A National Vote Highlighted by Complexity
In late spring 2026, some 55,000 unionized Canada Post employees completed a crucial voting process on their new five-year collective agreement, as reported by the newspaper Les Affaires. This vote, which spanned several weeks, represents a pivotal moment for labour relations within one of the country's largest Crown corporations. Beyond wage issues and working conditions, this event highlights a recurring structural challenge in the professional world: the extreme complexity and density of the legal texts that govern workers' lives.
A modern collective agreement often resembles a document of several hundred pages, written in dense legal and administrative jargon. For a front-line employee, whether a letter carrier, mail sorter, or clerk, finding a quick and reliable answer to a specific question can feel like an uphill battle. Whether it is calculating overtime in a specific context, understanding parental leave terms, or verifying safety rules, the barrier of regulatory language limits the practical exercise of individual rights.
The Legal Literacy Gap in the Workplace
This lack of accessibility is not an isolated case. According to analyses regularly published by the Quebec organization Éducaloi, simplifying the law is a major issue of social justice and equity. In the workplace, the inability to correctly interpret a collective agreement clause can generate unnecessary friction between employees, union representatives, and human resources managers. Labour relations departments often find themselves overwhelmed by repetitive information requests, while some employees simply give up on exercising their rights out of sheer discouragement when faced with the complexity of the texts.
This is where natural language processing technologies, when rigorously managed, open up new possibilities. Rather than forcing humans to become experts in legal interpretation, technology now makes it possible to translate this complexity into a fluid, accessible, and verifiable dialogue.
To achieve this without the risk of error, two key technical concepts must be deployed: word embeddings and retrieval-augmented generation (commonly known as RAG).
Understanding the Technology: From Vectors to Semantic Search
For a machine to help an employee navigate their collective agreement, it must do more than simply search for keywords like a traditional search engine from the 2000s. It must understand the meaning. This is the role of embeddings. This technique involves converting text segments into mathematical vectors in a multidimensional space. In this space, two sentences with similar meanings (for example, "leave for family obligations" and "absence for a sick child") will be geographically close, even if they share no words in common. This is known as semantic search.
Once the semantic search is complete, the RAG method steps in to formulate the response. Unlike consumer artificial intelligence models that try to guess answers based on general knowledge accumulated during training (which often causes hallucinations, meaning the invention of non-existent facts or clauses), RAG imposes a strict framework.
When a user asks a question, the system first queries the local document database (the indexed collective agreement). It extracts the exact paragraphs addressing the topic, then provides these paragraphs as the sole source of context to a language model. The model's only task is to rephrase the information clearly and concisely, explicitly citing the article and page of the original document. The risk of fabrication is thus neutralized.
The ProductivIA Approach: Document Library in Service of Transparency
Within the ProductivIA application ecosystem, this technological synergy comes to life through the Document Library application and the Central Assistant. Organizations, whether small businesses, municipalities, or public institutions, can upload all their collective agreements, internal policies, and employee handbooks into the Document Library application. These files (PDF, Word, or Markdown) are immediately vectorized and securely stored within the organization's logical silo.
Employees can then interact directly with the Assistant using natural language. A question such as "What is the notice period required to plan my summer vacation based on my seniority?" triggers an orchestrated query in the background. The Assistant calls the Document Library services, retrieves the precise clauses from the collective agreement, and formulates a neutral, objective, and referenced response. Users obtain their answer in clear English, along with a direct link to the official article for validation.
This approach offers a double advantage. For employees, it guarantees direct, autonomous, and confidential access to their rights, without fear of judgment or bias. For the organization, it eases the burden on human resources teams and union stewards, allowing them to focus on resolving complex cases rather than basic information requests.
Sovereignty and the Protection of Sensitive Data
Using artificial intelligence to handle questions about working conditions raises important privacy concerns. Employee queries can reveal sensitive personal situations: questions about sick leave, disability accommodations, or grievance procedures. Routing these queries through servers located abroad would pose a major risk of non-compliance with Quebec's Law 25 on personal information protection.
This is why the ProductivIA platform allows all orchestration workflows to be configured to run exclusively on sovereign infrastructure. By pairing the Document Library with the Matania model provider, hosted locally in Quebec, employee queries and collective agreement texts never leave the country. Organizations can thus ensure they offer a modern transparency tool while respecting the strictest security standards.
Toward New Collaborative Dynamics
Integrating such popularization tools does not aim to replace social dialogue, but to make it healthier. By eliminating misunderstandings linked to the misinterpretation of regulatory texts, local artificial intelligence helps ground discussions on shared factual bases. Unions and management can thus collaborate within a transparent information environment, where the written rule finally becomes accessible to everyone, regardless of legal expertise.