Real AI Use Cases in the Contact Centre with Luware Nimbus - Costs, Risks and Benefits

Marcel Gaufroid of Luware sets out a step-by-step route to AI in the contact centre with Nimbus, starting with post-call transcription and summarisation rather than customer-facing bots. He covers the three pillars of automation, augmentation and analytics, the 80% cost saving from post-call rather than real-time processing, and one insurance customer moving from 75% to over 90% routing accuracy with intent detection.

Marcel Gaufroid, Head of Sales – EU, at Luware shares a practical, step-by-step approach to implementing AI in the contact centre using Luware Nimbus, drawn from real-world experience with over a thousand customers.

• Luware's three AI pillars for the contact centre: automation, augmentation, and analytics

• Why post-call transcription and summarisation are the essential low-risk first step for any AI strategy

• How post-call processing slashes LLM costs by up to 80% compared to real-time transcription

• Building a feedback loop: using verified agent interactions to train virtual users over time

• Replacing traditional IVR with AI intent detection — achieving over 90% routing accuracy for one insurance customer

• Cost realities of deploying customer-facing virtual agents and why a step-by-step approach matters

Thanks to Marcel for the practical insights.

Chapters

• 0:53 Introductions and Luware's customer base
• 3:18 Three pillars: automation, augmentation, analytics
• 14:32 Do customers actually want to talk to a bot?
• Replacing multi-level IVR with intent detection
• 36:12 Start at the back: Companion and transcription first
• 36:42 Virtual user versus human agent terminology
• Building the labelled data set and feedback loop
• Bring your own LLM or flat-rate licensing
• 39:54 Why post-call processing costs 80% less
• Analytics chat client and trend spotting
• Compliance, consent and transcription versus recording
• Licensing, per-call costs and practical advice

Key insights

• Luware has over a thousand clients using its product on top of Microsoft Teams, spanning nonprofits, banks and insurers.
• Transcribing every interaction is the mandatory first step, converting human conversations into machine-understandable data.
• Processing transcripts and summaries post-call rather than in real time cuts LLM cost by a factor of five, 80% cheaper.
• One insurance customer replaced a multi-level IVR at roughly 75% routing accuracy with intent detection, reaching over 90%.
• After around three months of running Companion, customers had over 100,000 agent-verified questions and answers per service line.
• A full real-time virtual user using the latest GPT real-time models is estimated at around $2 per caller for triage alone.
• Luware uses "agent" for humans and "virtual user" for AI, and Companion is licensed per user while virtual users are consumption-based.
• Start with a low-risk queue such as an internal IT help desk, and keep the manual IVR as a one-click fallback.

Insights summarised by AI from the episode transcript, reviewed by the Empowering.Cloud team.