It's Time to Rethink MES for Complex Production Tracking

Ask ChatGPT which MES is best suited for complex production tracking, you’ll likely see a list of familiar names. Siemens. Rockwell. GE. Honeywell. Dassault. AVEVA.

Plants that assemble the same products on predictable cycles can probably pick a solution from that list and be okay. But plants asking the same question, with high-mix portfolios, aggressive NPI calendars, multi-site footprints, and regulated change cycles, probably will not.

The reason is a shift in what "complex" means. Legacy solutions like those listed above were architected in a time when complexity meant depth of route. But in 2026, complexity shows up in the rate of change. That mismatch shows up in change cycles measured in quarters, in operators absorbing what the system cannot, and in multi-site rollouts that stall before plant three.

This post makes the case for a different filter on the 2026 shortlist. The sections ahead cover what changed about complex production, what analyst data from LNS Research, Symestic, and Excellerant says about the cost of architectural rigidity, the five properties a fit-for-purpose MES has to have, and a framework for vetting composable platforms against a real evaluation, and where Tulip sits inside that frame.

By the end, you’ll have an evaluation lens to take into your next vendor conversation that reflects the reality of today’s production environments.

Marketing content from legacy MES vendors treat complex production as a genealogy problem. The assumption is multi-stage routings, split and merge operations, rework loops, serial and lot granularity recorded back to raw material, and a part that can be tracked through forty steps with every one of them reconstructable on audit request.

This criteria is table stakes for an enterprise execution system.

What breaks is the velocity of change. A forty-step route that runs unchanged for three years isn’t necessarily a challenge. A twelve-step route that changes weekly across nine sites is.

Three production patterns expose this clearly:

• High-mix, low-volume operations where product variants launch frequently and each variant carries workflow differences that have to propagate quickly.

• NPI-heavy businesses where new products move from engineering release to validated production workflow in weeks, not quarters.

• Regulated multi-site operations where a spec change in one market has to propagate to every site with controlled rollout, version history, and revalidation.

Under any of those patterns, the architectural question shifts. Most vendors can answer "Can this MES track a complex route?" without hedging. "Can this MES absorb route change without a vendor-led rebuild?" tends to produce longer answers, qualifications, and references to professional services.

Legacy MES was built on a stable-process assumption. The data model is rigid, configuration paths run through the vendor or a specialist systems integrator, and significant changes mean a controlled rebuild, a revalidation cycle, and a calendar measured in quarters. That design no longer serves manufacturers in a world where change is a daily expectation.

LNS Research's 2026 "Architecture Calls Your Bluff" series puts a name on what this gap practically looks like on the floor. When the system cannot support new requirements, the extra work gets pushed on to the people running the process. A process engineer keeps a spreadsheet of workflow changes the MES cannot capture. A quality manager re-emails revised instructions after every spec update because the official system takes a quarter to refresh.

LNS describes that load as a “cognitive burden”, and the underlying condition an “architectural compression”. Both describe a system that has run out of room to manage what a manufacturer needs.

Symestic's 2026 reporting on composable MES puts the timing of that failure into numbers. Typical change-management cycles in legacy MES environments run 6 to 18 months per significant workflow change. Equivalent cycles in composable environments run roughly 3 weeks. The midpoint matters less than the shape. Every week between "we observed a problem" and "the fix is live on the floor" compounds into scrap, rework, and missed commit dates.

Excellerant's February 2026 data adds the plant-level cost of data disconnect. Manufacturers lose an average of 25 production hours per month to unplanned downtime driven by fragmented data. The same article reports that production adjustments can take more than 48 hours to implement in disconnected environments.

Implementation time is the final piece of the same picture. Legacy MES deployments typically run 18 to 36 months to first site go-live, with multi-site rollouts extending years beyond that. A timeline that long assumes the world sits still during implementation, which is a bet that has not paid off in any manufacturing environment over the past decade.

The specifications for a system that can keep up in a change-heavy environment cluster into five areas.

Independent capability evolution: A change to deviation management, work instructions, or digital dashboards should not require a full-system release or a requalification of unrelated functionality. When capabilities are coupled, every small change becomes a large change, and the cost of a large change is usually measured in quarters.

Operator-facing design as a first principle: Frontline execution has to be the starting point of the system. When an MES is architected as a database with a clunky UI added on top, it creates challenges that show up in training time, in workaround behavior, and in the kind of errors quality teams spend weeks investigating.

Site-level composition inside governed standards: Each site needs to compose the capabilities it uses within a globally governed framework. We’ve found the right framing for this is global standardization with local flexibility. Both halves of the phrase are important. Standardization without flexibility produces rollouts that ship the same workflow everywhere and quietly accumulate workarounds at each site. Flexibility without standardization produces the audit findings that make quality leadership stop trusting the system.

A data architecture that survives the next five years of change: Capabilities with clear, exposed data surfaces are easier to integrate, easier to extend, and easier to adapt when new requirements arrive. That last category is the one most architects underappreciate. The function-specific AI agents that will be embedded in quality, NPI, and root-cause analysis over the next two years need clean data surfaces to work against. A monolithic data model with opaque internal logic offers none.

Governed adaptability: Regulated environments treat versioned workflow history, role-based permissions, staged rollouts, and auditable change trails as a non-negotiable. At multi-site scale, those controls become necessary regardless of regulation, because the alternative is a version drift no one can reconcile six quarters later. Traceability and controlled change both live here, and both show up in evaluation criteria like traceable execution records.

A composable MES breaks the monolithic application into independently deployable capabilities. Composable systems generally align with MACH principles (microservices, API-first, cloud-native, headless) and can be assembled from Packaged Business Capabilities (PCBs) which could include anything from electronic batch record to visual work instructions, production dashboards, or equipment integration. Each capability is a discrete unit with its own data surface, its own release path, and its own governance envelope.

Tulip goes a step further by providing the operator-facing execution layer, the data and connectivity model, and the governance surface within the platform. The manufacturer composes an MES-class system from shared building blocks. Templated app suites provide a starting point built on years of supporting manufacturers across a variety of complex and regulated industries.

While the composable framework has been dominant in e-commerce and customer-facing tech stacks for over a decade, its credibility in industrial software is finally catching up. Gartner's 2022 Magic Quadrant for MES projected that sixty percent of new MES deployments would be assembled from composable technology by 2025, a prediction we’ve seen come to fruition in recent years.

The 2025 Gartner Market Guide for MES includes composable-native vendors alongside the incumbents, and today’s Representative Vendor list looks a lot different than it did three years ago.

For an Industrial Architect defending a composable choice inside a steering committee, that shift is important. The architectural argument no longer has to be invented from scratch. It has been published, benchmarked, and picked up by the analyst firm that many IT-led organizations use as their shortlist anchor.

FDA 21 CFR Part 11 and EU GMP Annex 11 are architecture-agnostic. Both specify controls a GxP-relevant system has to provide: authenticity, integrity, controlled access, audit trails, and electronic signatures. Neither specifies that those controls have to live inside a single monolithic application. A composable platform that provides the controls meets Part 11 and Annex 11 on the same terms a monolith does.

EU GMP Annex 22, entering force in 2026, extends Annex 11 to AI and machine-learning-driven GxP systems. It requires traceability of model behavior, data lineage, and human oversight of automated decisions. Composable architectures, where each capability has a transparent data flow and a clean interface, are structurally better positioned to satisfy Annex 22 than monolithic systems where the internal logic is opaque to the people who need to validate it.

The 2010s narrative that regulated enterprises require a monolith had an architectural basis at the time, because cloud-native deployments could not demonstrate the governance surface regulators expected. That condition no longer applies, and we’ve helped dozens of pharma, biotech, and medical device manufacturers implement composable platforms in validated GxP environments.

The compliance question has shifted from "can composable meet the bar?" to "does this specific composable platform provide the controls?" That is a change from an architectural-category question to a vendor-evaluation question, which moves compliance review back inside the individual RFP where it belongs.

Gartner projects that by 2027, half of generative AI deployments in the enterprise will be function-specific. In manufacturing, that means embedded agents and copilots inside specific workflows like deviation review, NPI routing, and root-cause analysis. Each agent works against a discrete capability.

AI needs clean data and well-structured processes to work against. A deviation-review agent needs access to data with the right context and the right permissions. It does not need, and usually cannot use, the entire MES data model. Monolithic MES systems with obscure internal logic make this nearly impossible. The agent usually gets narrow, surface-level access that limits what it can do, and the workaround is a custom integration into a proprietary stack with its own release cadence.

In a composable architecture, each app or PBC exposes its own data surface. Agents are given access to only the necessary context and permissions, and can be evaluated, deployed, and retired independently, which is what function-specific AI looks like in practice.

That is a structural property, and it cannot be retrofitted into a monolith by adding a built-in AI feature list. The MES that will be ready for the AI wave of the next two years is the one whose capabilities can accept new agents on their own release cadence, with their own governance, and without a platform-wide release cycle.

Seven criteria separate a composable platform that matches a change-heavy environment from one that just uses the vocabulary.

1. Measure change velocity honestly before shopping: Count the process changes, NPIs, material substitutions, and regulatory updates your operations encountered over the last twelve months. If the number is high and rising, architectural fit will matter more than feature completeness.

2. Ask for the mean change cycle, with customer names: "How long between 'we need to change this workflow' and 'the change is live and validated on the floor'?" Any composable platform should be able to answer with specific customer examples in weeks; a response measured in months is the signal to ask harder follow-up questions.

3. Probe the extension surface: What does it take to add a new capability or integrate a new system? The spectrum runs from native configuration the platform supports out of the box, to composition inside the platform by a technical user at the customer, to a vendor engagement with its own statement of work. Where a platform lands on that spectrum tells you how much vendor lock-in the architecture implies.

4. Validate multi-site governance: Can the platform standardize globally while allowing local composition? What does controlled rollout of a single workflow change across several sites look like in practice? If the answer is a roadmap item, assume it will remain one.

5. Evaluate AI readiness as architecture, not feature list: Where do agents plug in? What data surfaces do they have access to? What governance controls apply to agent actions? These are structural questions, and a list of built-in AI features answers none of them.

6. Confirm specific regulatory controls: For whatever compliance regime applies, whether Part 11, Annex 11, Annex 22, AS9100, or ISO 13485, walk the controls one at a time. A general claim of compliance is not a substitute for a control-by-control review.

7. Use the 2025 Gartner Market Guide for MES as the shortlist anchor: The Representative Vendor list is one of the cleanest independent filters for narrowing the MES evaluation. Vendors outside that list have a higher burden to justify inclusion in an enterprise shortlist.

The default shortlist we highlighted at the beginning of this post answers a question from a decade ago. The 2026 question is which architecture can absorb change at the rate the plant changes.

If your business has stable, long-running product lines, a low change-velocity profile, and a single deeply-integrated historian, a traditional MES can still be defensible. Fewer plants meet that description every year, and the ones that do are not the audience for this post.

For the more common 2026 case, which involves high change velocity, NPI pressure, multi-site rollout ambitions, or a regulatory environment that forces controlled change, a composable manufacturing platform like Tulip is the category that matches the problem.

The most useful next step is taking the seven-question evaluation framework above into the next vendor conversation. Architecture-first questions tend to surface real differences between composable-native platforms and incumbents that have added a cloud option.

If you're interested in exploring how Tulip can help improve your operations, reach out to a member of our team today!