or lifetime. That single classification can change provisioning materially. This is why controls around SICR (Significant increase in credit risk) are critical.

Slide 6 — Where AI adds value in this example

A) AI‑assisted SICR (Stage movement)

Model flags exposures likely to have significant risk increase using features like delinquency trend, rating migration, restructuring flags, sector stress indicators

Many references discuss the practical use of thresholds/past-due indicators and rebuttals as part of staging discussions

B) Probability‑weighted scenario overlay For Exposure B (Stage 2), assume:

Base PD 12% (weight 60%)

Upside PD 9.6% (weight 20%)

Downside PD 18% (weight 20%)

Weighted PD = 0.6×12% + 0.2×9.6% + 0.2×18% = 12.72%

Revised ECL = ₹0.50 cr × 12.72% × 50% = ₹3.18 lakh

This aligns with the principle of probability-weighted outcomes and use of forward-looking information.

Summary:

“AI is not ‘deciding provisioning’; it’s helping you (1) stage consistently and (2) apply scenario logic with transparency.”

Slide 7 — Controls & governance

Non‑negotiables

Data lineage + DQ checks (completeness, duplicates, overrides, reason codes)

Model validation + independent review before and after deployment; governance and monitoring are emphasized in implementation discussions

Explainability: why did Stage change? why PD changed? who approved overlay?

Audit trail: versioning of inputs/assumptions/scenarios

Summary:

“If you present AI with this governance frame—human‑in‑the‑loop, documented overrides, reproducible runs—it lands well with both auditors and regulators.”

Slide 8 — Takeaways + how to start

3 takeaways

Ind AS 109 ECL is data + judgment heavy; AI helps with scale and consistency

Best early wins: data readiness + staging exception flags + scenario automation

Controls decide success: DQ checks + validation + audit trail

Starter checklist (say this quickly)

Do we have PD/LGD/EAD + rating migration + default history + macro variables?

Are SICR rules documented + consistently applied with rebuttal evidence?

Can we reproduce last quarter’s ECL with the same inputs/versioning? (audit trail)

Close:

“AI makes Ind AS 109 reporting faster and more defensible—when built with strong governance.”

Software

Wolters Kluwer OnesumX

KPMG’s Tool

ECL Square ( RVSBELL Analytics)

DILL Analytics ECL Tool

Python/R