Legacy systems in manufacturing: modernise or replace?
Should manufacturing firms modernise or replace legacy systems?
Modernise the enterprise layer and replace only at the edges. A manufacturing estate is not one system, and the components age at different rates. Wrapping the ERP while replacing shop-floor clients and spreadsheets is usually the right shape; replacing everything at once is the recognisable failure mode.
Modernise the enterprise layer and replace only at the edges. A manufacturing estate is not one system: an ERP, an MES, a historian, a quality module and a layer of spreadsheets each age at different rates and carry different risk. Wrapping the ERP while replacing shop-floor clients and the spreadsheets around them is usually the right shape. Replacing everything at once is the recognisable failure mode.
This article takes your estate apart component by component, gives a verdict for each, sets out a five-phase sequence that survives a live production schedule, and shows where AI attaches in phase one rather than in year two.
Your legacy estate is not one system
The first mistake in a manufacturing modernisation debate is arguing about the ERP when the pain is somewhere else. Ask where the work actually happens and the answer is rarely the system of record. It is a supervisor's spreadsheet that schedules the line, a quality inspector's clipboard, a maintenance log in a shared folder and an MES screen that three people know how to use.
That matters because those artefacts are cheap to replace and expensive to keep. The ERP, by contrast, is expensive to replace and usually adequate to keep, because it does the two things it must do: it holds the master data and it closes the books. A modernisation plan that starts with the ERP is starting with the hardest, lowest-return component.
The second mistake is treating integrations as plumbing rather than as assets. The connection between your ERP and your MES encodes years of decisions about part numbering, work centre mapping and unit conversion. Rebuilding it from scratch means rediscovering all of that in production.
A component-by-component verdict
Run this table against your own estate and mark each row. Most manufacturers find one or two replace rows and the rest wrap or keep.
| Component | Usual verdict | Why | Where AI attaches |
|---|---|---|---|
| ERP core and finance | Wrap, do not replace | Holds master data and statutory records; replacement means migrating years of history | Natural language reporting over a read model of it |
| MES and production execution | Wrap, replace modules selectively | Deeply coupled to equipment; full replacement stops the line | Anomaly detection on cycle times and scrap rates |
| Historian and sensor data | Keep, add an export path | Purpose-built and usually fine; the gap is access, not capability | Condition monitoring and maintenance scheduling |
| Quality module and paper records | Replace | Often paper or spreadsheet; low switching cost, high return | Vision-assisted inspection with a full decision log |
| Shop-floor clients and terminals | Replace | Unsupported runtimes, poor usability, no offline handling | Guided workflows and an SOP assistant in local languages |
| Supervisor spreadsheets | Replace | They are the real scheduling system and nobody owns them | Forecasting and constraint-aware scheduling support |
| Integrations between the above | Rebuild behind one layer | The business logic hidden here is the asset worth preserving | The read model everything else is built on |
A five-phase sequence that survives a production schedule
Manufacturing gives you no pause. The sequence below is designed so that no phase requires stopping a line, and each one delivers something usable before the next begins.
Phase one: get the data out, change nothing
Build the export path from ERP, MES and historian into a read model with the data structure you wish you had. Nothing operational changes. This phase alone usually reveals that three systems disagree about part numbering, which is information worth having before you commit a budget.
Phase two: put an anti-corruption layer in front of the old systems
Every new service talks to legacy systems through one translation layer rather than reaching in directly. Microsoft's architecture guidance describes the anti-corruption layer pattern as isolating a new subsystem from a legacy one by translating between the two models, and in manufacturing it is where the accumulated knowledge about unit conversions and work centre mapping finally gets written down.
Phase three: replace the edges
Shop-floor clients, the quality module and the worst spreadsheets are replaced with new applications built against the read model and the translation layer. These are the changes operators notice, which makes them the changes that earn goodwill for the rest of the programme.
Phase four: move a capability, not a system
Now, and only now, consider moving something the ERP currently owns: production scheduling, maintenance planning, supplier quality. One capability, run in parallel against the old one until the numbers match, then switched per plant or per line. Phased ERP delivery explains how module-by-module go-live is sequenced.
Phase five: decide about the core with evidence
By this point most of the value has been delivered and the ERP is doing less. Often the honest conclusion is that replacing it is no longer worth the money, which is a good outcome, not a failed plan.
What actually breaks in a manufacturing rewrite
When a big-bang replacement fails in a factory, the cause is usually on this list rather than in the code.
- Master data. Part numbers, BOM revisions and units of measure that never fully agreed are now required to agree on a single date.
- Traceability history. Lot and batch genealogy must survive migration intact because a recall depends on it, and partial migration is not an option.
- Costing continuity. Standard cost, variance and WIP valuation computed differently by the new system means a finance team that cannot close the month.
- Equipment interfaces. Every machine integration is bespoke, and the vendor who wrote the original driver may no longer exist.
- Operator retraining at scale. Three shifts across multiple plants, often in several languages, with production targets unchanged during the transition.
- Customer-mandated systems. An automotive or aerospace customer may require specific EDI formats or quality reporting that the new system does not support on day one.
- The peak that does not move. Order books and shipping commitments do not pause for a go-live weekend.
Why enterprise software implementations fail on adoption covers the human half of that list, which is the half that decides the outcome.
Where AI connects in the first phase
This is the strongest argument for modernising rather than replacing. Once the phase-one read model exists, AI does not wait for the platform. Maintenance manuals and standard operating procedures become searchable in the languages your operators use. Supplier invoices and goods receipt notes can be matched automatically with an exception queue. Production and scrap data can be queried in plain language without a report request. None of that requires the ERP to change, and all of it is deliverable in the first eight to sixteen weeks.
The pattern and its limits are described in Embedding AI into legacy systems without a rewrite, and the specific ERP automations that remove real work are set out in AI in ERP. What the new layer must log for auditors is covered in Compliance and data rules for AI in manufacturing.
When replacement is genuinely right
Three cases justify replacing a core manufacturing system. The vendor has ended support and there is no upgrade path, which makes continuing to run it a security and insurance problem. The data model cannot represent the business you now run, which shows up as whole product lines managed outside the system. Or the annual licence and hosting cost has grown to approach the cost of a rebuild, which happens with older per-seat industrial suites.
There is also an honest argument against phasing. Running two systems in parallel has a real carrying cost: duplicate reconciliation, two on-call rotas and a team split across both. If your estate is small, a single plant with one system and few integrations, a clean replacement over a quiet fortnight can be cheaper than eighteen months of careful phasing. Incremental modernisation is risk management, and where the risk is low it is overhead.
Cost, timeline and what we would quote
A Legacy-to-AI Modernization programme starts at $31,500 or ₹22.4 lakh and runs to $105,000 or ₹72 lakh and beyond depending on how many capabilities move. Where the work is mostly building new operational software around a retained core, custom ERP and CRM development starts at $28,000 or ₹18.4 lakh. Before either, a ten-day Sprint Zero at $3,250 or ₹2,00,000, credited to the build, produces the component verdict table for your estate and the phasing order. Starting prices for every programme are on the pricing page.
Most first phases reach a live capability in eight to sixteen weeks. Afterwards a Care Plan from $1,000 or ₹68,000 per month covers patching, monitoring and the on-call cover a two-system estate needs while both are running. Manufacturing work reaches us through modernisation and custom ERP rather than a dedicated practice page; the sectors we publish are on the industries hub.
A worked example
The closest published parallel is a fifteen-year-old institutional ERP we modernised without a rewrite, described in the legacy ERP modernization case study. The estate had the same shape as a mid-size manufacturer: a core nobody wanted to touch, a ring of departmental spreadsheets doing real work, and an academic calendar that functioned exactly like a production schedule in refusing to pause. We built the read model first, replaced the edges, and left the core in place.
Related reading
Custom ERP for manufacturing: modules that matter covers which modules justify custom work, The real cost of a custom CRM or ERP sets out the budget, and What AI costs in manufacturing prices the AI layer on top. If you want a component verdict for your own estate, start on the contact page.
The question in a factory is never whether the old system deserves replacing; it is which shift you are willing to lose if the replacement goes wrong.
Frequently asked questions
Should a manufacturer replace its ERP or modernise around it?
▾
Modernise around it in most cases. An ERP holds master data and statutory records, so replacement means migrating years of costing and traceability history. Wrapping it with an export path and a translation layer delivers value in weeks, while replacement is a multi-year programme with a concentrated cutover risk.
Can AI be added to a manufacturing estate before modernisation finishes?
▾
Yes, and it should be. Once an export path and a read model exist, SOP search, supplier document matching and plain-language production reporting can all go live in eight to sixteen weeks. None of them require the ERP, MES or historian to change, which keeps the risk contained.
What does manufacturing legacy modernisation cost?
▾
Eazyware's Legacy-to-AI Modernization programme starts at $31,500 or ₹22.4 lakh, and custom ERP and CRM development at $28,000 or ₹18.4 lakh. A ten-day Sprint Zero at $3,250 or ₹2,00,000, credited to the build, produces the component verdict and phasing order before you commit to a programme.