Saying value should be verified is easy. Building the machinery that verifies it, on a real production line, with seasonality, mix changes and three other initiatives running at the same time, is the hard part. Here is how it is done honestly.
The first article argued that manufacturing software lives in a proof crisis: claims without baselines, definitions or attribution, discounted by every experienced buyer. This one is about the machinery of proof itself. Not the philosophy, the mechanics: what has to be measured, in what order, with what controls, for a plant's finance function to accept that a system created the value it claims.
The structure is a chain with five links, and like any chain, it is only as strong as the weakest one.
Before anything is improved, the current state is measured, by the same instrument that will measure the after state, under a definition that is written down and will not move. This ordering matters more than any other single decision. A baseline reconstructed after the fact, from shift logs and memory, is not a baseline; it is a negotiation.
Practical rules that make baselines honest: the window must be long enough to cover normal variation, typically four to eight weeks of representative production, not the two good weeks before the steering meeting. The definition must be the honest one from day one, which means the baseline will usually be lower than the plant's previously reported numbers, because honest measurement catches the stops that manual logs miss.
This is uncomfortable and must be said out loud at the start: your OEE is about to go down before it goes up, not because the plant got worse but because the number got true. Plants that skip this conversation spend the next quarter arguing about the instrument instead of improving the line.
An honest baseline usually delivers bad news first. That is how you know it is honest.
Every intervention, whether it is a changeover standard, a maintenance plan revision or a scheduling rule, gets an expected impact attached before it is implemented: which metric it should move, in which direction, by roughly how much, visible from when. This converts improvement from storytelling into forecasting, and forecasting is checkable.
The discipline sounds bureaucratic and is the opposite. It takes one line per intervention, and it changes behavior immediately, because teams stop proposing actions whose expected impact they cannot state. It also creates the raw material for the only question that matters later: did reality match the forecast?
Over time, the ledger of expected versus realised impact becomes a track record, and a track record is the difference between a system that recommends and a system that is trusted.
Attribution is where most improvement claims die, because plants are not laboratories. The honest answer to the confounding problem is the same one medicine found: a control. In a multi-line plant, roll the intervention out line by line, and the lines not yet covered become the comparison group.
If OEE rises three points on the covered lines and one point on the uncovered ones over the same period, the same season, the same order book, then two points is a defensible claim and three is not.
Line-by-line rollout is often treated as an implementation compromise, something to apologize for in the project plan. It is the opposite: it is the strongest attribution instrument a plant has, and it costs nothing. The same logic extends across plants in a group, where covered sites can be compared against not-yet-covered ones, with the usual caveats about comparing like with like: same archetype, same demand conditions, normalized per unit.
Normalization is the quiet workhorse here. Volumes change, mix changes, a heavy quarter flatters throughput and a light one damns it. Value must therefore be expressed per unit produced, or per operating hour, before any before-and-after comparison is made. A claim of X euros saved means little; X euros per thousand units at constant mix means something.
Points of OEE persuade engineers. Money persuades everyone else. The conversion is straightforward if the inputs are honest: recovered time multiplied by the line's demonstrated rate multiplied by contribution margin per unit, using the plant's own numbers, not industry averages. Where the recovered capacity has no demand behind it, the honest value is cost avoided rather than revenue gained, and the ledger should say so, because a CFO will notice the difference even if the slide does not.
The output of this link is a value ledger: a running account, in currency, of realised impact, each entry traceable back through the chain to the intervention, its expected impact, its control comparison and the raw events underneath. Not a dashboard. A ledger. Dashboards are for glancing; ledgers are for auditing, and the entire point of this machinery is to survive an audit.
The last link is organizational, not technical. The value ledger is reviewed periodically, quarterly is a natural rhythm, with the plant's finance function in the room, and entries are accepted, adjusted or rejected. This is the step vendors avoid, because it surrenders control of the story.
It is also the step that makes the story worth anything. A savings figure the CFO has signed becomes part of the plant's own accounting of reality; it will be quoted upward in budget discussions, defended internally, and remembered at renewal. A savings figure the vendor computed alone remains marketing, however accurate it may be.
There is precedent for all of this working commercially. The strongest-growing operational software companies of the last decade did some version of it: savings verified by customer finance functions, published benchmark data across their deployments, improvement targets committed inside a defined period. It is not a coincidence that those companies also commanded the industry's premium valuations. Proof compounds.
The chain, complete: frozen baseline, declared expectations, control comparison, currency conversion, finance sign-off. Break any link and you are back to decoration.
What remains is the interesting question: if outcomes can be proven, what happens to how software is bought, priced and renewed? That is the third article.
This is the machinery behind our own decision layer: every recommendation carries a declared expected impact and is checked against the measured result on the line. Want to see a value ledger built from your own production data? Book a demo.
Part of the Proof, Not Promises series by Fabrico. Fabrico is a manufacturing operations platform combining OEE monitoring with computer vision, a full CMMS, MES capabilities and production planning in one data model.
Anmeldung
Blog
