Every vendor claims a ten percent OEE improvement. Every buyer discounts it. The result is a market where trust is priced at zero, good software is punished alongside bad, and plants keep buying on hope. There is a way out, but it is not better marketing.
Ask a plant director what their last software vendor promised, and you will get a number. Ask what it delivered, and you will get a pause. That pause is the most expensive silence in industrial technology, and both sides created it.
Vendors created it by making claims that cannot be checked. Ten percent OEE improvement. Thirty percent less downtime. Payback in months. The numbers are not necessarily false; some deployments genuinely achieve them. But they are presented without the three things that would make them checkable: the baseline they were measured against, the definition of the metric being improved, and the method that separates the software's contribution from everything else that changed at the same time.
A claim without those three things is not evidence. It is decoration, and experienced buyers treat it accordingly.
Buyers created it too, in a quieter way: by accepting proof standards from software that they would never accept from equipment. When a plant buys a new filling machine, the purchase contract specifies output, speed, and quality levels, and an acceptance test verifies them before final payment.
When the same plant buys performance software, it signs for licenses and hopes. The machine gets an FAT and a SAT. The software gets a kickoff meeting.
We hold a €200,000 machine to a stricter standard of proof than a €200,000 digital transformation. That is a choice, and it is the wrong one.
Three structural problems corrupt almost every improvement claim in this industry, including, if we are honest, some of the claims made by companies we respect.
The floating baseline. An OEE improvement is a difference between two numbers, and the first number is usually soft. If the before state was measured by manual logs that missed a third of the stops, and the after state is measured by an honest system that catches everything, the improvement is understated.
If the before period was chosen at a seasonal low, it is overstated. Few case studies disclose how, or when, the baseline was measured. Without that, the delta means nothing.
The definition problem. Two plants with identical physical performance can report OEE scores fifteen points apart, depending on how changeovers are treated, whether performance is measured against nameplate or demonstrated rate, and whether micro stops exist in the data at all.
A vendor who improves your OEE by redefining it has improved nothing. This happens more often than anyone admits, and usually without malice: the new system simply counts differently, and nobody reconciles the definitions.
The attribution problem. Plants are not laboratories. In the same quarter a software system goes live, the plant may also hire a new maintenance planner, lose a difficult SKU, change shift patterns and receive a rebuilt gearbox. When OEE rises four points, what caused it?
The honest answer is usually: several things, in unknown proportions. The vendor's case study will claim all four points. The plant's own improvement team, with some justification, will claim them too. Both cannot be fully right.
This is not an abstract complaint about marketing hygiene. Unprovable value has a price, and three parties pay it.
Plants pay it in misallocated improvement budgets. When you cannot verify which interventions worked, you keep funding the memorable ones rather than the effective ones, and the chronic losses, the boring ones, survive every budget cycle.
Good vendors pay it in discounted trust. When every claim in the market is decoration, buyers apply a uniform discount to all of them, and the vendor whose deployment genuinely would return its cost in six months is priced and doubted like the one whose deployment will quietly die after the champion leaves.
Markets without verifiable quality converge on price competition, which is exactly what has happened to large parts of the OEE software market.
And the industry pays it in slow adoption. Manufacturing digitization moves slower than its economics justify, and the standard explanations, conservatism, legacy systems, skills, are only half the story. The other half is rational skepticism: plants have been burned by unverifiable promises before, and unverifiable promises are still mostly what they are offered.
The alternative exists, and it is not complicated, just disciplined. A provable outcome in manufacturing software needs five properties, and the useful test of any vendor, ourselves included, is how many of the five they will commit to in writing.
An honest baseline: the before state measured by the same instrument and the same definition as the after state, over a stated window, before any intervention begins.
A disclosed definition: exactly how the metric is computed, published, so that any number can be recomputed by a skeptic.
Machine-measured evidence: data captured at the source, not recollected in review meetings, with disputed events resolvable by replay rather than by seniority.
An attribution method: some honest way to separate the intervention's effect from everything else, which is harder than it sounds and is the subject of the second article in this series.
Finance sign-off: the savings converted to currency using the plant's own margins and validated by the people who own the P&L, not the people who bought the software. A saving the CFO has not accepted is a slide, not a saving.
None of this is exotic. It is roughly the standard the industry already applies to energy-efficiency projects, where measurement and verification is a formal discipline with protocols and sign-offs, because money changes hands based on the result. Performance software has simply been allowed to skip it.
The vendors who adopt this standard voluntarily will take the market from the ones who have to be forced. The next two articles describe how the proof is built, and what it changes when it exists.
This is the same proof discipline behind our decision layer, where every recommendation logs its predicted impact and is checked against the measured result, and the practical questions buyers ask are answered in plain terms. Want a value ledger you could actually audit? 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.
Connexion
Blog
