A Tier 1 automotive supplier with three plants across central Europe finished its 2025 capex review with a slide that had been written four years running.
Eighty-six robots installed across the body-in-white area, cycle time at every cell inside its commissioning envelope, OEE figures the line manager could quote without checking.
Then, on the next slide, a line-level throughput figure that had moved 4% in three years against a planned 22%. The robotics team was performing. The line was not. The audit, when it came, traced the same pattern across all three plants.
Cell-level performance was being measured every shift. The integration software wrapping the cells, written by a small internal IT team in PLC-bridge code, MES write-back routines and a homegrown OEE pipeline, had been carried as overflow work since the first cell went live. None of it appeared on the slide.
That pattern is showing up in plants across automotive, electronics and warehouse logistics through 2026. The hardware is doing what the supplier promised. The integration layer between the cells, the line and the back office is where the ROI is leaking.
The reporting gap
The metric set most plants inherit from cell-level commissioning does not survive contact with line-level scale-up. At the cell, the figures travelling up to the operations director are the ones the OEM has been measuring for thirty years: cycle time, availability, mean time between failures, robots installed against plan.
At the line, the figures that matter are different. End-to-end throughput against takt. Reconciliation of the WMS pick view against the actual material on the floor. MES-to-ERP write-back failure rates.
Manual hours absorbed by line leaders running spreadsheets to bridge what their systems should be telling them. None of these are reported in the same pack as the cell metrics. Most are not reported at all.
The result is a slide deck the operations director and the finance committee can both read confidently, while the integration backlog underneath sits invisible to both. A plant manager who has lived with the gap for four years will describe it accurately. A plant manager who has lived with it for one will tell you the robots are not yet performing.
What the figures actually say
The IFR World Robotics 2025 report puts the global operational stock of industrial robots at around four million units, with China alone running close to two million and accounting for 54% of 2024 annual installations.
Sixty-four percent of industrial robots in the global electronics industry are now installed in China. Growth concentrated in electronics and automotive has filled the world’s plants with hardware. The corresponding integration spend has not kept pace.
Forrester research cited in a UK practitioner guide to integration scoping puts the three-year ROI on well-planned manufacturing integration at 354%, with a typical plant freeing 10 to 15 hours per employee per week of manual reconciliation work once the data layer is fit for purpose.
The same source notes UK SMEs running an average of six or more software tools daily and a UK integration market growing at 12% a year. None of those numbers are remarkable on their own.
The pattern they form, sitting next to a robotics capex slide, is the point. Plants are spending freely on the cells and conservatively on the layer that lets the cells deliver line-level value.
The Department for Science, Innovation and Technology’s 2025 review of UK central-government technology classified 28% of the estate as legacy, and supplier estimates inside large UK manufacturers have tracked broadly the same line.
Industrial estates have an estimate-by-estimate pattern: the older the plant, the higher the share of value depending on integration code nobody on the current team wrote.
IBM’s 2025 Cost of a Data Breach put the average UK financial-services breach at £5.74m, and while the breach figure does not transpose directly onto a manufacturing plant, the underlying point that legacy controls bind to legacy integration code travels well across sectors.
Integration spend looks small as a share of capex. It is large as a share of the ROI the rest of the capex was meant to release.
Why integration sits in the gap
Three honest reasons, none of them anyone’s individual fault. Robot OEMs sell, and are paid against, commissioning. The contract closes when the cell is signed off against its FAT criteria.
The OEM’s field engineers are excellent at what their contract covers. The integration code that sits outside the cell envelope, talking to the MES on one side and the part-traceability database on the other, is somebody else’s contract by the time the FAT is signed.
ERP, MES and WMS vendors sell, and are paid against, the boundary of their own product. SAP support stops at the edge of SAP, Rockwell support at the edge of Rockwell, Manhattan at the edge of Manhattan. The layer between them is custom by definition, and the vendors involved are clear, in writing, about whose problem it is.
The third reason is the one doing most of the damage. The custom layer ends up either under-scoped to a stretched internal IT team running it on top of a service-desk workload, or over-scoped to a generic systems integrator with a preferred technology stack and a margin model preferring a transformation programme to a targeted intervention.
Plants whose integration layer works tend to have someone, named, accountable for that layer alone, working to a written target architecture rather than picking up whatever each vendor field engineer happens to leave behind.
What Good looks like
A controlled approach treats integration as a first-class engineering deliverable, not as the residual after the cells, the MES and the ERP have all been paid. The shape of the work is by now well understood.
A six-to-ten-week paid discovery, run by engineers who will be on the build, produces something a plant director can govern: a fixed-price programme proposal, a defined scope, a target architecture that names the integration layer’s components, a sequenced delivery plan, written cut-over and rollback plans for each cut-in, and a data-lineage and reconciliation approach surviving the parallel-running period.
Red Eagle Tech’s UK practitioner guide to system integration scoping sets the deliverable out at the level of detail a plant CIO can put in front of a finance committee without a translator.
The thing that has changed the financial model in the last two reporting cycles is delivery speed. Senior engineers using current AI tooling against a defined target architecture are now compressing the well-scoped parts of integration work, particularly migration of legacy adapters, generation of regression suites against existing behaviour and drafting of data-lineage documentation, by 40 to 60% on engagements where the scope is tight and test coverage is adequate.
One UK manufacturer’s 400,000-line core integration migration, costed at eighteen months a year ago, ran nine. The acceleration is not in the engineering judgement. It is in the calendar time spent on the parts of the job that used to slow the rest of it down.
The speed change makes the discovery-led approach affordable for plants that have carried this work on the roadmap for three planning rounds, with the fixed price and the risk profile of the scope both coming down as a result.
What kind of supplier runs this work well
The kind of firm running this work well, at a scale a plant board can sign off without losing sleep, is more specific than “a systems integrator”. Three properties travel reliably together. The engineers on the build are named on the proposal and available for the buyer to meet before a contract is signed.
The discovery is a paid, fixed-price piece of work with a defined deliverable, not a pre-sales exercise the supplier writes off against an expected programme. And the firm builds full custom platforms from the ground up as a primary part of its offering, so the recommendation at the end of discovery is not biased by a single product preference.
The work is being done well by bespoke software firms running the senior-plus-AI delivery model, and the better ones are equally comfortable scoping a small targeted intervention or a full custom platform depending on what discovery shows the plant actually needs.
The screening conversation a plant CIO should be willing to have is short. Ask the firm to set out the discovery deliverable, the fixed price, the named team, the target-architecture approach to the integration layer specifically, and the post-discovery decision rights.
A reputable firm will answer all five inside a single call, and where that conversation needs three meetings and a workshop to land the first answer, the firm is showing the buyer something useful about what the engagement will feel like in practice.
The question for the next planning round
The question to put on the next planning-round agenda is narrower than “should the plant invest in better integration”. It is whether there is, in writing, a named owner of the integration layer between the robotic cells and the line-level scheduling system, and whether that named owner is working to a target architecture with a defined data model or to whatever each vendor field engineer leaves behind at the close of their FAT.
Plants whose answer points to a vendor field engineer have an integration layer continuing to absorb ROI between every shift and the next capex slide deck. The fix, once an accountable owner and a written target architecture are in place, tends to land inside a single planning cycle and stops the ROI leak before it gets reported as a capacity problem.
