Why 73% of MES Implementations Fail, and How to Pick the Right Approach
Every "build vs. buy MES" guide you've read was written by a company selling MES software. Platform vendors argue for their platform. No-code vendors argue for no-code. Those guides can be useful, but none of them will tell you when their product is the wrong answer for your floor.
We build custom manufacturing execution systems. We also tell clients when off-the-shelf is the better call. This is our attempt at a guide that helps you decide.
What MES software does, and why it's hard to buy off-the-shelf
Manufacturing execution systems sit between your ERP and your shop floor. The ERP handles business logic: orders, inventory, financials. The shop floor is the physical side: machines, operators, materials. MES translates between the two in real time. It routes work orders to the right workstations, tracks WIP, enforces quality checkpoints, collects production data from equipment, and feeds results back upstream.
That sounds straightforward until you look at a real factory floor. Your CNC machines from 2009 communicate over Modbus. Your newer machining centers use OPC-UA. Your legacy assembly line runs a proprietary PLC protocol that predates USB. Your quality inspection process involves a torque wrench with a Bluetooth radio and a barcode scanner running firmware from 2014. Your ERP is SAP. Your shift supervisors use tablets. Your maintenance team uses paper.
That's a normal manufacturing operation, and it's the main reason off-the-shelf MES software causes so much friction. Vendors built their data models around how they imagined a factory floor works, and your floor almost certainly works differently.
The full cost of off-the-shelf MES
The sticker price of commercial MES software is the least useful number in the buying decision. The number to look at is five-year total cost of ownership, and it's almost always larger than manufacturers expect.
Gartner's analysis found 70% of ERP and MES implementations fail to meet their stated objectives. Discrete manufacturing specifically sees 73% failure rates with average cost overruns of 215%. As of 2023, 47% of global enterprise software implementations exceeded budget, and most projects ran 3-4x over their initial projections by project close. These numbers aren't flukes. They're what you'd expect when generic software gets bent to fit specific operational requirements.
The hidden costs stack up in several layers.
Customization comes first. Vendors charge $150-$300 per hour to adapt their product to your workflow, and every non-standard requirement is billable. Nearly everything that's specific to how you compete counts as non-standard.
Integration is second. Your machines, PLCs, and sensors weren't designed around the vendor's API, so each integration point is a negotiation between your hardware's data format and their schema. When one of those integrations breaks at 2am during a shift, figuring out who owns the fix turns into a conversation about SLA clauses.
Data migration is third. Production history, genealogy records, and quality data built up over years have to be mapped, cleaned, and validated against the new system's structure. Organizations routinely find that their "clean" production database contains years of inconsistent formats, duplicate records, and missing serial number chains.
Training is fourth. Floor workers who have built muscle memory around one workflow push back on new interfaces that come close to their mental model without matching it. Retraining costs money and reduces output during the transition.
License lock-in is the long-term cost that's easiest to miss. Per-seat or per-module pricing grows with your headcount and usage. Vendors get acquired. Products reach end-of-life. New owners change pricing. The MES you picked because it fit in year one may look very different commercially in year five. Companies that choose off-the-shelf often spend 2-3x more on customizations, integrations, and workarounds over five years than a custom build of the same capabilities would have cost. The spending is spread across so many invoice lines that the pattern is hard to see until a post-mortem.
The full cost of building custom
Custom MES development has real disadvantages, and an honest advisor should say so.
Upfront cost is higher. A custom build that does what an $80,000 commercial license does will cost more in year one. Time-to-first-value is longer. A vendor demo environment can go live in days, while a custom system needs requirements gathering, architecture design, development, integration testing, and validation on the floor. Depending on complexity, initial deployment typically takes 4-10 months.
You own the maintenance. When something breaks, there's no vendor support queue. The fix falls to your development partner or your internal team. That's manageable with the right partner and good documentation, but it's a real operational responsibility.
Commodity functions aren't worth building. Basic OEE reporting, standard shift scheduling, and generic production dashboards are solved problems. Building them from scratch delays the work that differentiates your operation and spends development budget reinventing the wheel.
Custom development is also the wrong choice for manufacturers who are still standardizing their processes. If your floor operations are inconsistent or changing quickly, a custom system locks today's process into software. Off-the-shelf systems built for configurability handle operational change more easily during that phase.
The five questions that decide it
How unique is your production floor?
Count the machine types, communication protocols, and quality process steps that differ from the industry standard. If there are a lot of them, each one is a customization cost inside an off-the-shelf product. If your floor is mostly standard, commercial software was probably built with you in mind.
How many integration points does your MES need?
List every system the MES has to exchange data with: ERP, SCADA, quality management, maintenance, inventory, customer portals. Each integration the off-the-shelf vendor doesn't have a prebuilt connector for becomes a custom project inside their platform. It's billed at their professional services rates and maintained by a team that benefits from keeping you dependent on those services.
What does the five-year license cost look like?
Take the per-seat price, multiply by your projected headcount in year five, add estimated module expansion costs, then add a 30% buffer for price changes after an acquisition or contract renewal. Compare that number to a one-time custom build plus annual maintenance at 15-20% of the build cost. If you do this math honestly, the crossover point usually becomes clear within a few minutes.
Does your MES need to manage physical hardware?
Connecting to sensors, managing edge devices, updating firmware on connected tooling, and processing real-time machine data at low latency are things commercial MES vendors treat as integration problems. Custom systems treat them as core requirements. If IoT hardware is central to your MES rather than an add-on, the architecture favors custom development from the start.
What happens when it breaks during a shift?
Decide how much downtime per incident you can accept. Then read your vendor SLA carefully. What's the guaranteed response time for a production-blocking issue, and what does escalation look like at 3am on a Saturday? For custom systems, the answer depends entirely on your partner and your support agreement. Neither option is automatically better, but the tradeoffs differ, and manufacturers often find out the hard way that the commercial SLA they counted on protects them less than they thought.
Where most manufacturers end up
Most manufacturers who get this right don't pick one extreme. They run a hybrid: off-the-shelf for commodity functions and custom software for the parts that set them apart from competitors.
A precision parts manufacturer running 24/7 production might use a commercial OEE platform for standard shift reporting and build a custom real-time SPC system that enforces its proprietary quality tolerances across 200 measuring instruments, because that quality process is why customers buy its parts. An automotive supplier might use a vendor's production scheduling module and build custom tooling integration that tracks torque data from its proprietary assembly fixtures, because its OEM customers require that traceability.
So the useful question is which parts of your operation set you apart and which are table stakes. Buy commodity software for the table stakes. Own the parts that differentiate you.
What one team owning the stack means for MES
A manufacturing execution system spans hardware and software in ways most enterprise software doesn't. Sensors report to edge devices. Edge devices push to on-premises servers or cloud infrastructure. Servers run business logic and expose interfaces to PLCs, operator terminals, quality stations, ERP systems, and management dashboards. Mobile apps push alerts and approvals to supervisors on the floor.
When that system is assembled from several vendors (a hardware integrator, an MES platform vendor, an IoT middleware provider, and an app developer), nobody clearly owns the integration. Each vendor's support team has a natural incentive to find the problem in someone else's system. Incidents that cross the stack without a clear owner get fixed slowly, or never, while production waits.
A single team that owns the full stack, from the sensor firmware to the supervisor dashboard, deals with that problem differently. Nobody argues about whose API is at fault. The system either works or it doesn't, and one team is accountable for fixing it.
That argument applies beyond custom development. Whatever you choose, look at how many organizations will be involved when something breaks and who is accountable across that relationship.
The decision in summary
Off-the-shelf MES fits manufacturers with standardized floors, limited integration complexity, shorter time horizons, and workflows that are still being formalized. The five-year total cost stays manageable when customization needs are low.
Custom MES fits manufacturers with complex integration environments, proprietary processes that give them a real competitive edge, IoT hardware at the center of their data collection, and operations mature enough that the current workflow is worth putting into software. When integration and customization needs are high, the five-year total cost of custom often comes in below off-the-shelf.
A hybrid approach fits most manufacturers: evaluate each functional area on its own, buy what's commodity, and build what's distinctive.
Before anything else, do the five-year TCO comparison honestly, with realistic integration labor estimates on the commercial side and realistic build-plus-maintenance estimates on the custom side. That number will tell you more about the right answer for your floor than any product demo or analyst report.
If you're working through this decision and want a technical conversation about what a custom build would involve for your operation, reach out at hello@detroitcomputing.com. We'll tell you if it makes sense.