# Bilingual voice AI for an auto-finance call centre | YAID

> The routine 80% of call volume automated end to end in English and French, live in production.

Source: https://youraidept.com/case-studies/auto-finance-call-centre

[All case studies](https://youraidept.com/case-studies) Case study · Automotive captive finance

# Bilingual voice AI for an auto-finance call centre

A white-label customer-service operator automates the routine 80% of its call volume in English and French, without losing control of its own roadmap.

## 01 The situation

Roughly 80% of the operator's call volume is routine and scripted: payout quotes, banking and address changes, payment history. The system of record is a 1970s green-screen mainframe, and six months of negotiation with the incumbent platform's AI vendor had gone nowhere. Leadership wanted to be, in their own words, "masters of our own destiny," not locked into a stack that could be obsolete in six months.

"It got very clear when we looked at how many calls we get a day, and what percentage are about changing personal information on their files. It happened to be, like, 80%." L.M. President

## 02 What we built

Bilingual voice AI on Claude Sonnet 5 that resolves ~79% of total call volume end to end, in English and French.

Airtight authentication designed to a zero false-acceptance standard, hardened with one-time passcodes for bank-partner compliance.

Seamless human handoff via custom middleware, so a warm transfer hands the agent the caller's identity, intent and summary. No one starts a call cold.

Compliant and live hosted in-region for data residency, rolled out canary-first on a single brand, now running in production.

## 03 Before and after

The cost, before

~$44 CAD / routine call

On a capacity basis, each scripted call effectively costs the business ~$44 to resolve.

~$2.75M CAD / year

Skilled bilingual-agent time tied up in ~62,000 scripted calls that follow a playbook.

1970s mainframe dependency

A customer-facing operation exposed to a legacy stack and a stalled vendor, with no roadmap control.

30 to 45% annual attrition

Scripted, low-agency calls are the work that burns agents out.

The impact, after

~$21 CAD / resolved call

Effective cost per resolved call projected to fall from ~$44 to ~$21.

~$1.45M CAD / year, projected

Salary savings at the mid-case, with 60% of routine calls resolved by AI.

Client-owned voice stack

An independent stack running alongside the mainframe, with the team trained to own the roadmap.

~19 agents redeployed

Bilingual agents freed from scripted calls to move onto more complex work.

~5 months

Projected payback

The build is modelled to pay for itself in roughly five months, after which the recurring savings compound.

In their words

"We'd already taken the easy 30% out of this operation years ago. What we didn't think was possible was going after the rest of it without hurting the customer experience. That's exactly what this will do."

L.M.

President

Smaller project? [Hire a Claude-certified engineer](https://youraidept.com/hire) on YAID Hire: one brief to up to three engineers, each with published prices.

## Ready to put AI to work?

[Get your free AI audit roadmap](https://youraidept.com/audit-roadmap)