Voice AI proposal · Delhaize Customer Support

The voice layer that pays for itself.

Cutting ramp time and attrition by taking the repetitive calls off your team — best-in-class voice, as a neutral layer on top of what you already run.

01 · Introductions
Daniël Schaart

Daniël Schaart

ElevenLabs — Enterprise Account Executive

I did my homework before this session. What follows is built specifically for Delhaize's Belgian CS operation — the numbers, the languages, the agents. Everything on this page is live: you can talk to both agents, and stress-test the business case in real time.

02 · Understanding · Where it hurts

Two numbers you told us cost real money.

52 days
for a new support agent to reach productivity
15%
of CS agents leave before their first year is out

Every AI vendor breaks at the same point: the moment a customer asks something real, the agent sounds robotic, stutters, or stalls. On a phone call, that's exactly where the business case collapses.

03 · The fix

We solve the quality layer.

Everyone else added voice to a chatbot. ElevenLabs started as the voice model — the thing we're best in the world at. On a phone call, that's the only thing that matters.

Best-in-class voice, as a neutral layer on top of what you already run.

Sub-second latency, real-time

State-of-the-art turn-taking — it understands pauses and "ums" instead of talking over people.

70+ languages, auto-detected

Dutch, French and English switched mid-conversation, no manual config. Built for a bilingual Belgium.

Neutral layer, no rip-and-replace

We sit on top of Genesys, Salesforce, or whatever you run today. No lock-in.

Proof · It's not theoretical

Regulated, high-volume brands already run us.

Revolut
Revolut resolves customer queries faster with our agents
Klarna
35M
Klarna customers served by us as first-line phone support

Retail runs on peaks. Black Friday triples your volume — you can't hire 100 agents who ramp in 52 days. A voice layer breathes with the volume.

Under the hood

Enterprise-ready, not a demo toy.

Quality & latency

Sub-second turn-taking model — natural, not robotic.

Languages

70+ with automatic in-conversation detection.

Knowledge — integrated RAG

Pulls live from Delhaize's own knowledge base. No hallucination.

Compliance

SOC 2 Type II · ISO 27001 · GDPR · EU data residency · Zero-Retention.

Deployment

Full inbound + outbound telephony, SIP trunking, fits your stack.

Scale

Built for hundreds of concurrent calls, reliably.

On-premise & data sovereignty

Run the voice model inside your own network — audio never leaves your perimeter. EU residency and zero-retention supported.

How we fit

On top of your stack. Not instead of it.

The answer to every platform objection: coexistence, not migration.

ElevenLabs — voice & conversation layer
Natural voice, 70+ languages, RAG, turn-taking. The one thing that makes agents work on the phone.
Your platform
Genesys · Salesforce Service Cloud · your CCaaS
Your systems of record
Order data · CRM · knowledge base
04 · Live demo

The quality carrying all of this.

Three Delhaize agents I built in ElevenLabs. Pick one below — then hit Start conversation.

Disclaimer: for demo purposes these agents speak English instead of Dutch (Flemish) or French. Production agents run natively in NL and FR with mid-call auto-switch.
Lisa
Lisa
Customer Service Agent
Live

Handles order tracking, faulty-product returns, and store hours — in Dutch or French, switching mid-call.

  • Order tracking with getOrderStatus
  • Faulty return → refund + replacement via processReturn
  • Store hours via getStoreHours
Model
GPT-4o · v2 turn-taking
Languages
NL · FR (auto-switch)
"Hi, you're speaking with Lisa from Delhaize. How can I help you today?"
Honest answer

Couldn't your own team build this?

Revolut — an engineering-first fintech — tried exactly that. Their prototype proved the concept, but productionising it (speech-to-text, language models, real-time turn-taking, all with PCI and zero-retention) proved a platform-scale lift. So they tested every major provider and chose ElevenLabs. When world-class engineers conclude it isn't worth building in-house, that's the strongest signal there is.

Where we're different

Same question, different center of gravity.

OpenAI

Powerful generic model, but a developer toolkit — not an enterprise contact-center platform. Telephony, compliance and RAG you build yourself.

We ship the whole layer, production-ready.
Sierra

Strong CX agents, but a full-stack platform that pulls you toward their way of working.

We're the neutral voice layer — you keep your stack.
Parloa

Solid contact-center player. The differentiator stays voice quality and latency under pressure.

Plus 70+ languages with live detection for a bilingual Belgium.

And unlike all of them: ElevenLabs embeds Forward Deployed Engineers to get you live — Revolut called them the most responsive partner they evaluated. You're not handed an API and left alone.

05 · Business case · live

Two effects, one payback.

Move the sliders. The numbers update in real time — we'll drop your actual figures in during discovery.

Assumptions
Delhaize CS baseline
Share of calls that are FAQ-like (order status, delivery slots, product info). Voice AI can resolve ~90% of these end-to-end — suggested deflection ceiling ≈ 63%.
Replacements + growth. Sam shortens ramp time for every new hire. Defaults to ~15% of the headcount as a starting point.
Platform pricing

LLM tokens billed on top of voice minutes. Telephony billed separately.

Net annual impact
1.689.705
2.033.853 gross savings − €344.148 ElevenLabs platform cost · 4.9× ROI
Calls deflected / yr
798.600
of 1.452.000 total
New AHT
204s
was 240s
Ramp time
21 days
was 52 days
Attrition rate
10.0%
was 15%
Ramp savings
€84.240
faster time-to-productivity
Attrition savings
€81.675
5.9 FTE retained
ElevenLabs platform cost
Enterprise subscription · 484.000 AI minutes / month
344.148
per year
Voice platform (minutes)321.706
LLM inference ($0.0042/min)22.442
Where the savings come from
Call volume · deflection + AHT1.867.938
Ramp time · Sam onboarding coach84.240
Attrition · retention lift81.675
Next steps

Prove it on your calls.

01

A 4-week paid pilot on one call type — order tracking, 500 live calls.

02

Success criteria agreed up front — containment measured on your own data.

03

Follow-up with your IT team to lock the integration scope.

Stan — does this take real load off your team? Ale — does it change your onboarding? Robin — does the business case hold?

Let's make Delhaize's phone line sound human.

Daniël Schaart · ElevenLabs — Enterprise Account Executive