Speech Analytics for Compliance Monitoring at Lenders in Vietnam
Speech analytics lets Vietnamese lenders monitor up to 100% of customer calls automatically — transcribing Vietnamese conversations, then flagging mis-selling, missing disclosures, and abusive collection against the State Bank of Vietnam (SBV) consumer-finance conduct rules. It replaces sampling a handful of calls with full, consistent coverage.
This is an explainer, not legal advice.
Why does call compliance matter for Vietnamese lenders?
Lenders in Vietnam — banks and licensed finance companies — sell and service credit largely over the phone: sales calls, collection calls, and servicing queries. The conduct on those calls is regulated. Circular 43/2016/TT-NHNN on consumer lending by finance companies, as amended by Circular 18/2019/TT-NHNN, sets out how a finance company may urge and recover debt. It caps debt-reminder contact at a maximum of five times per day, requires reminders to follow the method agreed in the contract and to fall within permitted hours (no earlier than 7 a.m.), prohibits threats, and does not permit contacting people other than the borrower or a stated guarantor.
Consumer-protection duties reinforce this. Vietnamese law expects credit institutions to disclose product terms clearly and treat borrowers fairly, and personal data on every call sits inside the perimeter of Decree 13/2023/ND-CP on Personal Data Protection and the Personal Data Protection Law effective from 1 January 2026. On a ₫30 million consumer loan, the way the product is described, whether fees are disclosed, and how a borrower in difficulty is treated all sit inside the compliance perimeter.
The practical problem: a manual quality-assurance (QA) team can only listen to a small sample of calls. If a team reviews 2% of calls by hand, 98% of conduct risk goes unseen.
What is speech analytics and what does it check?
Speech analytics — a form of conversation intelligence — transcribes every recorded call and analyses the transcript for defined patterns. Instead of a human sampling, software reviews the whole population of calls consistently.
On a lending call, it can check for signals such as:
- Mandatory disclosures — were the interest rate, fees, and key terms actually stated?
- Mis-selling language — pressure tactics, overstated benefits, or promises not backed by terms.
- Prohibited or abusive conduct — threats, profanity, calls outside permitted hours, or contacting third parties in breach of the consumer-finance rules, especially on collection calls.
- Consent and identity — was the borrower verified and did the required consent language occur?
- Complaint cues — a customer expressing dissatisfaction that should trigger the grievance process.
Because it is automated, the same rule is applied to every call the same way — the consistency a manual sample cannot give.
Sampling vs full-coverage monitoring
Aspect | Manual QA sampling | Speech analytics |
|---|---|---|
Calls reviewed | A small sample (often ~1–5%) | Up to 100% |
Consistency | Varies by reviewer | Same rules every call |
Languages | Limited by staff | Vietnamese, including dialects |
Time to flag a breach | Days after the call | Near real time |
Audit evidence | Notes on sampled calls | Full, searchable record |
How does AI help lenders monitor compliance?
YuCI is conversation intelligence that transcribes and analyses every customer call in Vietnamese, then scores each against the lender's own compliance checklist. It flags missing disclosures, mis-selling language, and abusive-collection cues so a compliance officer reviews exceptions rather than random samples. Patterns across thousands of calls — a particular script, product, or agent driving breaches — become visible and fixable. Because every call is transcribed and searchable, the lender holds concrete evidence for internal audit and for demonstrating conduct oversight to the SBV. Coverage moves from a fraction of calls to the full population, without expanding QA headcount. This builds on what speech analytics is in banking. This is a general explainer, not legal or compliance advice.
How does this support SBV consumer-finance conduct expectations?
Circular 43/2016/TT-NHNN and its amendment Circular 18/2019/TT-NHNN direct finance companies to reminder and recover debt within strict limits — capped frequency, permitted hours, no threats, and no contacting third parties. Speech analytics does not interpret the law for a lender, but it gives the evidence base to show those obligations are being met at scale. A compliance team can demonstrate that calls are monitored comprehensively, that breaches are caught and remediated, and that agent coaching is driven by real data. Personal data in call recordings must also be handled under Decree 13/2023/ND-CP and the 2025 Personal Data Protection Law. That is a materially stronger position than a sampled spreadsheet — the same discipline that governs AI voice agents for loan collections and the broader set of voice AI use cases in retail banking.
FAQ
Does speech analytics replace the compliance team? No. It reviews every call and surfaces exceptions, but a compliance officer investigates and decides. It scales the team's reach rather than removing human judgement.
Can it handle Vietnamese and regional accents? Yes. Vietnamese customer calls span northern, central, and southern accents and heavy diacritics, and speech analytics is built to transcribe and analyse them, including code-switching within a single call.
Is this legal advice on SBV rules? No. This is an explainer, not legal advice. Speech analytics provides monitoring and evidence; interpreting Circular 43/2016/TT-NHNN, Circular 18/2019/TT-NHNN, and related issuances for your institution is a matter for your legal and compliance functions.
How much of our call volume can be monitored? Up to 100%. Unlike manual sampling, automated analysis reviews the full population of recorded calls with the same rules applied consistently.
What kinds of breaches can it detect? Missing mandatory disclosures, mis-selling or pressure language, calls outside permitted hours or beyond the five-per-day limit, contact with third parties, consent and identity gaps, and complaint cues that should trigger the grievance process.
How does it help during a regulator review? Every call is transcribed and searchable, so the lender can produce concrete evidence of monitoring, breach detection, and remediation rather than relying on a small sampled record.
Conclusion
For Vietnamese lenders, conduct risk lives on the phone — and manual sampling leaves most of it unseen. Speech analytics turns full-coverage monitoring into a practical reality, giving compliance teams the evidence and the reach to meet SBV consumer-finance conduct expectations at scale.
Move from sampling to full call compliance coverage. Talk to the YuVerse team.
References
- Circular 43/2016/TT-NHNN — consumer lending by finance companies — https://english.luatvietnam.vn/circular-no-43-2016-tt-nhnn-dated-december-30-2016-of-the-state-bank-of-vietnam-prescribing-consumer-lending-by-finance-companies-112176-doc1.html
- Circular 18/2019/TT-NHNN — amending Circular 43/2016/TT-NHNN — https://english.luatvietnam.vn/circular-no-18-2019-tt-nhnn-dated-november-04-2019-of-the-state-bank-of-vietnam-on-amending-supplementing-some-articles-of-the-circular-no-43-2016-178364-doc1.html
- Decree 13/2023/ND-CP on Personal Data Protection — https://english.luatvietnam.vn/decree-no-13-2023-nd-cp-dated-april-17-2023-of-the-government-on-personal-data-protection-249791-doc1.html
- State Bank of Vietnam (SBV) — https://www.sbv.gov.vn/
