Most plants already have the four systems. SCADA is on the wall, the historian has years of tags, MES books the order, and ERP closes the week. The data is related because it describes the same assets and the same stretch of production. That still does not mean the systems talk. In a lot of sites they share zero communication that a supervisor can use without a person in the middle.
This page is a diagnosis, not an integration architecture. The split keeps arriving as three causes at once: technical (protocols and stores), organizational (IT and OT), and historical (bought in different decades). Each cause has a repair, and each repair has a cost band. Those bands are not prices. They are a way to stop treating every gap as a platform project.
Manufacturing control functions and business functions do not share one language by default. ISA developed ISA-95 as an abstract model for information exchange between those functions, with most of the standard aimed at the interface between manufacturing operations at level 3 and business planning at level 4. Yan Lu, KC Morris, and Simon Frechette put the same problem in research language: the ability of disparate systems to exchange, understand, and exploit product, production, and business data rests critically on information standards. NIST standards landscape
If you arrived from when the plant does not answer, stay with the systems this time. That article is the feeling of asking one question and receiving fragments. This one is why those fragments were never going to arrive as a single record.
Four systems, four clocks, one plant
SCADA, the historian, MES, and ERP can sit in the same building and still keep four clocks. The live graphic answers what the state is now. The archive answers what a tag did over a window. MES answers which order, batch, or routing step is in work. ERP answers what the plant promised, issued, received, or costed in a planning period. Each answer can be true. They still do not form a conversation unless someone maps object, time, and authority by hand.
ISA’s public overview of ISA-95 is useful because it refuses to treat the plant as one application. Level 0 is the physical process. Level 1 is sensing and actuation. Level 2 is monitoring and supervisory control, including PLCs and distributed control systems. Level 3 is manufacturing operations management, where ISA places MES and also names SCADA as a system that can manage operations. Level 4 is business planning and logistics, including ERP. The standard primarily deals with the interface between levels 3 and 4. That last sentence is easy to skip, and it is the whole problem. The plant bought systems for different layers, then asked them to behave as if the interface already existed.
The historian is the system people most often mistake for that interface. Wayne Matthews, writing in ISA InTech, says process historians were once simply software packages that stored and presented real-time process data. Older historians were fed mainly by a DCS and relied on a proprietary database that limited access. Modern historians can receive data from control systems, laboratories, ERP, and asset systems, and plants use them for diagnostics, maintenance, alarms, and production. ISA on process historians Breadth is not a shared clock. A historian can store a temperature every second, or only when the value leaves a deadband. MES may close a batch at a shift boundary. ERP may post the same event to a fiscal period. If those three times are not named, the join looks clean and is still wrong.
NIST’s manufacturing standards report treats this as a structural fact rather than a local failure. The ability of disparate systems to exchange, understand, and exploit product, production, and business data still rests on information standards, not on proximity in the same plant. NIST standards landscape The four systems already hold related facts. They do not hold the same objects. SCADA knows a tag. The historian knows a compressed series. MES knows an order and a work center. ERP knows an item and a posting. A person can see that they belong together. The systems cannot, unless that relationship is stored somewhere other than someone’s memory.
NIST’s OT security guidance adds a harder constraint: OT and IT were not built as one timing regime. OT systems are generally time-critical, while many IT systems optimize for throughput and can tolerate delay. Many OT processes are continuous, and unexpected outages often have to be planned days or weeks ahead. NIST SP 800-82 A live control system will not wait for an ERP posting. An ERP close will not wait for a historian exception record. Four clocks is how the plant was specified.
The practical test is ordinary. Pick one asset and one hour. Can SCADA, the historian, MES, and ERP name that asset the same way, close the same time window, and say which record is allowed to decide the next action? If the answer is no, the plant does not have a communication problem in the vendor sense. It has four truthful records and no shared sentence.
The technical split: protocols and stores
The first reason the systems do not talk is unromantic. They do not speak the same protocol, and they do not keep the same kind of store. A field device may sit on Modbus, PROFIBUS, Foundation Fieldbus, or a vendor protocol that never left the cabinet. A PLC may reach an HMI over Ethernet/IP, PROFINET, or EtherCAT. A historian may accept OPC or SQL. MES and ERP usually live in relational or transactional stores with their own APIs. NIST’s landscape maps this stack without pretending it is already unified. Shop-floor communication is a mix of fieldbuses and industrial Ethernet. Links from the shop floor to MES and enterprise systems depend on later integration standards such as OPC, OPC UA, MTConnect, PackML, and BatchML. NIST standards landscape
Matthews describes the older historian end of that stack with useful bluntness. Older historians relied on a proprietary database, which often limited access. Later, interfaces such as OPC, Ethernet, and SQL let more systems send and access historian data. ISA on process historians OPC can move a value. It does not decide whether the value is a tag, a batch attribute, or an inventory quantity. SQL can expose a table. It does not tell MES that two aliases are the same pump.
CISA’s ICS guidance explains why those old protocols persist. Many ICS environments still run legacy technologies and proprietary protocols because they were designed for operability and reliability, not cybersecurity, and they often sat on isolated networks. Many legacy devices still use older operating systems and protocols that lack encryption or authentication. CISA ICS Opening that stack to MES or ERP is not a cable job. It is a change to a system that was specified to stay quiet.
The stores make the split worse. SCADA keeps a live image. The historian keeps a compressed series, sometimes by exception. MES keeps orders, batches, and status. ERP keeps inventory, confirmations, and cost. NIST notes that even when manufacturers apply data analytics, two barriers remain: selecting the right tools, and integrating those tools with data acquisition and decision support. Many analytics tools still use proprietary interfaces and specialized formats. NIST data analytics A SCADA-to-ERP pipe will not invent context on the way through.
Names fail next. The same asset can be a tag path, a historian point, a work-center code, and an ERP equipment record. Units drift. Time fails after that. Matthews notes that historians use filtering, time stamping, and reporting by exception when a value stays inside a deadband. ISA on process historians MES may use a shift calendar. ERP may use a posting date. A join that ignores those rules looks decisive and still starts an argument about which number is real.
The Low-cost repair for this cause is a point-to-point connector: one known source, one known destination, one documented mapping, one owner. That is the right first move when the plant needs a recurring transfer and already knows the objects. It is the wrong move when the plant is trying to make four clocks agree by hiding the mapping in a nightly job. The Med-cost repair, a context, name, and time layer, belongs later. First admit that the protocol and the store are doing the job they were bought to do. They were not bought to be a plant language.
The organizational split: IT and OT
Even a clean protocol fails when the people who own the systems do not share a job. NIST’s OT security authors write this without romance. Control engineers will play a large role in securing OT, but they will not be able to do so without collaboration and support from both the IT department and management. The cultures of control engineering and IT are often significantly different, and their integration is essential for a collaborative security design and operation. NIST SP 800-82 The same sentence applies to data. OT owns the meaning of the process. IT owns the store, the identity path, and the backup. A connector that neither side wants to maintain is not an interface. It is a future outage.
The priorities differ for reasons that are not political theater. OT security objectives typically prioritize integrity and availability, then confidentiality, with safety as an overarching priority. Unexpected outages of systems that control industrial processes are unacceptable. NIST SP 800-82 IT may treat data loss or office-system availability as the urgent problem. An OT engineer will refuse a query that loads a controller. An IT owner will refuse a file share that bypasses authentication. The plant then lives with a USB export and a spreadsheet, which is how the Excel trap in production becomes the unofficial integration bus.
ISA wrote ISA-95 in part to give those two groups a vocabulary. The society says the framework helps manufacturing personnel and IT personnel collaborate by determining key terms for integration projects, and that it reduces the risk, cost, and errors associated with implementing them. Plants skip that step. They start with a vendor demo instead of agreeing what an order, a batch, or a confirmation means. Without those terms, IT builds a pipe and OT refuses to trust it.
Security architecture makes the organizational split visible in hardware. NIST recommends logical separation between corporate and OT networks, including firewalls, unidirectional gateways, and a DMZ that prevents direct traffic between those networks, plus separate credentials for corporate and OT users. NIST SP 800-82 CISA describes the older pattern: ICS environments that functioned in isolated networks with limited external connectivity. CISA ICS Isolation was a safety habit. Communication has to be designed through that habit, not around it. Opening the firewall is not integration. It moves a control system onto an office threat model.
Ownership is the part that software cannot fake. Someone has to own the mapping when a tag name changes or ERP posts a confirmation late. NIST says a cross-functional cybersecurity team should include IT staff, a control engineer, an operator, a security expert, and management, and should coordinate with site management and the CIO or CSO. NIST SP 800-82 Data integration needs a smaller version of the same table: an OT owner for meaning, an IT owner for the path, and a named decision about who can change the mapping. If that table does not exist, the plant should not pretend the systems talk.
The High-cost repair for this cause is an IT-OT program: roles, a change path, and a rule for what may cross the DMZ. That is expensive in attention more than in licenses. It is still cheaper than a connector farm that no department will admit it owns.
The historical split: bought in different decades
The third cause is the one plants feel in their purchase orders. SCADA, the historian, MES, and ERP were rarely bought as one design. They arrived with different capital cycles, different vendors, and different promises. ISA-95 itself is a dated witness. Technology has evolved several times over since ISA-95 was established in the mid-1990s, yet the framework is still offered as a technology-agnostic model because its scope prioritizes activities, not products. ISA-95 The standard exists because the split was already old when the first MES and ERP waves were being sold as the fix.
Matthews gives the historian half of that timeline. Historians were first fed mainly by a DCS and stored in a proprietary database. Later they became a hub that could receive ERP and laboratory data and distribute it across intranets, the internet, and the cloud. ISA on process historians MES and ERP have their own arcs: on-prem packages, then suites, then cloud tenants, often with the plant still running the version validated when the line was commissioned. The systems can look current in a brochure and still belong to different decades on the floor.
CISA calls the resulting condition by its field name: brownfield. A cybersecurity challenge unique to ICS is brownfield deployments, meaning new technologies layered onto existing legacy infrastructure. CISA ICS The same layering happens with MES upgrades, historian migrations, and ERP rollouts. The new system sits beside the old one. Operators keep the old screen because it still runs the line. Finance keeps the ERP because it still closes the books. Engineering keeps the historian because it still holds ten years of tags. The plant now has four generations of truth.
NIST’s OT guidance explains why replacement is slow. Initially, OT systems had little resemblance to IT systems because they were isolated and ran proprietary protocols on specialized hardware. Ethernet, IP, and wireless devices are now replacing older technologies, which increases connectivity and also increases cybersecurity exposure. OT systems increasingly resemble IT systems as they adopt standard computers and network protocols. NIST SP 800-82 A controller that still has to be deterministic will not follow an office patch window. A validated batch record will not move because a cloud MES looks cleaner.
There is a second historical residue: the unofficial bridge. When the bought systems do not talk, people invent a fifth system. Sometimes it is a historian report. Often it is a workbook that joins extracts and becomes the morning number, which is the same trap described in the Excel trap in production. NIST’s operations-driven measurement work says making full use of manufacturing data is impeded by a lack of well-defined methods and standards, and that performance improvements are currently gained through targeted applications. NIST performance measurement Targeted applications are how brownfield plants survive. They are also how the unofficial fifth system becomes load-bearing.
The honest repair for a historical split is rarely a High-cost platform replacement on day one. Replacement is justified when the current system cannot carry the objects, the time model, or the approvals the plant already needs. Until that is proven, document what each decade’s system is allowed to decide, then add a connector or a context layer around the decisions that already hurt. Living with the split is legitimate when the unanswered join is rare and reviewable. Buying another generation of software because the last one never met the one before it is how plants spend a decade and still have four clocks.
A realistic repair path, with cost bands
A repair should match the cause. A protocol gap is not an organization problem. An ownerless interface is not a missing ERP module. Use three bands only, and keep them qualitative.
| Band | Repair | Use it when | Do not use it when |
|---|---|---|---|
| Low | point-to-point connector | One recurring transfer has a named source, destination, mapping, and owner | The plant is trying to invent a shared language by hiding joins in jobs |
| Med | context/name/time layer | The same asset, batch, or hour has several names and clocks, and people keep reconciling them by hand | The plant has not yet named the decision the layer is supposed to support |
| High | IT-OT program or platform replacement | Roles, access paths, or the current system of record cannot carry the objects and approvals already in use | The project is being sold as one system of record that will retire SCADA, the historian, MES, and ERP |
Start with evidence, not a vendor map. Write down one operational question that currently needs two or more of the four systems. Name the object, the time window, the source of each field, and the role that is allowed to act. NIST’s analytics project says manufacturers spend much of their effort identifying objectives, selecting the tool, and managing input data. For smaller manufacturers those applications can be prohibitively complex and expensive. NIST data analytics If the plant cannot state the question, a connector will only move the confusion faster.
The Low band is a point-to-point connector. MESA’s Metrics Working Group formed in 2004 because manufacturer members needed to understand what affects performance and to make a case for improvements in processes and IT systems. The later analytics materials still say to align the organization to the same goals and not ignore data governance. MESA manufacturing analytics A connector that copies one confirmed quantity from MES to ERP can be enough. Give it an owner, a mapping, a failure sign, and a review date. If it has to guess which asset is which, it is the wrong band.
The Med band is a context, name, and time layer. It does not replace MES or ERP. It keeps the relationships the four systems refuse to share: tag to asset, batch to order, shift clock to posting date, source to approval limit. That is the same problem treated as plant context in industrial data context, and it is why a historian can support five useful analyses without becoming the plant’s conversation, as in five analyses hidden in the plant historian. NIST already pointed at implementation languages for part of this job, including MESA’s B2MML as an ISA-95 implementation for linking ERP with manufacturing systems such as MES. NIST standards landscape Someone still has to govern aliases, clocks, and conflict rules. If that governance is missing, the layer becomes a prettier unofficial fifth system.
The High band is an IT-OT program or a platform replacement. Use the program when the block is organizational: no shared terms, no DMZ path, no owner. NIST says the most successful method for securing OT is a collaborative effort between management, OT engineers and operators, IT, and a trusted OT advisor. NIST SP 800-82 Data movement needs that same table. Use platform replacement only when the current MES or ERP cannot hold the object, the time, or the approval the plant already uses. Replacement does not create one system of record by slogan. SCADA will still own the live state. The historian will still own the long series. MES will still own execution. ERP will still own the commercial record.
WizeeMind belongs in that Med band, not the High one. It is a plant assistant that keeps source, time, asset, and approval limits attached to evidence. It does not replace MES or ERP, and it does not invent one system of record. NIST’s AI Risk Management Framework is intended for voluntary use, to help organizations incorporate trustworthiness into the design, development, use, and evaluation of AI systems. NIST AI RMF An assistant can carry the citations. It cannot approve a confirmation or declare that four clocks are now one.
Decide in that order. If the unanswered join is rare and reconstructable, live with the split and write the reconstruction down. If the join is frequent and the mapping is known, use a point-to-point connector. If people spend the week reconciling names and times, fund a context layer. And if no department will own the path, start an IT-OT program before anyone talks about replacement. The plant does not need one more system that claims to speak for the other four. It needs a named repair that matches the split it has.
Frequently asked questions
Why do SCADA, the historian, MES, and ERP fail to communicate?
The four systems hold related plant data, but they were bought for different jobs, store it in different forms, and sit on different sides of the IT-OT boundary. Related data is not the same as a shared conversation. ISA-95 exists because manufacturing control functions and business functions need an explicit interface, not an assumed one. ISA-95 NIST’s landscape adds that disparate systems can exchange and use product, production, and business data only when information standards carry the meaning. NIST standards landscape A plant can therefore be full of true records and still have zero communication that a supervisor can act on without a manual join.
Is a point-to-point connector enough to fix the split?
A point-to-point connector is a Low-cost repair when one recurring transfer has a known owner and a stable mapping. It does not fix names, clocks, or approval limits across the plant. Matthews’s caution still holds: open interfaces such as OPC, Ethernet, and SQL can move more data, but they do not decide what the value means in MES or ERP. ISA on process historians NIST’s analytics work makes the same cut. Integrating analytics with data acquisition and decision support remains a barrier. NIST data analytics Use a connector for a known transfer. Do not ask it to become the plant language.
Who should own an integration between plant systems?
Ownership has to sit with a named pair: an OT owner for the process meaning and an IT owner for the store and access path. A connector with no owner becomes another unofficial system. NIST’s OT guidance says control engineers cannot do this work without IT and management, because the two cultures are often significantly different. NIST SP 800-82 CISA’s brownfield note is the practical test. If a new feed is layered onto legacy infrastructure, someone has to own the mapping when the old name and the new name diverge. CISA ICS
Should the plant replace MES or ERP to make the systems talk?
Platform replacement is a High-cost move. It is justified only when the current system cannot carry the objects, time, or approvals the plant already needs. Most plants should not start there. ISA-95 still describes activities rather than products, so a new MES or ERP does not erase the level 3 to level 4 interface. ISA-95 MESA’s analytics materials keep the management test in view: align the organization to the same goals and put data governance in place before treating software as the strategy. MESA manufacturing analytics Replacement can still be right. It is the last band, not the first diagnosis.
When is living with the split the right decision?
Living with the split is reasonable when the unanswered question is rare, the manual join is reviewable, and a connector would create a silent authority nobody can audit. NIST’s operations-driven work already treats targeted applications as the common path, because well-defined methods and standards for using the data are often missing. NIST performance measurement An assistant can keep source, time, asset, and approval limits attached while the plant lives with that split. NIST’s AI Risk Management Framework is voluntary guidance for trustworthy use of AI systems, not a license to invent one system of record. NIST AI RMF If the join is rare, write the reconstruction down and spend the High-cost attention on a problem that actually repeats.
Sources
- ISA-95 Standard: Enterprise-Control System Integration
- ISA InTech: New roles for process historians
- NIST: Current Standards Landscape for Smart Manufacturing Systems
- NIST: Guide to Operational Technology (OT) Security
- CISA: Industrial Control Systems
- MESA: Manufacturing Analytics
- NIST: Data Analytics for Smart Manufacturing Systems
- NIST: Operations-driven Performance Measurement for Smart Manufacturing Systems
- NIST: Artificial Intelligence Risk Management Framework