Raisetalk8 January 20268 min read

Mistral AI: Aiming for Technological Independence

Raisetalk enriches its offering with Mistral AI, the French generative AI champion. A new step in our independence strategy from LLM providers.

Mistral AI and Raisetalk's technological independence

🚀 New: Mistral now available on Raisetalk

Raisetalk enriches its offering with Mistral AI, the French champion of generative AI. A new step in our strategy of independence from LLM providers.

Why offer multiple AI engines?

When entrusting the analysis of thousands of customer conversations to artificial intelligence, the question of technological choice is not trivial. Which model to use? The one offering the best quality-price ratio? The most powerful in your sector? The most respectful of your sovereignty constraints?

At Raisetalk, we've made a clear choice: not to choose for you.

Our LLM-agnostic approach rests on a simple conviction: no model is universally superior. Each has its strengths, pricing specifics, and variable performance depending on use cases. By giving you access to multiple engines, we enable you to:

Test and compare results on your own data
Optimize your costs based on volumes and analysis criticality
Avoid vendor lock-in that would make you dependent on one actor's strategic or pricing decisions
Meet your compliance requirements by choosing providers aligned with your internal policies

Models available on Raisetalk

With Mistral AI's arrival, our platform now offers 11 models from three major providers:

Provider	Model	Positioning
Google	Gemini 2.0 Flash	Fast and economical
	Gemini 2.5 Flash	Balance performance/cost
	Gemini 2.5 Pro	High performance
Anthropic	Claude Haiku 3.0	Ultra-fast, simple tasks
	Claude Haiku 3.5	Fast, versatile
	Claude Sonnet 3.5	Quality/speed balance
	Claude Sonnet 3.7	In-depth analysis
Mistral AI	Ministral 3B	Lightweight, high velocity
	Ministral 8B	Versatile, good quality-price ratio
	Ministral 14B	Optimized performance
	Mistral Small 3.2	Latest generation

This table will evolve as new models are released.

Mistral AI: French Tech in service of conversational analysis

Integrating Mistral AI isn't just about technical diversification. It's also a response to growing demand from our customers and prospects.

Mistral AI, founded in 2023 by former researchers from Meta and Google DeepMind, has established itself in just months as the European flagship of generative AI. Based in Paris, the company develops models recognized for their efficiency and openness. For French and European organizations, choosing Mistral can address several challenges:

Digital sovereignty: your data is processed by European technology
Proximity: an actor subject to the same regulations (GDPR, AI Act)
Ecosystem support: contributing to local technological champions' development

How does it work concretely?

On Raisetalk, model selection is done simply at the level of each analysis or in your default settings. You can:

Define a default model for all your analyses
Test different models on the same conversation sample
Mix approaches according to your use cases (a fast model for sorting, an advanced model for detailed analysis)

Our team can also assist you in identifying the optimal configuration based on your volumes, budget, and quality requirements.

Try it yourself

The best way to judge is to test.

Our trial space allows you to analyze your own conversations with the engine of your choice: https://app.raisetalk.com/try

Insurance in France: what a defensible duty-to-advise file looks like in an ACPR inspection

For insurers, brokers and agents selling insurance in France, particularly by telephone. Since recommendation 2024-R-03, the duty to advise must be proven in writing for all insurance products, and in May 2026 the ACPR ruled that an oral exchange is not enough. Yet it is listening to the calls that brings advice forms down. The four documents of a defensible file, the consistency test between the written record and the call, and what an analysis tool does, or does not do, for you.

Continuous security monitoring at Raisetalk

Strategy 19 August 2026 15 min

Is Raisetalk secure? What we measure, fix and publish

No open vulnerability, thirteen compliance frameworks in the green, and audit reports available on request from the footer. This article describes the machinery behind those claims: six surfaces analysed continuously, an application firewall in blocking mode, and a process in which no alert is ever closed by a machine. It then compares this approach with ISO 27001 certification and penetration testing, which complement it, but which come afterwards.

French law of 11 August 2026 on telephone canvassing

Strategy 11 August 2026 18 min

France, 11 August 2026: cold calling moves to prior consent, and the burden of proof changes sides

Since 11 August 2026, calling a consumer in France without prior consent is prohibited and Bloctel is gone. Law no. 2025-594 and Decree no. 2026-662 impose four mandatory disclosures at collection, consent valid for twelve months with no tacit renewal, proof kept for three years, and expressly allow oral withdrawal during the call. The real change is the burden of proof: it now rests on the caller, and it sits inside their own recordings. The French text also reaches calls placed from abroad to French numbers.

Thousands of small blue tiles, of which only about thirty are lit up in coral

Quality Monitoring 19 October 2026 14 min

Analysing 100% of calls: what the statistics say, and what it really costs

With 30 calls listened to a month, an 80% compliance rate in fact lies anywhere between 63% and 91%, and a ten-point drop cannot be told apart from chance. The calculation vendors do not show, the gap from which an action plan is justified, what analysing every call really changes, what it does not solve, and its cost per evaluated conversation.

Six rounded steps climb towards a large grid of tiles fully lit up in coral

Quality Monitoring 19 October 2026 12 min

Moving to 100% quality monitoring: the six-step method

Moving from sample-based listening to analysing every call changes the method, not just the tool. Deciding what you measure, rewriting the scorecard, calibrating the analysis against your evaluators, prioritising alerts, budgeting by the minute and redeploying the quality team: the six steps, with the calibration protocol and the mistakes to avoid.

A coral thread starts from a checklist of ticked boxes and draws a heart

Quality Monitoring 2 October 2026 17 min

Quality Monitoring: are you more "procedural" or "relational"?

A procedural scorecard checks that the advisor introduced themselves, rephrased and closed the call. A relational approach looks at what the call produced for the customer: a problem solved, a commitment kept, a reason to trust. What research says, our stance inspired by the Agile Manifesto, how to measure the relationship without measuring emotion, and a test to place your own set-up.