Whitepaper · October 2027 · 14 pages
The cost of a
shared word.
What happens inside a company when one term means three things, measured across 41 organisations over two years.
Contents
Summary
We tracked 41 companies between fifty and eight hundred people for two years, recording every occasion on which two teams reported a different value for the same named metric. The median company did this eleven times a year. The mean time to resolve a single disagreement was 5.4 working days, and in 38 percent of cases the resolution was that both figures were correct.
The cost is not the analyst hours. It is the decisions deferred while the argument runs, and the slow erosion of trust in every number that follows.
Method
Participating companies logged each disagreement in a shared form at the moment it surfaced, rather than reconstructing it later. We recorded the metric name, the two values, the teams involved, the elapsed time to resolution and the eventual cause. Companies were not told what we expected to find.
Self reported logs undercount. Three companies later found disagreements in their own archives that were never logged, which suggests our figures are a floor rather than an estimate.
Findings
Disagreements clustered around four words: revenue, active, churn and cost. Together they account for 71 percent of all logged cases. None of the four is ambiguous in isolation; each becomes ambiguous the moment two systems compute it on different boundaries.
Figure 1. Share of logged disagreements by metric name, 41 companies, 2026-2027.
Resolution time rose sharply with company size, but not linearly with headcount. The better predictor was the number of systems that could independently compute the metric, which grew faster than the company did.
Figure 2. Days to resolution against number of systems able to compute the metric.
What worked
Nine companies adopted a written definition held in version control during the study. In those nine, logged disagreements fell by a median of 74 percent within two quarters. The three that adopted a definition but kept it in a document rather than a reviewed file saw no measurable change.
The mechanism appears to be review rather than documentation. A definition that can change without anyone noticing behaves like no definition at all.
Limitations
Self selection is the obvious one: companies willing to log their own disagreements may already be unusually attentive. Our sample skews toward software, and the four clustering words may differ in other sectors. The two year window is too short to say whether the improvement holds.
References
Brightwater, N. and Oyelaran, M. (2027). Denominators. The Margin, No. 198.
Holloway, I. (2025). Review as a design constraint. Larkmoor Press.
Reed, D. (2024). Deleting infrastructure. Self published.
Sample whitepaper. The study, the companies and every figure are invented.
Single HTML file with no build step. Type: Newsreader, Inter Tight, served by Google Fonts under the SIL Open Font License.