For most of their existence, institutional archives have simply been kept in storage. We consult them only when a dispute requires precedent or an anniversary calls for a photograph. Otherwise, they generally just collect dust – static, preserved and largely unread – because their volume and the time needed to interpret them exceed any one person’s capacity.
The UPU’s archive begins with the 1874 Treaty of Berne – the founding agreement that established the organization – and spans 150 years of Congress decisions, Conventions, resolutions, compendia and photographs: a transparent record of the institution’s reasoning on the evolution of the postal sector that it was created to serve.The challenge has now shifted. The accumulation that once made the archive hard to exploit is what makes it so valuable today, because it represents a rich network of human decisions and the global events that drove them. Powered by artificial intelligence (AI) agents that can process this complexity, the archive stops being a repository to dip into and becomes the foundation for an ontology – a structured model of knowledge that connects documents, concepts and decisions – enabling us to think with the archive, rather than simply searching it.
From static data to dynamic ontology
What has changed is the cost and time needed to read large bodies of text. An AI system that can move through tens of thousands of pages, follow citations across Congresses and locate the exact article at the basis of an issue can answer questions that the archive was never designed to address, such as tracing how an idea evolved across decades or identifying connections between decisions made at different Congresses.
This is what we explored in challenge 2 of the 2026 edition of the UPU Innovation Challenge, where participants applied frontier agentic AI to real-world postal challenges. Our team focused on transforming part of the UPU historical archives into a dynamic, interconnected knowledge base.
Our search focused on the archive’s latent intelligence – the knowledge already contained within its documents but hidden within decades of institutional records. We needed a way to connect decisions made years apart and uncover long-term patterns visible only when the archive is examined as whole. A mandate for cost-based terminal dues, for instance, appears clearly in 1974, 1979 and 1984, yet the technical architecture needed to implement it arrived only in 2012 – 38 years later. That long trajectory becomes visible only when the archive is read as a single knowledge base, rather than as a collection of scattered files.
To go further, we modelled five core axes – freedom of transit, terminal dues, customs, postal financial services and quality of service – and built a graph connecting their relationships. The results show how institutional knowledge accumulates. Freedom of transit, established in the Constitution in 1874, has endured and adapted to new threats, while postcode automation, adopted in 1969, became the infrastructure that later made service quality measurable. These patterns represent the institution’s own meta-knowledge, going beyond the content of any single document.
Human-machine collaboration
This is where the difference between automation and connection becomes practical. The agents did not interpret the archive on the institution’s behalf; they connected documents that no individual could hold in mind at once and let those connections surface to identify the patterns. We designed the system to help a person think, reason and infer, rather than to do it for them. This is the future that I believe in: humans learning to think alongside machines, each enhancing the other.
Human involvement remains non-negotiable. Every claim was traced back to a specific resolution, Congress and year, and clearly identified as either documented in the record or inferred from it. The challenges are as much about ethics and governance as technology. Archives contain their own silences and biases, and AI systems can infer relationships that are not supported by the historical record. The discipline that addresses these challenges is the same: cite every claim, distinguish documented evidence from informed inference, and maintain the human institution as the final author of meaning. For records carrying legal and historical weight, that transparency is what makes this approach viable.
Beyond the UPU
This principle extends far beyond the UPU. Every postal operator holds decades of regulatory filings, board decisions, correspondence and operational data in the same preserved state. The same approach – connecting, citing and interrogating institutional knowledge – enables any operator to consult its own memory before repeating decisions that its predecessors have already resolved.
If the UPU's institutional memory could speak across its 150 years, much of it would be a reminder: innovation is built on the patience of those who documented the past, preserving debates, decisions and lessons for future generations. Until recently, we lacked the technology to ask the archive what it could teach us. We have one now.
The effort only counts if we preserve the stories behind the outcomes, not the outcomes alone: the reasoning that guided us, the assumptions we tested and the mistakes we resolved along the way. Done well, future generations will inherit more than our conclusions; they will inherit our process. Data architecture becomes the container for that institutional memory and a lasting legacy for the global postal community as it navigates an uncertain yet profoundly exciting future.
Edgar Barroso, PhD
Founder of Horizons Architecture and Professor at the School of Government and Public Transformation, Tecnológico de Monterrey