brands supported
sites across Europe
countries
Context
Armatis is a leading player in customer experience outsourcing in Europe. Its business covers the entire customer relationship value chain: omnichannel customer service (voice, email, chat, messaging), sales activities, back-office processing, digital relationship management and outsourced administrative management.
Against a backdrop of market transformation — generative AI, automation, margin pressure — Armatis is undertaking a strategic shift towards a technology-driven model.
Project challenges
The Armatis Technology programme aims to deliver a structural pivot:
- Build a differentiating SaaS software suite — SquAire — that can be sold independently of outsourcing services
- Set up a software-vendor organisation: product governance, industrialised delivery, scalability capacity
The ambition: move from a labour-intensive model to a scalable technology platform model.
An agentic architecture at the heart of SquAire
Steamulo was selected to design and implement the new SquAire software suite, built on artificial intelligence, in order to transform Armatis from a services player into a differentiating technology player.
The architecture relies on an orchestrated agentic approach, structuring the suite's intelligence around:
- Business digital products addressing specific use cases (customer relationship, customer knowledge management, steering, quality…)
- Specialised AI agents, autonomous and interconnected, enriching and automating each functional building block
These agents collaborate under the supervision of an orchestrator agent, ensuring correct understanding of context, activation of the right processing and overall process consistency. The first building blocks, co-built using Agile methodology, are SquAire Performance (analysis, supervision and steering of customer interactions) and SquAire Knowledge (agent copilot: suggestions, augmented responses, real-time guidance).
Specialised, shared AI agents
Each module relies on dedicated AI agents (understanding, generation, decision, control), shared across the platform:
- Diarisation — speaker identification and structuring of voice streams
- Transcription — transcription of conversations by speaker
- Anonymisation — anonymisation of conversations
- Summary — automated generation of interaction summaries and structured archiving
- Sentiment analysis — quality scoring, compliance, Voice of Customer, weak-signal detection
- Quality evaluation — overall interaction score based on multiple criteria
- Statistical data querying
- RAG agent — vectorisation of training catalogues and a question/answer chat via the recommendation agent
This approach automates customer relationship processes, augments agents' capabilities, industrialises operational performance and creates a differentiating, scalable technology foundation.
A sovereign architecture, deployed on Kubernetes
Given the high sensitivity of the data involved (voice, customer exchanges, personal data), SquAire relies on a mixed architecture that is sovereign and controlled:
- Self-hosting of AI and application building blocks on French cloud infrastructure
- Full control of data flows, with no imposed dependency on external APIs
- Strict GDPR compliance and sector-specific requirements
- Deployment in dedicated or isolated environments (multi-tenant or single-tenant)
The whole is orchestrated on Kubernetes: deployment of microservices and AI agents on clusters, horizontal scalability of processing (real-time AI), isolation of critical components, high availability and industrialised deployments (CI/CD, rolling updates). The AI models (LLM, STT, NLP) are deployed in a controlled environment, without exposing sensitive data, and specialised to Armatis' business contexts.
Our other projects.
Ready to scale up your projects?
Our experts are ready to audit your strategy and identify your first AI acceleration levers.