Most transformation value is declared, not delivered. Verification is the difference.
The audit standard Markham applies before an outcome is allowed to be called delivered.
Most transformation value is declared, not delivered. Verification is the difference.
When we audited 312 outcomes that programmes had reported as delivered, 44% failed at least one of three tests: the baseline had moved, the benefit was double-counted, or the result did not survive twelve months. The industry’s delivery statistics are built on declarations. This paper publishes the standard we apply instead — the tests an outcome must pass before it may be called delivered in any Markham engagement or dataset.
The standard is deliberately simple: a frozen month-0 baseline, a single attribution line per benefit, an independent reader of the evidence, and a twelve-month survival check. It is published in full so that clients, and competitors, can apply it without us. Every figure in the annual cohort study passes it; that is what the numbers in our research mean when they are cited.
Share of outcomes reported as delivered that failed verification — moved baselines, double counting, or non-survival at month 12.
Tests in the core standard: frozen baseline, single attribution, twelve-month survival. Simplicity is the design.
Median overstatement of programme value in self-reported delivery figures, against verified figures for the same programmes.
Share of figures in the annual cohort study that pass the standard. Declarations do not enter the dataset.
Freeze the baseline before the programme starts and put it under change control. A moved baseline is the most common way value is invented.
Give every claimed benefit exactly one attribution line. If two initiatives claim the same saving, one of them is wrong.
Have the evidence read by someone who does not benefit from the answer.
Re-test at month 12. An outcome that did not survive a year was a project artefact, not a delivered result.
Markham Institute, Measuring transformation: a verification standard, MKM-R-2025-006, v1.0 (September 2025). Citation permitted with attribution.