Knowledge
What dormant knowledge costs
Studies redone, clauses forgotten, know-how gone with a colleague: what the knowledge sleeping in documents costs, and which part can be measured.
Every organisation holds a considerable body of knowledge, built at great cost, which it mostly uses only once. Reports, studies, opinions, notes, contracts, minutes: the document is written, delivered, archived, and what it contains then stays within reach without anyone going to look for it. Everyone knows the situations that follow: the study redone because nobody knew it existed, the answer only one colleague knew, the deadline discovered the day after. They are so common that we tend to file them under the ordinary friction of organisational life. They nonetheless have a cost.
Knowledge that sleeps
In an organisation, knowledge sleeps in thousands of reports, opinions, notes, minutes, contracts and datasets. Each cost days of work. Very few are reread after delivery.
Leaders know it. In February 2025, Gartner reported, based on a 2024 survey of data management leaders, that 63% of organisations either lack, or are unsure whether they have, the right data management practices for AI. The firm drew a prediction from it: through 2026, organisations would abandon 60% of AI projects not supported by AI-ready data.
The prediction remains to be confirmed. It nonetheless frames the stakes well: usable knowledge is becoming a condition for AI projects to succeed.
Two scenes, and an institution
The most telling reality needs no statistics.
You may have lived this scene. You are asked for a study, due in six weeks. You spend the month on it: interviews, public data, a synthesis of existing work. The day before delivery, someone asks whether you drew on the study carried out three years earlier on the same subject. You had never heard of it. It was sleeping in the document management system, under a title containing none of the words you searched for, and it already answered half your questions.
The other scene is slower. A colleague retires after more than twenty years in the organisation. She knew which contract had been renegotiated and why, which clause had caused trouble with which supplier, which opinion carried authority. All of it was written down somewhere. The path to finding it ran through her, and it left on the day of her farewell drinks.
The same scene plays out at the scale of institutions. An advisory assembly keeps thirty years of opinions in its archives. Each took weeks of hearings, debate and drafting. In theory, everything can be consulted. In practice, these opinions are reread by those who remember writing them, and fade from collective memory when those people leave.
In all three cases, the information had not disappeared. It was written down, archived, open to consultation. What was missing was the path to it.
What can be measured: contracts
The easiest cost to measure sits in contracts, where every forgotten clause translates directly into money. An indexation clause nobody applies on the due date, a late-delivery penalty never claimed from the supplier, a volume discount forgotten when paying the invoice.
Each time, the information existed, signed and filed in the document management system, and nobody was there to reread it at the right moment.
Added together, these small losses weigh heavily. The professional association World Commerce & Contracting estimates average value erosion at 8.6% against the expected outcomes of contracts, citing several causes. It also notes that contract data is spread on average across 24 different systems: the knowledge exists, and it stays hard to find when it matters.
These are estimates from a professional association. The fragmentation they describe, anyone can observe in their own organisation.
What is harder to measure
The other costs are harder to quantify and weigh just as much: work redone, inconsistencies between two reports that contradict each other without anyone noticing, slowness against a competitor who finds in an hour what takes you a week.
Generative AI changes the game. For the first time, an organisation can question everything it knows, connect documents written ten years apart, and be alerted when a forgotten text becomes useful again. The promise is real, and it comes with a requirement: the value of this intelligence depends entirely on the quality of the sources it relies on.