Call Conduct Analyser
Checks recorded sales calls for disclosures and fair treatment.
Disclosure check F1
0.93
Pressure-selling flag precision
0.71
Unlogged complaint recall
0.84
QA review coverage
2% to 100% of calls
About this model
Reviews recorded sales and service calls for fair-treatment and conduct checks: required disclosures read (fees, cooling-off period, risks of investment products), no pressure selling, vulnerable-customer cues, and complaints expressed but not logged. Works on English, Bahasa Malaysia and code-switched calls, and points the QA reviewer to the exact timestamp.
Intended use
Selecting and pre-reviewing calls for Quality Assurance and Compliance, replacing a 2% random sample with risk-based coverage of every call. Findings are reviewed by QA before any coaching or conduct action.
Training data lineage
Two stages: transcripts from the contact centre speech-to-text model on contact-centre-audio, then a fine-tuned multilingual classifier for 14 conduct checks, trained on 9,200 QA-scored calls and on complaints-corpus cases about mis-selling. Trained and registered in Vertex AI; audio stays in the Malaysia region.
Limitations & bias notes
Inherits transcription errors, especially on noisy calls. Checks based on tone (pressure, frustration) are less reliable than checks based on words. Not designed for chat or branch conversations.
Ownership and sensitivity
Who approves access
- Owner, Contact Centre, Siti Hajar Ismail
Business-sensitive. Models and data products scoped to named business units.
Entitlement per business unit, approved by the owner; conditions attach.
- Customer PII
- Contains or processes personal data about customers (PDPA 2010).
Try it
Live sandboxsales_call_specimen_2203.wav
fixtureEvaluation only: restricted models run against synthetic or masked sample data until your business unit's access request is approved.
Feedback
Details
- Updated
- 2026-07-22
- Latest version
- 1.3.0
- Licence
- BU Restricted
- Access
- Restricted
- Framework
- Transformers
- Language
- Bilingual