Conversations for enterprise applications

Making something new from something old (conversations)

ChatGPT demonstrated how natural language enables complex searches with easily usable results. Natural language is human language, the language we use every day.

In late 2022, OpenAI released ChatGPT, revealing generative AI capabilities to the general public. The novelty lies in the difference from previous search methods: users previously searched via keywords and received links to explore and synthesise. ChatGPT and similar LLMs provide complete, organised responses that users can iterate upon.

Historical context: Alan Turing conceptualised speaking machines in 1950; early chatbots like ELIZA appeared between 1966 and 1970; commercial applications emerged around 2000. But they suffered from poor contextual understanding, rigid interactions and limited learning capacity.

When users meet professional applications

Three major evolutionary stages:

First type: the Web (1990s-2000s)

Internet development enabled Cloud adoption and web interfaces for business applications, eliminating the need for dedicated client software.

Second type: mobile applications (from 2007)

Apple’s iPhone and App Store revolutionised user experience by offering mobile access. Success spread rapidly from personal to professional spheres, meeting needs for simplicity, speed and mobility.

Third type: conversational access

Conversational access has now spread well beyond smart speakers: according to OpenAI, ChatGPT had 900 million weekly active users in February 2026, up from 400 million a year earlier, more than doubling in twelve months. A Capgemini study revealed: “While 76% of enterprises benefit quantifiably from voice/chat assistants, companies must better address client needs for full potential exploitation.”

Conversational interfaces must overcome challenges specific to the professional sector: technical (security, privacy, IT integration), functional (application variety, usage versatility) and economic (spreading infrastructure costs over a user base that’s inherently smaller than in B2C).

Agora’s response to enterprise conversation challenges

A sovereign, trustworthy solution

Security and data confidentiality anchor the technical architecture. The objective: deliver a “sovereign, trustworthy solution compliant with regulations (GDPR)”.

Agora’s AI specialises in precise intent recognition while requiring minimal computing resources. This enables:

• Dedicated NLP functions per project, eliminating data and intent contamination between projects
• Administrator-defined training data automatically configuring neural networks
• Dedicated IT resources (Docker containers, AI/NLP, databases) per project
• Sensitive element anonymisation
• SSO function respecting client authentication policies

The platform deploys across three geographically distributed French hosting infrastructures with orchestration mechanisms for load-sharing and automatic failover handling.

Personalisation

The universality of human language enables conversational projects across all sectors: HR, ERP, Finance, Logistics, Industry, Municipalities, Home Care, etc.

A quality conversational experience requires understanding operational specifics: proposed actions, proper nouns, specialist vocabulary. In fleet management software, for example, a dispatcher might ask “where’s truck 12?” or “which vehicles are due for inspection this month?”, vocabulary (license plates, sites, inspection types) that means nothing to a generic language model never trained on these specifics.

Agora offers no-code and low-code studios enabling precise interaction definition. Defining intentions involves:

• Integrating specialist vocabulary as needed
• Building training datasets; entering example phrases automatically generates most training data via integrated LLMs
• Testing and deploying intentions; automated training produces dedicated language models within seconds

A genuine efficiency and profitability lever

Agora adapts to most enterprise architectures and applications for disruption-free deployment. The primary prerequisite: Internet-accessible applications with documented APIs.

Conversational interfaces integrate with existing solutions:

• Professional collaborative applications (MS Teams, Slack, Google Workspace)
• Bots integrated into web interfaces or mobile apps
• Consumer messaging apps (WhatsApp, Messenger, SMS)

After the Web, then mobile, conversational access is establishing itself as a third way to reach enterprise applications, driven by the same forces (simplicity, speed) but with constraints specific to B2B: security, business vocabulary, integration with existing systems. Security and sector-specific customisation aren’t optional extras: they’re what determines whether a conversational project makes it past the demo stage.