Walk into a tier-1 components plant outside Pune at 11 AM and you will see the same scene we have seen in plants across Maharashtra, Tamil Nadu, and Gujarat. CNC machines from the last five to ten years, equipped with sensor packages that emit hundreds of signals a second. SCADA dashboards mounted above the lines, displaying OEE figures and cycle counts. A production engineer with a laptop running Excel. A quality engineer at a bench, magnifier in one hand, micrometer in the other, inspecting a casting.
The casting was made by a machine that knows, to two decimal places, the temperature, pressure, and dwell time at every step of its forming. The machine knows, in fact, every parameter that meaningfully predicts whether the casting will fail inspection. The quality engineer’s magnifier knows none of this. She is doing, by hand, a job the machine could partially automate, because nobody has built the bridge between the data the machine generates and the workflow she runs.
This is the gap. It is not a technology gap. It is not a capital gap. It is an analytical-workflow gap, and it is wider in Indian manufacturing than in almost any other industrialized country we have worked with.
What the data is worth, in numbers
The cost of the gap is easier to feel when you put rupees on it.
A medium-tonnage forging press — say a 1600-ton mechanical press making automotive components — produces revenue of roughly ₹40,000 to ₹1,20,000 per hour when running, depending on product mix, contract terms, and material costs. When it is down unplanned, the loss is the contribution margin on that hour, plus penalty exposure if a customer schedule slips, plus the cost of the breakdown itself.
Most plants we audit run those presses at 60–75% OEE. The number industry consultants quote as achievable is 85% or higher. The 10–25 percentage points of OEE that nobody is recovering, on a press worth ₹6–8 crore, is real money. For a single press, going from 65% to 78% OEE — a believable target with the kind of analytical work we are about to describe — is on the order of ₹2–4 crore a year of recovered output.
PPM — parts per million rejected — is the other big number. A tier-1 supplier shipping safety-critical components is typically expected to be at 25–50 PPM externally. Internally, before final inspection, rejection rates of 1000–3000 PPM are common, and the rework, scrap, and labour that goes into pulling the rejects out before they ship is rarely costed properly. For a mid-size component supplier doing ₹150–300 crore of revenue, internal rejection that is 1000 PPM higher than it needs to be costs in the range of ₹1.5–3 crore a year, conservatively.
These are not exotic numbers. They are visible to anyone who walks the floor with a calculator. They are also the numbers that almost never enter the analytical work the plant does, because the analytical work is happening in Excel and the data is sitting in OPC servers nobody at the analyst’s pay grade has access to.
Why the gap persists
Three reasons, all structural.
// the data is in places the analysts cannot reach
The PLC controllers feed an OPC server. The OPC server is, sometimes, scraped to a Wonderware or AVEVA PI Historian. The Historian is, sometimes, exposed to a corporate BI tool. By the time the data reaches anyone with the time and skill to model it, it has been aggregated to fifteen-minute averages, the high-frequency signals that actually predict things have been stripped out, and the analyst is back to looking at hourly OEE numbers in Power BI.
The fix is not exotic. It is a small engineering project — half a person for a quarter, basically — to set up a parallel pipeline that pulls the high-frequency data into a database an analyst can query. We have done this for plants where it took six weeks and produced more analytical capability than the previous five years of MES rollouts. But it is not the kind of project an MES vendor will pitch, because it does not extend their footprint, and not the kind of project the in-house team will do, because nobody has been given the budget or the charter.
// the talent gap is real, and the corporate structure makes it worse
A production engineer in a tier-1 plant earns ₹8–12 lakh a year. A data engineer who can build an analytical pipeline of the kind we are describing commands ₹25–40 lakh in Indian metros. The plants that have these people are the exception. The plants that have them often have them as a single individual buried under one of the corporate IT functions, with no authority to touch production data and no relationship with the operations leadership who would benefit from their work.
Our consulting engagements are often about bridging this exact gap. We come in for eight to twelve weeks, do the analytical work that the in-house team cannot do, and produce something that the in-house team can maintain — a model that runs nightly, a dashboard that the production manager actually checks, a quality-prediction service that fires alerts. We are not trying to replace the in-house talent. We are trying to demonstrate what that talent could be doing, so the corporate budget for it gets approved.
// the industry 4.0 vendor pitch is the wrong shape
A multinational platform vendor will quote ₹3–5 crore for a “smart factory” rollout that promises everything from predictive maintenance to AR-assisted inspection. The actual high-leverage work in most of these plants is ₹30 lakh to ₹1 crore, scoped to one machine or one defect mode, with results visible in three to six months. The platform pitch is too big to approve quickly and too generic to apply to the specific problem; the right-sized pitch is rare because the right-sized vendor is rare.
What the work looks like, concretely
Four examples of the kind of analytical project we have seen pay back inside a year, drawn from work in this sector.
CNC tool-wear monitoring. Vibration sensors plus spindle current plus the cycle parameters; train a model that predicts the remaining useful life of an end mill or insert. Replace the rule-of-thumb tool change schedule with one that responds to the actual condition of the tool. A single avoided tool crash on a multi-axis machining center, where a broken tool can damage the spindle or the part, is a ₹3–5 lakh event. Plants we have seen avoid one or two of these per quarter once the system is in place, on top of the more boring savings from running tools to actual end-of-life rather than swapping early.
Casting defect prediction. Take the process parameters from each shot — die temperature, fill pressure, holding time, cooling-curve shape — plus a thermal image of the casting taken at a fixed station after demoulding. Train a classifier on the inspection outcomes. Move from manual visual inspection (where defect-detection rates of 70–80% are typical, depending on defect type) to automated triage of the obvious-pass and obvious-fail castings, with human inspection focused on the genuinely ambiguous middle. Quality engineer time stops being spent on castings that anyone could classify; rejection rate at final inspection drops by a multiple.
Press maintenance scheduling. Pull five years of breakdown records, run a survival-analysis model over the components that fail (bearings, clutch packs, hydraulic seals), and produce a maintenance schedule that is risk-weighted instead of calendar-based. Plants typically see 12–18% reductions in unplanned downtime within two quarters, which on a press worth ₹6–8 crore in revenue per year is direct margin improvement.
SPC migration. This one sounds boring and is enormously valuable. Most plants we walk into have SPC charts being maintained on paper, or in Excel templates that are updated once a shift. The control rules are applied informally, the alerts are generated by humans, and the response time when a process drifts is often hours. Move the same SPC to a live database with automated rule firing — Western Electric or Nelson rules, depending on what the QMS calls for — and the response time goes to minutes. The cost is a few weeks of work; the gain is reducing the number of bad parts produced before someone notices the drift.
These are not exotic projects. None of them require deep learning. None of them require a fundamental rethink of the plant. They are well-defined analytical projects, scoped to one machine or one defect mode, with measurable outcomes. They are the projects the in-house team would be doing if they had the time, the data access, and the right pair of hands beside them.
How we engage
When we take on a manufacturing engagement, the first week is mostly listening. We walk the floor with the production manager. We look at the actual SCADA dashboards. We ask the maintenance head what wakes them up at night. We ask the quality manager what they would build if they had a data engineer to work with for a quarter.
Out of that week comes a written assessment: three or four candidate projects, each with a believable business case in rupees, each with a clear definition of done. The plant picks one. We pilot it for six to eight weeks. If it works, it goes into production and we build the next one. If it does not — and roughly one in five does not — we say so plainly, and we do not bill for the next phase.
The gap is not capital. It is analytical workflow. The data exists; the analysis does not.
If you run a manufacturing plant and you have looked at your SCADA dashboards lately and wondered whether the data flowing through them could be doing more for you, that is exactly the question we are good at answering. Write to us with one paragraph about what you have, what you make, and what is keeping you awake. We will tell you within two business days whether we think we can help, or recommend someone better suited if not.