The numbers that survive an IPO, and the ones that don't
Sitting inside a steelmaker's transition to a public company changes what you consider a metric. A short account of which numbers held up.
Gunung Raja Paksi is Indonesia’s largest private steelmaker, and while I was there it was becoming a public company. I owned the digital and data agenda through that transition: ERP across 25+ departments, and a consolidated reporting layer that gave the board one set of numbers instead of several.
That process is the most useful education in metrics I have had, because it applies a test most internal reporting never faces: someone outside the company, with no stake in your narrative, asks how the number was produced. A surprising proportion of numbers do not survive that question. Here is what separated the ones that did.
A number needs one owner, not one source
The standard prescription is a single source of truth. Necessary and insufficient. We had figures that came from exactly one system and still could not be defended, because nobody could say who decided what the number meant — which transactions were in scope, how returns were handled, when the period closed.
What survived scrutiny was a number with a named human owner who could answer definitional questions without consulting anyone. One source stops contradiction. One owner produces an answer.
If the definition changed, the history is fiction
The most painful discoveries were metrics whose definition had drifted. Someone had improved the calculation two years earlier — correctly — and nobody had restated the prior periods. The trend line therefore showed a change in the business that was actually a change in the arithmetic.
This is invisible internally, because everyone remembers the improvement and mentally adjusts. It is glaring to an outsider reading the series cold. Now I treat any definition change as requiring either a restatement or a visible break in the series. Silent improvement of a metric is a form of data loss.
Operational numbers get promoted, and they are not ready
A plant runs on figures produced for operational use: yield, downtime, inventory movement. They are accurate enough for the decisions they were built for, which often means accurate to a shift, reconciled informally, with local conventions about edge cases.
In an IPO those same figures get promoted into reporting where the tolerance is entirely different. The number does not change; the standard it is held to does. Anticipate which operational metrics will be promoted, and harden them before someone external asks — because the alternative is doing it under time pressure, which is when you find that two plants counted the same thing differently for a decade.
Timeliness is a control, not a convenience
Before, a monthly consolidation taking three weeks was an irritation. Under public reporting it is a risk: a long close means a long window in which the numbers are unknown, and errors found late are found expensively.
Shortening the close was not a reporting improvement, it was a control improvement. That reframing is what unlocked the investment — as an efficiency argument it had been declined for years.
What I took with me
Ask of any metric: who owns the definition, has it changed, what standard was it built for, and how long after the fact do we know it? Four questions, and they separate numbers that inform decisions from numbers that decorate slides.
It is also why, on any programme since, I get the ROI model agreed with Finance before the work starts rather than after. Not governance theatre — I have simply watched what happens to a number that nobody agreed to in advance when someone finally examines it.