MES for Small Manufacturers: An Honest Buyer Guide (50-500 Employees)
Most MES content is written for plants with hundreds of machines and a validation team. This guide is for the rest: factories with 50 to 500 employees, running on spreadsheets and whiteboards, getting pressure from auditors or big customers to produce data they cannot currently produce. You sit between "Excel is fine" and "we need an enterprise rollout." Here is what buying actually costs, what you actually need, and how to avoid the demos that waste a quarter.
You May Not Need a MES Yet
Start with the uncomfortable paragraph, because it will save some readers real money. If you cannot answer "what is our OEE on line 2?" — not from memory, from data — you are not ready to buy a system that reports OEE. Measure it manually first, on paper or a spreadsheet, for a month. The exercise teaches you where downtime actually goes and gives you a baseline to check any vendor's numbers against. Skip it and you will buy software, then discover your real problem was unclassified downtime nobody was trained to log.
The second honest case for waiting: fat margins, no customer audits, no traceability demands, and no compliance pressure. A MES pays for itself by recovering lost minutes and surviving audits. If neither problem exists at your plant, the money is better spent on maintenance or capacity. If you do face audits, traceability requirements, or chronic overtime from expediting, keep reading. First understand what a MES actually is, because vendors will happily sell you one that solves the wrong problem.
The Real Cost Stack
Vendors love quoting "license price" because it is the smallest number in the stack. Budget all five layers:
License. Typical small-factory models: per-machine pricing around $720 per machine per year; per-seat around $49 per user per month; per-interface pricing of $100-250 per connection with a 10-interface minimum floor; or a flat site license. On a 20-machine floor with 10 terminals, per-machine and per-seat models land in the same ballpark — a few thousand dollars a year either way. Watch the per-interface floor: if you need 4 connections but the floor is 10, you are paying for 6 you don't use. Ask which model fits your shape, not which is cheapest on paper.
Implementation and integration. This is usually the largest line item and the one least often quoted. Connecting terminals to machines, mapping your work orders, configuring downtime reason codes, testing genealogy. Ask for a range in writing before the second demo.
Hardware at stations. Terminals, scanners, and network at each critical station. Industrial touchscreens cost more than tablets, and cheap tablets die on a hot floor.
Training. Operators must log downtime with reasons and confirm output at terminals. If training is one afternoon, adoption will fail quietly in week three and your data goes back to fiction.
The recurring cost of bad data. This one never appears on a quote. If operators skip scans or fat-finger downtime codes, you pay the license and get nothing. This is a process cost, not a software cost, and it is why the pilot approach below matters more than the vendor.
Most vendors publish none of these numbers. That is a fact, not an accusation — but it means your job is to demand a written total-cost estimate covering all five layers before demo #2. A vendor who won't put numbers in writing is telling you something.
What a Small Factory Actually Needs
At 50-500 employees, the feature list that pays for itself is short:
- Work orders with terminal execution. The operator sees the job, confirms counts, and flags problems at a station terminal — not on a printed traveler walked back to the office.
- Downtime with reasons. The single highest-value dataset most factories have never had. Without it, OEE is a number without a cause.
- Traceability and genealogy. Which lot went into which unit, and where each unit shipped. This is what auditors and recalls actually ask for.
- OEE per machine and line. Availability, performance, quality — with the downtime reasons to explain them.
- Quality checks at the point of production. In-process checks that stop the line on failure, not a clipboard audit a week later.
What you do not need at this size:
- Full advanced planning and scheduling (APS). Finite-capacity optimization is heavy to maintain and worthless with garbage inputs. A disciplined scheduling spreadsheet beats an unconstrained optimizer. If scheduling is your main pain, read that comparison before buying anything.
- Digital twin. Simulation of a plant whose real data you don't yet trust is theater.
- Validated QMS modules. Full validation is for regulated industries with the staff to maintain it. Basic in-process checks cover most small-factory quality needs.
Every one of these three will appear in a demo as "included in the platform." Included does not mean free: you pay in license tier, implementation hours, and the training burden of features nobody uses.
The Buying-Process Checklist
Pilot one line first. Not one machine, not the whole plant. One line with a real pain and a cooperative supervisor. Two to three months, real orders, real scrap. A pilot that runs on test data proves nothing.
Define the win condition before starting. Written, numeric, boring. Example: "downtime logged with reasons on 90% of stops, OEE baseline established for the line, genealogy report generated for one customer audit." If you cannot state the win condition, the pilot becomes a science project that fades.
Confirm data ownership and export. Your production data must be exportable in open formats — CSV at minimum — by you, without the vendor's involvement. Test the export during the pilot, not after you have two years of data hostage.
Ask what happens to your data if you leave. Retention period, export format, whether historical records are wiped. Get it in writing before signing.
Test offline behavior. Networks drop on real floors. When the connection fails, do terminals keep running and sync later, or does production stop logging? Ask for a demonstration, not a reassurance.
Nail down who does integration. Machine connections and ERP handoffs are where implementations stall. Is integration done by the vendor's own engineers or a systems integrator you contract separately? Who is accountable when the PLC on the 1998 packaging machine won't talk? An unnamed SI in the fine print is a red flag in itself.
Red Flags in Vendor Conversations
Module tours before diagnosis. A vendor who demos 40 minutes of features before asking what hurts is selling a catalog, not solving a problem. You should be talking more than they are in the first meeting.
"Contact us for pricing" plus a 12-month commitment. Quote-required pricing hides the number until you are emotionally invested, and the annual lock-in appears with the quote. Smaller vendors publish prices; ask why this one won't.
"Transformation program" language. If the pitch involves journey maps and change-management frameworks instead of "here is your downtime data on one line by Friday," the implementation will be as abstract as the pitch. Good MES deployments are specific: stations, terminals, reason codes, go-live date.
One counterexample pattern worth trusting: a vendor who asks to see one line, proposes a pilot with a numeric win condition, and quotes a total cost across all five layers of the cost stack. That is what a serious partner looks like, regardless of company size.
Frequently Asked Questions
How much does a MES cost for a small factory?
With typical models — $720 per machine per year, $49 per user per month, or $100-250 per interface with a 10-interface floor — software for a small deployment usually lands in the low thousands of dollars annually. Implementation, station hardware, and training typically add more than the license itself in year one. Always demand a written estimate covering all five cost layers before demo #2; if a vendor won't give numbers, walk.
Can we start with just one line?
Yes, and you should. One line, two to three months, a written win condition, real orders. If the pilot hits its numbers, expand to the next bottleneck line; if it doesn't, you have learned where your data discipline breaks for the price of one line. Any vendor who insists a small factory must roll out plant-wide at once is optimizing their revenue, not your risk.
MES vs ERP: which should we buy first?
If your ERP handles orders, inventory, and finance acceptably, buy the MES — it fixes what the ERP cannot see: what happened at the machine. If you have no ERP at all and basic order tracking is chaos, that comes first, because the MES will need something to talk to. The systems answer different questions; the full MES vs ERP comparison breaks down the boundary in detail.
One Line, One Pain, One Pilot
Voltrus MES starts the way this guide recommends: one line, real orders, a numeric win condition. Work orders at terminals, downtime with reasons, traceability, hourly OEE. See how your OEE compares by industry, then bring us your worst line.
See Voltrus MES