What Custom Software Development for Manufacturers Really Looks Like: Phases, Risks and Success Metrics

Custom software development for a manufacturer is a staged project that connects the machines (PLCs, SCADA, sensors) with the people and systems that use their data. A small project, such as data capture from one machine with a dashboard, takes about 3 to 5 months. A multi-line MES module with ERP integration takes 9 to 18 months (both are our rough estimates, not published figures). Only 29% of software projects are delivered on time, within budget and with the agreed functionality, according to the CHAOS Report by the Standish Group (2024), and plants with 2003-era PLCs, proprietary protocols and OT teams that put availability first make it harder. Below: the phases, the seven decision gates where projects should stop or go, and the costs that quotes leave out.

Why do industrial software projects fail?

Technology is rarely the root cause: industrial software projects die from requirements, scope and communication problems.

The Standish Group (2024) reports that 29% of projects are successful, 52% arrive late, over budget or with reduced functionality, and 19% are cancelled. Technology failure ranks last among the causes it identifies. A study by J.L. Partners (2024) of 481 software engineers found that projects with documented requirements before development began were 97% more likely to succeed.

On method the sources disagree. The Standish Group reports a 64% success rate for Agile projects against 49% for Waterfall, while the J.L. Partners result favours specifying first. In an OT environment, where an unplanned patch can halt a production line, that tension has physical consequences.

Aggregators such as Gitnux and Raftlabs publish sector figures for scope creep and legacy-integration problems. We could not trace them to a primary study, so we do not quote them here.

The clearest description of where the work lives comes from Globental (2026): “A factory with an ERP and a digitised plant still has the same gap: between what the administrative system records and what actually happens on the line, bridged by paper forms, spreadsheets and someone typing at the end of a shift. That gap is where custom industrial software lives. It doesn’t replace the ERP or the machine controller. It connects them.”

How do you capture requirements when OT and IT speak different languages?

Run separate sessions first. The process engineer, the IT manager and the plant manager often describe the same problem in three vocabularies, and nobody notices until the system is half-built.

ISC2 (2024) puts it this way: IT professionals are “unfamiliar with OT realities such as PLCs, safety systems, vendor maintenance models and downtime constraints”, while OT engineers have “limited exposure to identity risks, threat modelling or modern governance expectations”. BCG (2024) estimates the value of IT/OT convergence at approximately $60 billion globally.

A three-session format we suggest (engineering judgement, not a published standard):

1. OT/process only. Output: a process flow with the required data points, tolerances, alarm conditions and non-negotiable quality criteria. The language is P&ID, ladder logic and OEE.
2. IT/systems only. Output: the data model, API surface, security requirements and integration architecture. The language is endpoints, latency and user roles.
3. Joint session. Present both outputs and build a conflict register, with each resolution documented and signed by both sides. It prevents scope disputes six months later.

For a real example of what the output of such work looks like, our inspection software for Enusa imports files from drone and measurement-equipment exports and produces the charts, tables and 3D map on its own. Preparing a report went from about 2 working days of an engineer’s time to about 30 minutes, with zero errors in the data-processing stage.

Which PLC and SCADA protocols should you budget for?

List every protocol your machines speak before writing code. Discovering one mid-development is the most frequent cause of overruns in industrial projects.

FlowFuse (March 2026) documents that Modbus still dominates at the field-device level, even on new equipment installed as a deliberate first choice. OPC UA is growing, but mostly in SCADA systems, MES integrations and IT/OT gateways, not in field-level PLCs. A typical installation speaks Modbus RTU, Modbus TCP, Profibus DP, HART, EtherNet/IP, classic OPC DA and proprietary dialects. PLC Construction (2026) confirms that Modbus RTU, DH+, Profibus and DF1 are the most common in SCADA integration projects, and all need translation to OPC-UA to feed modern data layers.

OPC-UA gateway options for common PLC families

Engineering estimates, not published rates and not taken from Maedcore project data. Treat as indicative.

Watch for the 10,000-tag problem: a protocol driver that works in the lab with 50 tags can degrade or fail under production load with 10,000 active tags, especially on a noisy field network. Load-test the driver against the real tag count before approving the integration.

Regulated plants add another constraint: traceability. In our radiation inspection software for Enusa Industrias, all measurement data is recorded and retrievable for regulatory compliance, and the report includes more information and analysis than the manual version it replaced.

What is the 7-gate framework for industrial custom software?

We found no decision framework in the literature calibrated for industrial environments, so this is ours: seven points where a project should pause, be validated or be cancelled before the next phase spends budget on an unstable foundation. Each gate has exit criteria and the failure mode it prevents.

Gate 0: Process maturity (before any requirements work)

Exit criteria: The current process is documented, stable and measurable. Not “we’ll fix it in the software”.

Prevents: Automating an immature process. The team then blames the software for a problem that belonged to the process.

Gate 1: Asset and protocol inventory (before locking requirements)

Exit criteria: Every data source listed with make/model, firmware version, protocol, update rate and whether the manufacturer permits third-party read access.

Prevents: Finding mid-project that a 2003 PLC speaks a proprietary dialect over a serial link. Budget for at least one unexpected protocol.

Gate 2: Trilateral requirements sign-off (before architecture design)

Exit criteria: Requirements signed by the process/OT engineer, the IT/systems manager and the plant/operations manager, with the conflict register showing how discrepancies were resolved.

Prevents: Unclear requirements and poor communication, which the Standish Group (2024) identifies as primary causes of failure. A document signed by all three profiles is the most effective prevention and the one most often skipped.

Gate 3: Integration proof of concept (before full development sprints)

Exit criteria: At least one live reading from the most complex or legacy protocol, delivered to the target data layer, with latency and data-quality metrics documented and load-tested at the real tag count.

Prevents: The 10,000-tag problem, and learning about a protocol incompatibility after three months of development. Reference cost: gateway hardware €500 to €3,000 per unit; 40 to 120 hours of engineering per protocol family (engineering judgement).

Gate 4: Scope freeze and change control (at the start of the first sprint)

Exit criteria: A formal change process is active. Any new requirement needs a written impact assessment (time, cost, risk) before it is accepted.

Prevents: Scope creep. In plants, the typical trigger is process engineers discovering data points they “need” as soon as they see the system working. That is understandable, but it must go through the process.

Gate 5: Staged validation in the OT environment (before production deployment)

Exit criteria: A parallel run is completed (the new system runs beside the existing process and outputs are compared) with a documented delta below the agreed threshold. Cutover is scheduled in a planned maintenance shutdown.

Prevents: Post-launch correction, which according to CISQ (cited by Capgemini, 2025) can cost up to 30 times more than catching the defect in design. The IEC 62443-2-1:2024 revision updates OT asset-owner security programme requirements for the first time since 2010. Validation plans should follow its priority order: safety first, availability second, integrity third, the inverse of most IT frameworks.

Gate 6: Change management and adoption (before go-live sign-off)

Exit criteria: All shift operators trained and assessed; SOPs updated; a named “system owner” on the client side; a support plan for the first 30 days.

Prevents: Shelfware. As Artefact (2025) puts it, new tools “must be accompanied by change management and process adjustments; without adequate training and willingness from operational teams, new products become ineffective”. For a 40-operator-per-shift plant, training and SOP rewriting can add 15 to 25% to project cost (engineering estimate, no primary source).

What does custom industrial software really cost?

Beyond the build price, three items rarely appear in quotes.

Legacy protocol support. Each additional proprietary protocol adds 40 to 120 hours of engineering and €500 to €3,000 in gateway hardware. Gate 1 exists to find them before the quote is signed.

Fragile protocol adapters. They can break when the manufacturer updates PLC firmware without consulting the software team.

Regulatory compliance. In industrial contexts this means IEC 62443, NIS2 (which names IEC 62443 as a recognised OT security standard), and ISO 13849/IEC 62061 for machinery with safety functions. These enter the architecture at Gate 2, not during validation.

Indicative cost ranges and annual maintenance

Engineering judgement based on market observation; no primary source and not Maedcore project data.

Conventional software maintenance is typically estimated at 15 to 20% of initial development cost per year. With legacy protocol support we put it at 20 to 30% (engineering judgement), because protocol adapters need rework and the knowledge is scarce. Leaving this line out of the business case is the most frequent way a profitable project on paper becomes an unplanned recurring cost.

How do you measure success in an industrial software project?

Classic software metrics are not enough. The ones a plant manager cares about:

• OEE before and after. If the software links machine data to operational decisions, OEE should move within the first 90 days of real use. If it doesn’t, the project is not delivering operational value.
• Report cycle time. How long did the shift report take before, and now? Easy to measure and visible to operators. In the Enusa case above, it went from about 2 working days to about 30 minutes.
• Per-shift adoption. Below 70% of operators using the system 60 days after go-live is a warning sign of failed change management (rule of thumb).
• Integration incidents per month. A rising number means accumulating technical debt in the protocol adapters.
• Post-launch correction cost vs. maintenance budget. If corrections exceed 30% of the annual maintenance budget in year one, Gate 5 missed quality issues (rule of thumb).

Related reading

• Custom Software for Manufacturing: the five-filter decision on whether to build, buy, or leave it alone.
• Industrial Process Automation: how to pick the first process to automate, and which ones to leave manual.
• Radiation Inspection Software for Enusa Industrias: Maedcore
• Maedcore: Digital Solutions: what we build in this area.

Sources

• CHAOS Report on IT Project Outcomes — OpenCommons (Standish Group data synthesis), 2025
• 268% Higher Failure Rates for Agile Software Projects, Study Finds — Engprax / J.L. Partners, 2024
• Custom Software Development Industry Statistics: Market Data Report 2026 — Gitnux, 2026
• Custom Software Development Statistics 2026: 40+ Data Points — Raftlabs (citing Standish Group CHAOS Report 2024), 2026
• Next-Generation Industrial Software Development: Applied AI — Capgemini Spain (citing CISQ data), 2025
• Custom Software for Industry: 2026 Guide — Globental, 2026
• Why OPC UA Is Not Replacing Modbus (Yet) — FlowFuse, March 2026
• Common SCADA Integration Challenges and How to Solve Them — PLC Construction, 2026
• Converging IT and OT Will Boost Industrial Tech Value — Boston Consulting Group, 2024
• IT/OT Convergence — ISC2, August 2024
• IT/OT Convergence: The Path to Industrial Excellence — Artefact, 2025
• IEC 62443 Compliance Guide (ICS/OT) — Breach Craft (citing IEC 62443-2-1:2024), 2026
• SME Digitalisation in Spain 2026: Statistics by Sector and Size — 10Code / ONTSI, INE, Eurostat, 2026
• Industrial Digitalisation: Tax Incentives and Support Programmes — Tecnocim Innova (citing INE 2024, Eurostat 2024), 2026
• Radiation Inspection Software for Enusa Industrias — Maedcore