Flawed or confusing data is fed into central reporting systems until executives stop trusting them, and the person keeping a hand-built alternative becomes the source everyone asks for the real number.
An example
Job costs across forty sites were supposed to become visible at a glance when a commercial builder rolled out a central project-controls system, so the chief financial officer would no longer have to phone each regional vice president. Within two quarters the system had a reputation. In the southeast region, cost codes were applied inconsistently, change orders appeared under the wrong jobs, and committed costs were entered late, so margins swung from week to week. At one board meeting the chief financial officer presented a margin figure from the system that was off by three points, and the southeast vice president corrected it from his own spreadsheet. Afterward, executives began calling him for the real number before every review. His team blamed confusing new fields and a flaky integration. When finance finally traced the entries, a third of the miscoded records had come from two of his project accountants.
Why it works
Polluted data destroys the central system's main promise, that it can be trusted, and one visible error in front of the board poisons confidence in fifty correct numbers, so the system looks worse. Relying on it becomes costly: the risk of embarrassment, the reconciliation work, and blame for a number nobody can vouch for. The polluter wins by adding noise, a cheap operation against an expensive defense, and becomes the chokepoint everyone calls. It works when contamination is hard to trace to a person, a plausible cover exists (a flaky integration, say), data governance is weak, and executives have low data literacy and a great deal to lose from being wrong.
The thinking behind it
George A. Akerlof In The Market for "Lemons": Quality Uncertainty and the Market Mechanism (1970), Akerlof shows that when buyers cannot tell good from bad, they treat everything as average and the good is driven out. Executives who cannot tell which system numbers are sound treat all of them as suspect, and the hand-built spreadsheet wins by default. Data lineage that attributes every entry to a person lets users tell good data from bad again.
Daniel Kahneman Judgment under Uncertainty: Heuristics and Biases (1974), by Tversky and Kahneman, describes availability: people judge likelihood by how easily instances come to mind. A wrong number presented to the board is a vivid instance, and it dominates executives' sense of how reliable the system is. Showing corrections in the same rooms supplies new instances, which is how the system's credibility gets rebuilt.
Warning signs
Executives source board numbers from one team's hand-built spreadsheet instead of the system. Errors in the central system cluster in one unit or one kind of entry. The explanations are always technical: confusing new fields, a flaky integration, a mapping change. One person is consistently the answer to "who has the real number?" That person's unit has write access to the system and keeps its own parallel records.
How to defend yourself
A general argument for the system's reliability does little. Fix lineage and attribution.
- Trace errors to their source: who entered what, and when. Most systems record it, and most organizations never look.
- Fund data governance with the authority to audit entries and correct them at the source.
- Refuse manual overrides in high-stakes forums unless they come with a written reconciliation to the system of record.
- Show the system's figures with the corrections marked, so its credibility is rebuilt in the rooms where it was lost.
Hold off on going after the curator personally until the pattern is documented.
How a leader shuts it down
Invest in data lineage so every change is attributed to a person. Fund data governance with audit authority, and treat data hygiene as a management accountability measured by unit. When the system exists, stop asking a curator for the real number and ask for the reconciliation instead. Watch who gains standing whenever a shared system loses it.