A Historic Agreement Marked by Textual Complexity
The recent tentative agreement on a new three-year collective agreement between Ford Canada and the Unifor union, representing Canadian auto workers, marks a major milestone for the industrial sector. Reported by business media, including Les Affaires and La Presse, this agreement suspends the threat of a labour dispute and stabilizes operations in a key sector of the economy. However, behind the signing of such agreements lies an immensely complex administrative reality: the drafting, management, and application of legal texts that frequently span several hundred pages.
These official documents do not simply set wage scales. They detail with surgical precision pension plans, group insurance, grievance procedures, seniority rules, and occupational health and safety protocols. For human resources managers, shop-floor supervisors, and union stewards, navigating this maze of interconnected clauses is a daily challenge where an error in interpretation can quickly lead to costly litigation or a tense working environment.
The Challenges of Interpreting Labour Relations
In the field of industrial relations, precision is a legal obligation. Misapplying a seniority clause during a recall to work or misinterpreting overtime hours can trigger a formal grievance process. Traditionally, resolving these issues requires lengthy manual searches through physical binders or static PDF files. According to a Université Laval study on bargaining dynamics, the time lost searching for precedents and interpreting collective agreement texts is one of the main administrative pain points in medium and large organizations.
The arrival of large language models (LLMs) has sparked strong interest in automating these searches. However, using public models poses two major risks. The first is hallucination: a generalist artificial intelligence model, if not strictly constrained, can invent a clause or mix up provisions from two different agreements. The second, even more critical risk concerns confidentiality. Labour relations documents, draft agreements, and bargaining notes contain highly strategic data. Sending them to third-party servers located abroad to feed commercial models exposes the organization to information leaks and potential violations of privacy laws, such as Law 25 in Quebec.
RAG: Anchoring Artificial Intelligence in Real Text
To overcome these limitations, computer science research has developed an architecture called Retrieval-Augmented Generation, commonly known as RAG. Unlike a traditional chatbot that relies solely on the general knowledge acquired during its training, RAG forces the artificial intelligence to first read a specific document before formulating a response.
The process relies on the use of embeddings, meaning the conversion of text paragraphs into mathematical vectors that capture their semantic meaning. When a user asks a question such as "What is the deadline to dispute a disciplinary notice under the new agreement?", the system does not perform a simple keyword search. It identifies the passages in the collective agreement whose meaning is closest to the question, extracts these precise excerpts, and then presents them to the language model. The model then drafts a clear response, rigorously anchored in the official text, citing the exact article. This method drastically reduces the risk of hallucination and guarantees the verifiability of every claim.
The Sovereign and No-Code Approach of ProductivIA
It is precisely this scientific rigour that the ProductivIA platform integrates through its Document Library application. Designed to meet the requirements of corporate and institutional environments, this application allows highly dense organizational documents, such as the new Ford Canada collective agreement or employee handbooks, to be imported and instantly transformed into searchable knowledge bases.
One of the pillars of ProductivIA lies in its entirely no-code architecture. Unlike unguided rapid development approaches, sometimes referred to as "vibe coding," which generate code on the fly without audits and introduce security vulnerabilities, ProductivIA offers a standardized and secure environment. The user does not have to write a single line of code or configure any complex data pipelines. Documents are uploaded into an isolated logical space, called a silo, ensuring that organizational data remains confined and is never used to train external models.
The platform's orchestration allows the Assistant application to seamlessly query the Document Library. A manager can thus ask the Assistant to draft a response to an employee based exclusively on the clauses of the collective agreement stored in the Document Library. For organizations subject to strict data sovereignty requirements, the administrator can configure the platform so that queries are processed by the sovereign Quebec provider Matania, thereby avoiding any cross-border transit of sensitive information.
Toward Assisted and Secure Joint Collaboration
Integrating sovereign RAG technologies into labour relations management paves the way for a new era of collaboration. By providing a single source of truth that is instantly and reliably accessible, employer and union representatives can agree more quickly on text interpretation, thereby reducing sources of administrative friction.
Technology does not replace the human expertise of industrial relations advisors or the legitimacy of union representatives. Instead, it frees them from tedious document search tasks, allowing them to focus on what matters most: social dialogue, solving complex problems, and maintaining a healthy and equitable working environment.