Rockwell Automation’s July 2026 survey of 1,560 manufacturing decision-makers across 17 countries found that 93% of manufacturers already run MES software. Only 28% have it enterprise-wide, just 23% fully integrated across ERP, PLM, quality, and OT.
This is a vendor survey, not independent research – but its own numbers back the reading. 44% rank integration as the top MES buying requirement, 33% call it their biggest data integration problem.
“MES adoption is no longer the hurdle, but enterprise scale is,” said Anthony Murphy of Rockwell. Govindaraju and Putra reached the same ranking a decade earlier, with no stake in the answer.
Functional integration is “the most difficult challenge,” they found in a 2016 study of a Bandung steel plant. Adoption is solved. Integration is not.
Manufacturing Execution Systems Were Defined as Configurable in 1997
Low-code MES markets itself as new. It isn’t. MESA International defined the manufacturing execution system category in September 1997, in White Paper #6: “Users can configure this set of functionality to meet their corporate and plant objectives.”
The same paper is blunt about why: “the configuration and priorities may vary widely.” In 1997 vocabulary, configure meant choosing which functions a plant deploys, not modeling a workflow in a drag-and-drop builder.
Per-plant variability was never a defect low-code discovered. It was the specification MES shipped against from the start.
What ISA-95 Actually Says About Level 3, and What it Doesn’t
NIST GCR 19-022 (Georgia Tech’s Leon McGinnis) explains why MES gets configured rather than shipped finished. Below Level 3, “control consists of executing predefined operations,” a layer already “reasonably well-understood.”
At Level 3, scheduling is “an extremely difficult decision problem,” with no exact method that scales below exponential time. Shop floor automation below Level 3 is a solved control loop. Level 3 is a decision problem, discovered plant by plant, late.
Does ISA-95 actually say SCADA sits at Level 2? Its own committee page, fetched directly for this piece, never places SCADA or MES at any level. It describes the standard as the interface “between control functions and other enterprise functions,” specifically levels 3 and 4.
NIST says Level 2, a widely cited page says Level 3, practitioners argue Level 1 or 2, and ISA-95 itself answers none of them.
The Bottleneck Sits at the Machine, Not in the App Builder
Rockwell’s own gap, 93% adoption against 23% integration, points straight at machine data integration – measurable before evaluating a single MES feature.
On Time Edge’s Kim Burndred splits every asset into two categories: Category 1 machines have native connectivity such as OPC or MTConnect, Category 2 don’t and need an IIoT gateway. Count them, he says, “before there’s a rollout plan, and long before there’s even a timeline.”
TU Munich researchers needed 989 lines of code for data collection across three lab machines, communication layer alone. Dreher Consulting’s schedule delta tracks the same inventory: 4-6 months greenfield, 9-15 months brownfield, and even Dreher’s own low-code success story still ran 8 months and 1.5 technical FTE.
Low-code compresses the application half of a factory automation software project. It leaves the machine-connection half untouched.
Connectivity is Not the Same as Context
OPC UA was supposed to close this gap. Does turning on a protocol actually deliver meaning along with the signal? The OPC Foundation says its framework “turns data into information.” In practice, engineers report the model going unused.
Marcus Ilgner, after 18 months with OPC UA, found nodes “are just put into the namespace ns=1” (the generic bucket), because the full model carries overhead that “isn’t bound to yield any ROI in the short and mid-term.” Wautoma Biotech, an equipment OEM, calls the result “the difference between providing a cable and providing a map.”
That contextualizing work, not another protocol, is the job an industrial data management platform is actually asked to do. The claim isn’t false.
The model can carry semantics. What no protocol does is populate it for you, on any industrial automation platform, machine by machine, plant by plant. Production data management is mostly that mapping work, not the wire.
Where the Category’s Favorite Numbers Come From
Three outcome numbers repeated across MES vendor pages trace back cleanly, and none survives the trip. The “45% average reduction in manufacturing cycle time” traces to MESA’s White Paper #1, a small group “queried in 1993” – a paper whose own caveat vendors always drop, since results “will be different for each user.”
The same number resurfaces, unchanged, in MESA’s 1997 paper, and still promises real-time production monitoring gains on vendor pages in 2026. McKinsey published two ranges a year apart: 30-50% downtime reduction in 2017, unmethodologized, then 5-15% asset availability in 2018, alongside its own warning against treating predictive maintenance “as panacea.”
The “22% on-time delivery” and “19.4% net margin” figures attributed to MESA appear in neither paper anyone cites for them.
No One Has Measured This, Including the Company Publishing This Piece
No independent, controlled, multi-plant study establishes an average OEE uplift attributable to MES. Why not? Fraunhofer IVV explains it plainly – this material “usually hide(s) in sections like ‘lessons learned’… efforts put into the technical solution are not scientifically published.”
The one methodologized figure in this research, from a NIST brief built on 80 interviews, puts the entire addressable opportunity at “approximately 3.2% reduction in the shop floor cost of production” – an order of magnitude below what smart manufacturing platform vendors quote.
That absence of evidence applies to every vendor in this category, including the one publishing this piece. An article quoting a double-digit OEE figure is quoting a vendor, not a study.
Speed is Not Automatically an Advantage
Fraunhofer ISI’s 2019 regression on Germany’s manufacturing survey found digital technologies “did not have any statistically significant effects on total factor productivity.”
Combining robots with digital technology even caused “interference, resulting in a reduced impact of both technologies on productivity.” The data is from 2012 and covers digitalization broadly, not MES – both facts matter.
Hershey’s failed 1999 rollout shows what speed does to an unready process: three systems compressed onto a 30-month schedule instead of the recommended 48, testing shortened, go-live timed before the Halloween peak.
Food Industry Executive reports the same pattern in modern MES projects, with acceptance tests that “never simulate real changeovers, allergen clean-downs, or recall scenarios.”
A tool that lowers the cost of bending manufacturing workflow automation to a plant’s current habits, without fixing those habits first, only raises exposure to this failure mode.
A Rollout That Failed for People Reasons, Not Technical Ones
gbo datacomp describes a Bavarian firm’s MES rollout, carefully planned technically, that still produced poor data quality six months in. gbo’s own diagnosis was blunt – “the shift supervisors had never been involved in the conceptualization phase.”
The fix wasn’t code. Three supervisors joined as champions and helped design input forms, and within four months the data capture rate rose from under 60% to over 92%, with zero development. Even gbo, an MES vendor, admits “software is the smallest problem.”
Panorama Consulting, an independent auditor with no stake in either answer, lists the recurring causes as process design, master data, and supervisor buy-in. Change latency isn’t one of them.
Tulip, a low-code MES vendor with every reason to argue otherwise, points to what happens after go-live instead: a form change becomes “an IT ticket, a vendor engagement, and a release cycle… the Quality team stops asking and starts working around the system.” No independent source has measured that claim.
The Variable Worth Arguing About
Gartner’s July 2025 Magic Quadrant for Enterprise Low-Code Application Platforms evaluates twelve vendors, from Appian and Mendix to ServiceNow and Zoho. None builds an MES platform for factories. When an analyst says low-code, it means enterprise app delivery. When a plant says it, the word means something else.
Does that gap make the underlying gradient irrelevant? Not quite. A 2023 study in Empirical Software Engineering (Alamin, Uddin and Malakar) mined roughly 33,000 Stack Overflow posts across 38 low-code platforms and found Application Customization was the largest of 40 clusters, at 30%, because customization “that is not native to the LCSD platforms becomes difficult.”
Those are enterprise tools – the plant floor barely shows up on Stack Overflow. Applying that gradient to MES is this article’s proposal, not the paper’s finding: work inside what a platform ships natively compresses easily, work that leaves it does not.
A working integrator on PLCtalk made the same point in 2024: “If you only have basic MES requirements, such as OEE, these can be scripted directly in Ignition.
If it is full block track and trace, you may want to go on a SepaSoft training course.” OEE is arithmetic over tags, track-and-trace a model of material identity – that gap, not low-code versus code, is the variable actually worth arguing about.
One Architecture That Takes This Position
Iotellect’s MES module is one example built around that boundary. The vendor describes it as “ready-to-use MES modules on top of a fully editable low-code core,” covering planning, OEE, track and trace, and real-time monitoring – ISA-95 compliant, deployable cloud, on-premise, or hybrid/edge.
The pitch: shipped Iotellect low-code MES software as production management software for what plants already need, an open model underneath for what they don’t.
Everything above supports the shipped-modules half. The market really is split between enterprise low-code and industrial IoT MES, and named modules for named functions are a checkable claim.
The editable-core half is not the same kind of claim, and the evidence here leans against it. Alamin’s gradient says leaving what a platform ships natively is the hard case, not the easy one.
No documented, independently verified case turned up in this research – for this vendor or any other – of a low-code platform running full ISA-95 Level 3 execution at production scale, its configuration under version control and maintainable by plant staff after the integrator leaves.
That case may exist. It wasn’t found here, and the absence should be stated, not talked around.
The Data Layer Underneath it
None of that works without the layer described earlier, the one turning raw signals into modeled data. Iotellect’s broader platform, positioned as an Iotellect industrial IoT platform, connects “40,000+ types of devices and data sources” through 50+ protocols, on fixed-subscription pricing rather than per-device billing – a business fact, not a savings claim.
What that protocol count buys a plant, across connected factory systems, is a smaller version of the same problem: connectivity, not automatically context.
The Question a Marketing Page Cannot Pre-Answer
ISA-95’s own scope guarantees this problem never closes. KTH’s review lists “Management of configuration” as a Level 3 support activity – versioning the model isn’t a platform nicety, it’s a named part of what MES is supposed to do.
Mike Hadlow named the failure mode in 2012 – a team’s configuration language ended, years later, “back where we started… hard coding everything, except now in a much crappier language.” The real point came just before that line: “at a certain level of complexity, hard-coding a solution may be the least evil option.”
Versionable configuration is the minimum bar. It’s also trivially claimable: the answer is a bullet point, not an artifact. What no vendor page can pre-answer is harder – show the diff of a genealogy-model change that shipped last quarter. Who reviewed it. What broke. How was it caught.
Editability is where the clock starts.
It is not where it stops.
