Models / SME Lending

MSIC Industry Coder

Assigns MSIC 2008 industry codes from a business description.

InternalBNM reportableMinimal riskLanguage AI · 1.2.0 · msic-industry-coder
172 13K/30d 61

Top-1 accuracy (5-digit)

0.814

Top-3 accuracy (5-digit)

0.936

Section-level accuracy

0.962

Median latency

35 ms

About this model

Assigns a five-digit MSIC 2008 industry code to a business from its free-text description of activities, in English or Bahasa Malaysia, with the top three candidates and confidence. Keeps industry coding consistent across onboarding, credit concentration reporting and regulatory returns.

Intended use

Suggesting MSIC codes at SME onboarding and during portfolio clean-up; sector concentration reporting. The onboarding officer confirms the code.

Training data lineage

Fine-tuned multilingual sentence encoder with a classification head, trained on 64,000 business descriptions and confirmed codes from sme-portfolio-mart in BigQuery, plus the public MSIC 2008 code descriptions. Trained and registered in Vertex AI.

Limitations & bias notes

Conglomerates with several activities get the code of the dominant activity described. Very short descriptions ("trading", "perniagaan am") return low confidence and are sent to manual coding.

#msic#industry-classification#sme#onboarding#bahasa-malaysia#regulatory-reporting

Ownership and sensitivity

Owned by

Business BankingSME Lending

Accountable owner: Lim Wei Jie

Who approves access

No approval needed. Self-service: subscribe and use, logged for chargeback.

Internal

Open to all staff. Aggregates, policy text, public series and synthetic data.

Self-service: subscribe and use, logged for chargeback.

BNM reportable
Feeds a regulatory return or a figure reported to Bank Negara Malaysia.

Try it

Live sandbox

Feedback

Details

Updated
2026-06-15
Latest version
1.2.0
Licence
Group Reuse
Access
Open to all BUs
Framework
Transformers
Language
Bilingual