All Case Studies
CS / 05TelecommunicationsVietnamese Tier-2 Mobile Operator

Deploying AI Voicebot to handle 62% of contact centre volume in 90 days

Vietnamese voicebot deflects 62% of tariff-enquiry and plan-signup calls — agents focus on churn saves and upsell, CSAT up 14 points.

Telecommunications · Vietnamese Tier-2 Mobile Operator
01
62%
Deflection rate
02
+14
CSAT points
03
90 days
Time-to-production
04
−41%
Cost per contact
Project timeline

Challenge → Solution → Outcome

Step-by-step view of how we moved from the brief to the outcome.

Step 01Challenge

A 380-agent contact centre handled ~28k calls/day, 70% of them repeat questions (tariff checks, data-plan signup, password resets). New agents took 6 weeks to train and turnover ran at 55%/year. Meanwhile churn was high because agents had no time for real save-the-call conversations.

Step 1 / 10
01 / 10
Story arc

Problem. Solution. Result.

01
Problem

A 380-agent contact centre handled ~28k calls/day, 70% of them repeat questions (tariff checks, data-plan signup, password resets). New agents took 6 weeks to train and turnover ran at 55%/year. Meanwhile churn was high because agents had no time for real save-the-call conversations.

02
Approach
  • 01Analysed 90 days of call logs and identified 11 intents covering 78% of volume.
  • 02Deployed a Vietnamese voicebot on Cloudfon.ai, integrated via SIP through the existing Aarenet SBC — no PBX replacement.
  • 03RAG against the internal knowledge base so the bot answers from the current policy version.
03
Result
  • After 90 days: bot handled 62% of volume autonomously, bot CSAT 4.3/5 (agent 4.4/5).
  • Agent headcount down 35%, reallocated to churn saves and outbound upsell.
  • Cost per contact down 41%, system payback in 7 months.
02 · Solution

Solution

  • 01Analysed 90 days of call logs and identified 11 intents covering 78% of volume.
  • 02Deployed a Vietnamese voicebot on Cloudfon.ai, integrated via SIP through the existing Aarenet SBC — no PBX replacement.
  • 03RAG against the internal knowledge base so the bot answers from the current policy version.
  • 04Escalation logic: bot hands off to an agent on negative sentiment or out-of-scope intent.
  • 05Real-time dashboard for team leads to track bot vs agent performance.
03 · Outcome

Outcome

  • After 90 days: bot handled 62% of volume autonomously, bot CSAT 4.3/5 (agent 4.4/5).
  • Agent headcount down 35%, reallocated to churn saves and outbound upsell.
  • Cost per contact down 41%, system payback in 7 months.
  • Bot runs 24/7 — solving the previously underserved overnight shift.
04 · Before / After

Measurable shift

01
Daily call volume handled
Before100% by 380 agents
After
62% bot · 38% agent
02
Average wait time
Before94 seconds
After
11 seconds
03
CSAT
Before3.2 / 5
After
4.6 / 5
04
Cost per contact
Before~17,500 VND
After
~10,300 VND
05
Overnight coverage (22h–6h)
BeforeVoicemail only
After
Full 24/7 bot
In their words
We were skeptical about Vietnamese voice AI — accents, slang, mid-call topic switches. The team ran weekly listening sessions on real recordings until the bot stopped sounding like a bot. By month three, customers were thanking it.
Director, Customer Operations·VN Tier-2 Mobile Operator (under NDA)
Tech Stack
Cloudfon.aiAarenet SBCVietnamese ASR/TTSRAG + LLMWebRTCSIP/RTP

Have a similar challenge?
Let's talk.

ZaloContact