"Our documentation is fine" is the sentence every EPM Center of Excellence says until the one person who runs the forecast is on leave. The practice looked mature because the parts that were visible were good. A maturity score makes the invisible parts count, and it only works if the score is set by the weakest part, not the average.
Six drivers, measured from what is actually there
A score should be computed from the content, not from a survey. For each application, six drivers, each measured as a percentage of what the standard says should exist:
| Driver | What it measures | How it is counted |
|---|---|---|
| Documentation completeness | The design document, SOPs, specifications and templates the standard requires for this application | Required documents present, as a share of the list |
| Review discipline | Whether content is still true | Documents reviewed within their review period, as a share of all documents with a review date |
| Training completeness | Whether a new joiner can be onboarded without a person | Learning paths with every module filled, and completions by the people in the role |
| Ownership and backup cover | Whether the practice depends on who is in the room | Processes with a named owner and a named backup, as a share of processes |
| Run-book depth | Whether a process can be handed over | Run books with steps, calendar, controls and a handover pack, as a share of run books |
| FAQ and issues | Whether known problems are written down | Known issues with a workaround, and FAQ entries updated in the period |
Each driver maps to a level from 1 to 5: ad hoc, repeatable, defined, measured, optimized. Recalculate nightly, because content ages: a run book reviewed in January is a different asset in September.
Why the weakest driver, not the average
Take an application with these driver levels: documentation 4, review discipline 4, training 4, ownership and backup 1, run-book depth 4, FAQ 4. The average is 3.5, which reads as a solid practice. The ceiling is 1, which reads as what it is: a practice that stops when one person leaves, because no process has a named backup. The average rewards volume. A team can raise it by writing more documents, which is what teams do when a score is an average. The ceiling cannot be gamed that way; the only way to raise it is to fix the driver it names.
That is the second reason to use the weakest driver: it tells you what to do next. An average says "improve"; a ceiling says "name a backup for the forecast process". The next improvement is never a guess, and the person who owns the application can see it without a meeting.
What the levels mean in practice
Level 1, ad hoc: the process runs because someone knows how. Level 2, repeatable: it is written down somewhere, by someone, once. Level 3, defined: the standard's documents exist, with owners, and the twentieth application looks like the first. Level 4, measured: review dates are kept, training completions are tracked, and warnings raise themselves when content goes stale. Level 5, optimized: the value of the application is measured against a dated baseline and the roadmap follows it. Most practices sit at 2 and believe they are at 3; the difference is usually review discipline and backup cover, the two drivers nobody measures by hand.
Warnings that raise themselves
A score that is recalculated nightly can also tell you what changed. Content past its review date, a run book never reviewed, a process with no backup, a learning path with an empty module: each is a warning on the application's page, with an owner. The warnings are what keep the score honest between quarterly reviews, and they are what the CoE lead looks at on Monday instead of auditing a drive.
What to do this quarter
- Pick one application and write the list of documents, run books and learning paths the standard requires for it.
- Give every item an owner and a review date, and every process a named backup.
- Score the six drivers by hand once, take the lowest as the level, and fix that driver first.
- Repeat for the next application using the same list, so the practice is comparable across applications and regions.
The EPM CoE Portal computes the score nightly from the content in it, names the weakest driver, and raises the warnings. About the EPM CoE Portal.