CX Today Cutting Through the AI Hype: Here's How to Actually Measure What Matters
9 snips
Mar 26, 2026 Rémi Guinier, Head of AI Product at Diabolocom, specialises in AI for contact centres with a focus on quality monitoring and model calibration. He explains why generic models fail in noisy real-world calls. He outlines Shapeable AI that supervisors can configure. He covers auto-calibration using golden datasets and the three signs of a successful AI project.
AI Snips
Chapters
Transcript
Episode notes
Generic Models Hide The Data You Need
- Generic AI models act like black boxes and often hide intermediate outputs needed for measurement.
- Rémi Guinier says lack of transparency plus vendor overpromising prevents contact centers from proving real business impact.
Voice Benchmarks Don't Reflect Contact Centers
- Off-the-shelf voice models are trained on ideal datasets and fail on noisy, interrupted contact center audio.
- Rémi highlights benchmarks like Flores and Common Voice don't reflect real-life interference, latency, and interruptions.
Prefer Small Specialized Models For CX Tasks
- Use smaller, specialized models for focused CX tasks like summarization or quality grids to cut latency and hallucinations.
- Rémi recommends tailoring models to the exact output needed (e.g., structured summaries) to boost speed and accuracy.