MES software selection: requirements that survive go-live

MES selection fails in predictable ways: buying for the demo instead of the tenth line, underestimating integration ownership, and discovering at go-live that the "standard connectors" need six months of custom work. An MES lives for a decade or more and touches every operator, so the selection process deserves the rigor of a capital project. This guide structures it around what actually predicts success.

Start from operations, not features

Write the requirements from the shop floor's daily work before talking to vendors: order dispatch and sequencing, work instructions, data collection (manual and automatic), genealogy and traceability depth, quality checks and nonconformance handling, labor tracking, and downtime capture. For each, state the must-have versus nice-to-have and the transaction volumes — an MES that handles 50 work orders a day may struggle with 5,000 serialized units an hour, and the difference only shows under load testing.

Define the system boundaries explicitly using the MES vs ERP checklist: which system owns routings, BOMs, inventory moves, and quality plans. Most MES disappointments are boundary disputes discovered late — two systems both claiming genealogy, neither owning rework. If serialization and traceability are core requirements, read the track and trace basics first so the requirements use the industry's vocabulary.

Evaluate integration as the main risk

Integration — to ERP, PLCs, historians, lab systems, and label printers — is where MES budgets die. For every interface, demand specifics: named connector versions, supported protocol revisions, who configures the mappings, and reference customers running that exact interface in production. "We integrate with SAP" is not an answer; "our certified connector supports SAP S/4HANA 2023 via MII/oData with these IDoc mappings, configured by our consultants in roughly three weeks per plant" is the beginning of one.

Prefer standards over adapters: ISA-95/B2MML models for ERP exchange, OPC UA for equipment data, and a unified namespace strategy so the MES consumes one coherent model rather than fifty bespoke drivers. And clarify data ownership: your production history must be exportable in usable form without the vendor's professional services — negotiate the exit in the entrance contract.

Pilot like you mean it

Run the pilot on a real production line with real operators, real volumes, and real network conditions — not the vendor's cloud sandbox. Define pass/fail criteria in advance: transaction response times at peak rate, offline behavior when the network drops (a line that stops because the MES is unreachable is a design failure), label and device integration working end to end, and operator task times no worse than the current process. Score vendors on the pilot's evidence plus lifecycle factors: upgrade cadence and cost, multi-site deployment model, validation support for regulated industries, and the financial stability to exist in ten years. The cheapest license with the most expensive integration is the most expensive MES.