Connecting MES to ERP — Architecture Patterns That Work

Connecting MES to ERP is one of those projects that sounds straightforward in a vendor demo and turns into 6–9 months of scope discovery in the real world. The reason: ERP and MES were built in different decades for different jobs, and the data they exchange has to translate cleanly between two different worlds.

This piece covers the four architecture patterns we see succeed in US manufacturing — what each looks like, where each fits, and the failure modes to plan around.

Why MES–ERP integration matters

ERP plans production: work orders, BOMs, capacity, financial roll-up. MES executes production: real-time work-order status, cycle times, scrap, downtime, operator activity, quality data. When they don’t talk:

• ERP shows yesterday’s WIP. MES shows now. Planners pick.
• Cost roll-up uses standard hours, not actual hours. Margin decisions go wrong.
• Quality data lives in MES forever; finance never sees the cost of rework.
• Operators key data twice: once on the floor, once into ERP at end of shift.

Pattern 1: Direct point-to-point integration

ERP and MES exchange data directly via API or database links. No middle layer.

• Best for: single-site operations, simple equipment counts, both vendors with strong APIs.
• Pros: simplest topology, lowest latency, minimal moving parts.
• Cons: tight coupling. Either system’s upgrade can break the other. Doesn’t scale to multiple sites with different MES instances.

Pattern 2: Middleware/iPaaS hub

An integration platform (MuleSoft, Boomi, Workato, custom Node/.NET service) sits between ERP and MES. All exchanges flow through the hub.

• Best for: multi-site operations, multi-vendor MES across sites, manufacturers also integrating CRM, WMS, BI.
• Pros: loose coupling, audit logs of every message, easier to add a third system later, retry/replay built in.
• Cons: middleware has license cost and maintenance overhead. Adds a step that can fail.

Pattern 3: Data lake / event-streaming backbone

Both systems publish events to a streaming layer (Kafka, AWS Kinesis, Azure Event Hubs). Downstream consumers (BI, ML, custom analytics, the other side of the integration) read from the stream.

• Best for: manufacturers $500M+ with serious analytics ambitions, multiple plants, multi-cloud infrastructure.
• Pros: infinitely scalable, replayable, opens door to predictive analytics and digital-twin work.
• Cons: overkill for most mid-size operations. Requires data engineering capacity.

Pattern 4: Embedded MES inside ERP (or vice versa)

Some platforms bundle MES capability directly into the ERP (Plex, Epicor Kinetic with MES module, IFS Cloud) or build the manufacturing layer as native to the ERP data model. No cross-system integration because there’s no second system.

• Best for: greenfield manufacturers replacing everything, or those willing to commit to a single vendor for both planning and execution.
• Pros: integration problem doesn’t exist. Single data model.
• Cons: vendor lock-in deepened. Fitness for niche manufacturing scenarios varies.

The data flows you must define

Whichever pattern you pick, these flows must be designed explicitly:

• Master data sync: items, BOMs, routings, work centers, employees. Direction and frequency.
• Work order lifecycle: ERP creates → MES executes → MES reports back actuals → ERP closes and posts cost.
• Inventory transactions: MES consumption events must reduce ERP inventory in real time, not at end of shift.
• Quality data: MES capture → ERP for nonconformance, scrap, and inspection records.
• Operator labor: MES time-on-job → ERP for actual labor cost roll-up.

Failure modes to plan around

• Master data drift. Same part, different IDs in ERP and MES. Pin one as authoritative and replicate.
• Clock skew. MES events stamped seconds before ERP receives them lead to ordering issues. Use a single time source (NTP) and pass timestamps in every message.
• Network partitions. Plant network goes down for 20 minutes. MES queues events locally and replays. ERP must handle out-of-order replay.
• Validation failures. ERP rejects an MES message because the work order is closed. Define a dead-letter queue and a process to investigate.
• Schema evolution. One side adds a field. Plan versioning into every interface from day one.

FAQ

Should we integrate before or after replacing ERP?

Before. The current MES connecting to current ERP teaches you what data flows you need. Then design the new ERP rollout to handle those flows from go-live.

Can we use ISA-95 as a guide?

Yes. ISA-95 (Levels 0–4) is the canonical reference for what belongs in MES vs ERP. Use it as a contract negotiation tool with vendors and a discipline check on your own architecture.

How long does MES–ERP integration take?

Pattern 1 (point-to-point): 8–14 weeks. Pattern 2 (iPaaS): 14–24 weeks. Pattern 3 (event streaming): 6–12 months. Pattern 4 (embedded): no integration project, but the underlying ERP/MES rollout is 9–18 months.

Bottom line

Pick the pattern that matches your scale and ambition, not the pattern your vendor is selling this quarter. Most US manufacturers in the $50M–$500M range land on Pattern 2 (middleware) for good reasons. Get the data-flow design right early, plan explicitly for failure modes, and the ongoing maintenance becomes manageable.