The recent outcome of the collective agreement vote at Canada Post, widely reported by national media such as the daily newspaper Le Droit, highlights the inherent complexity of large-scale labour negotiations. With tens of thousands of unionized members called to vote on multi-year agreements, every clause, phrasing, and historical precedent carries significant weight. In this type of process, there is no room for error: a misinterpretation of a historical text or the omission of a prior agreement can disrupt a fragile balance and lead to major labour disputes.
Behind the headlines lies a colossal information management challenge. A modern collective agreement is not a simple linear document; it is a complex legal ecosystem made up of letters of understanding, arbitration decisions, administrative regulations, and pension plans accumulated over decades. For labour relations teams and union representatives, analyzing this mass of information under tight deadlines is often a herculean task.
The Challenge of Document Overload in Labour Relations
Traditionally, searching for precedents within an organization's archives relies on keyword search engines. This method quickly shows its limits when dealing with legal concepts or human relations. If a negotiator searches for clauses related to scheduling flexibility, a classic keyword search might overlook crucial paragraphs dealing with working time arrangements or work-life balance simply because the exact terms differ.
This is where artificial intelligence technologies prove useful, particularly Retrieval-Augmented Generation (RAG) and embeddings. These scientific concepts make it possible to transform raw text into mathematical vector representations. By calculating the semantic proximity between sentences, the system understands the intent and context behind a query, rather than limiting itself to a simple character comparison. As a result, asking a complex question about retroactive pay terms allows for the instant extraction of relevant sections across dozens of historical documents, even if the wording varies over time.
Data Sovereignty at the Heart of Sensitive Negotiations
In the context of collective agreement negotiations, the confidentiality of working documents is absolute. Draft agreements, financial impact analyses, and bargaining strategies constitute highly strategic data. According to the requirements of Quebec's Law 25 on the protection of personal information, public and private organizations must ensure that sensitive data is not exposed to third parties or transferred outside the province without a rigorous privacy impact assessment.
The use of consumer artificial intelligence services poses a major risk of information leaks, as the queries sent can be used to train third-party models or be stored on foreign servers. To guarantee the security of these critical processes, organizations must turn to siloed architectures where information processing is carried out locally or on proven sovereign infrastructures.
Perspective within ProductivIA
The ProductivIA platform meets this precise need for security and efficiency through its Document Base application. Designed to function as a secure organizational memory, this application allows organizations to ingest, vectorize, and structure all collective agreements, meeting minutes, and internal jurisprudence.
By pairing the Document Base with the central ProductivIA Assistant, human resources professionals and legal advisors can query their document corpus in natural language. For example, it becomes possible to ask the Assistant to compare the structure of proposed salary increases with the agreements reached during the three previous bargaining rounds. The Assistant then relies on Document Base services to extract the exact data, reducing the risk of hallucination by anchoring its answers solely in the official documents provided by the organization.
To meet the compliance requirements of Law 25, the organization's silo administrator can configure the Assistant to exclusively use the Quebec sovereign model, Matania. This choice ensures that all linguistic processing takes place within Quebec, without any strategic data or personal information transiting to foreign servers. The agnostic orchestration of ProductivIA allows for this seamless model transition without requiring software modifications or programming skills.
Looking Ahead
The integration of semantic search and RAG in the field of labour relations does not aim to replace human dialogue, which remains the foundation of any successful negotiation. Instead, it aims to eliminate administrative friction and provide stakeholders with reliable, verifiable, and secure data. By reducing the time spent on document research, negotiators can focus on what matters most: finding fair compromises and building harmonious labour relations.