Back to Blog
ROIOperationsCase Study

From 500 Agents to 150: The Math Behind AI Customer Service

12 February 20264 min readRelai Team

The Scenario

Let's model a real scenario. A mid-sized South African financial services company runs a call centre with 500 agents. Here are their current numbers:

  • **500 agents** across three shifts
  • **200,000 queries per month**
  • **Average salary:** R15,000/month per agent
  • **Fully-loaded cost per agent:** R28,000/month (salary + benefits + facilities + tech + management)
  • **Average handling time:** 8 minutes per query
  • **Average wait time:** 28 minutes
  • **CSAT score:** 67%
  • **Annual turnover rate:** 45%

Total monthly operational cost: R14M

Breaking Down the Queries

The first step in modelling AI impact is understanding what agents actually handle. Across financial services, the query breakdown is remarkably consistent:

  • **35% Balance and transaction enquiries** — "What's my balance?" "Did my payment go through?"
  • **20% Account services** — Address changes, card activation, PIN resets
  • **15% Product information** — "What are your interest rates?" "How do I apply for a loan?"
  • **10% Complaints and disputes** — Incorrect charges, failed transactions
  • **10% Technical issues** — App not working, online banking problems
  • **10% Complex advisory** — Loan restructuring, investment questions, fraud cases

What AI Can Handle

AI excels at the first three categories and handles parts of the fourth and fifth:

  • **Balance and transactions (35%)** — Fully automated. AI verifies identity and provides information in seconds.
  • **Account services (20%)** — 80% automated. Most address changes, card activations handled by AI. Edge cases escalated.
  • **Product information (15%)** — Fully automated. AI knows every product, every rate, every requirement.
  • **Complaints (10%)** — 30% automated. Simple complaint logging automated. Complex disputes escalated with full context.
  • **Technical issues (10%)** — 50% automated. Known issues and standard troubleshooting automated. Novel problems escalated.
  • **Complex advisory (10%)** — 0% automated. These require human judgment and empathy.

Total automation rate: 65%

The New Math

With 65% of queries handled by AI:

  • **Queries requiring human agents:** 70,000/month (down from 200,000)
  • **Agents needed:** ~175 (down from 500)
  • **Remaining agents handle:** More complex, rewarding work
  • **Agent satisfaction increases:** Turnover drops from 45% to 20%

Monthly Cost Breakdown — After AI

  • **175 agents × R28,000:** R4.9M
  • **AI platform cost:** ~R800K/month
  • **Total monthly cost: R5.7M**
  • **Monthly savings: R8.3M**
  • **Annual savings: R99.6M**

But What About Service Quality?

This is the surprising part. Service quality doesn't just maintain — it improves:

Speed - AI response time: <5 seconds (vs. 28-minute average wait) - Human agent response time: <2 minutes (vs. 28 minutes, because agents now handle fewer queries)

Accuracy - AI provides consistent, accurate information every time - No "I think the rate is..." or "Let me check with my supervisor" - Humans focus on complex queries where they add genuine value

Availability - AI operates 24/7/365 - No sick days, no shift transitions, no peak-hour bottlenecks

Languages - AI serves customers in 8+ languages - Previously, hiring Zulu, Xhosa, and Sotho-speaking agents was a constant recruitment challenge

Customer Satisfaction - CSAT improves from 67% to 89-92% - Customers get faster service in their preferred language - When they do speak to a human, the agent isn't rushed or overwhelmed

The Agent Experience

The 175 remaining agents aren't worse off. They're better off:

  • **Higher-value work:** They handle interesting, challenging cases instead of repeating "your balance is R2,450" 100 times a day
  • **Better tools:** AI provides them with customer context and suggested solutions before they even start the conversation
  • **Lower stress:** Manageable query volumes mean no constant queue pressure
  • **Career growth:** Agents become specialists rather than generalists

The Implementation Timeline

This transformation doesn't happen overnight, but it happens faster than most expect:

  • **Month 1-2:** Discovery, audit, and AI training
  • **Month 3:** Pilot deployment (10% of traffic)
  • **Month 4:** Scale to 30% of traffic
  • **Month 5:** Full deployment at 65% automation
  • **Month 6:** Natural attrition begins reducing agent count
  • **Month 9:** Target agent count reached without layoffs

Key principle: Reduce through attrition, not redundancy. With 45% annual turnover, natural attrition handles most of the reduction.

The Bottom Line

The math is straightforward:

  • **Investment:** R800K/month + R2M setup
  • **Return:** R8.3M/month in savings
  • **Payback period:** Less than 30 days
  • **12-month ROI:** 1,000%+

From 500 agents to 150 isn't about cutting jobs. It's about deploying humans where they create the most value — and letting AI handle everything else. The result is better service, lower costs, and happier teams.

That's the math behind AI customer service.