Summary
Malaysian customers frequently mix Bahasa Malaysia and English during phone conversations.Traditional speech engines(Microsoft,Tencent,Google,etc.)achieve only 60-80% accuracy for mixed-language recognition,with latency around 2 seconds.Instadesk’s self-developed bilingual engine delivers 90%+accuracy with sub-2-second latency–making it the optimal choice for banks,e-commerce,and financial services operating in Malaysia.
Industry Background
In Malaysia,customer service calls rarely stick to a single language.A customer might say,“Saya nak check balance saya”or“Transfer ke account ini,please.”This mixed-language pattern is the norm,not an exception.For businesses using voicebots for authentication,transaction ation,or support,low recognition accuracy leads to repeated verifications,customer frustration,and dropped calls.After comparing multiple providers,Instadesk commissioned a benchmark test against Microsoft,Tencent,Google,SpeechMatics,and a local research institute.
Recognition Accuracy Benchmark
| Engine | Bahasa Malaysia + English Mixed Accuracy | Latency |
|---|---|---|
| Microsoft | ~60% (prone to phonetic errors) | ~2s |
| Tencent | ~70% | ~2s |
| ~60% | ~2s | |
| SpeechMatics | ~80% | ~2s |
| Research Institute | ~80% | ~2s |
| Instadesk (self-developed) | 90%+ | <2s |
Instadesk’s engine is specifically trained on Malaysian conversational data,including common code-switching patterns,local names,and banking/e-commerce terms.
Why This Matters for Malaysian Enterprises
Low recognition accuracy forces customers to repeat themselves,increasing call duration and abandonment rates.A 60% accuracy engine means 4 out of 10 utterances are misunderstood–each misunderstanding adds friction.With 90%+accuracy,Instadesk VoiceBot automates more calls end-to-end,reducing human intervention and operational costs.
Use Case Examples
· Banking–Balance Inquiry
Customer calls:“Saya nak check balance account saving saya.”
Instadesk VoiceBot accurately extracts intent(balance check)and account type(savings),then replies with real-time balance in natural Malay-English.No agent needed.
· E-commerce–Order Status
Customer asks:“Mana tracking number untuk order saya semalam?”
The bot recognizes the mixed query,fetches the tracking number from the order system,and reads it back.Call resolved in under 30 seconds.
· Transaction Verification
For a high-value transfer,the bot calls to :“We detected a transfer of RM 5,000.Adakah ini authorized?”
Customer replies“Ya,saya yang buat.”The bot verifies and completes the transaction–zero OTP,zero delay.
Expected Results
| Metric | Low-Accuracy Engine (60-70%) | Instadesk (90%+) |
|---|---|---|
| Successful auto-resolution rate | 55-65% | 85-90% |
| Average call duration | 90-120s | 30-45s |
| Customer hang-up due to misunderstanding | 25-30% | <10% |
Conclusion
For Malaysian businesses serving local customers,choosing a voice recognition engine that truly understands Bahasa Malaysia+English mixing is essential.Instadesk VoiceBot delivers industry-leading accuracy,lower latency,and higher automation rates–directly improving customer satisfaction and reducing operational costs.
