Growth · Global BPO
Cost per trained support agent cut by about 75% with AI roleplay
Twelve AI customer personas now train support agents. Each trainer covers 200+ agents instead of 50, and new client accounts start in about half the time.


The problem
The BPO trains a large customer support workforce. Each trainer could cover about 50 agents, and ramping agents onto a new client account was slow. That delayed revenue on every new win.
What we built
A multi-persona AI training system where agents practise with AI customers before they take real calls.
- 12 customer personas, each with a named owner accountable for what it teaches and how it changes.
- Real call data feeds back into the personas weekly. Drift detection flags when a persona stops behaving like real customers.
- Customer data is masked before it goes into persona refinement.
- Bias and quality checks per persona, cohort and geography gate every session.
- Every interaction, grading decision and trainee response is logged and traceable.
- Confidence scoring on every persona response, with a kill switch to pull a persona that starts misfiring.
Results
- ~75% lower cost per agent trained.
- Trainer-to-agent ratio went from 1:50 to 1:200+, with quality holding.
- 4x training throughput on the same trainer headcount.
- Agent ramp time on new client wins cut by ~40%.
- New client accounts onboarded in roughly half the previous lead time, so revenue on new wins starts sooner.
- More consistent objection handling across geographies, lifting CSAT on the largest accounts.
- Zero training quality incidents traced back to bad persona patterns.
What you get · Growth
What this means for you
AI roleplay can onboard your new support or sales hires in weeks, not months, without a full-time trainer.




