AI Can’t Fix What India’s Factories Can’t Measure
India’s component makers are adding factory robots in growing numbers, but the payoff now depends on skills, process discipline, and production data that managers can actually trust.
News
- TCS Ties Up With Anthropic to Push Claude Into Enterprise AI
- Karnataka Shortlists 256 Deeptech Startups for ELEVATE NxT Finale
- Reliance-Backed Addverb Eyes $100 Million for Robotics Push
- Anthropic Releases Claude Fable 5 and Mythos 5
- Meta to Lease First AI Data Center in India From Reliance
- Musk Says Space AI Data Centers Are Within Reach as SpaceX Nears IPO
Key Takeaways
01
India’s component suppliers are now the country’s fastest-growing buyers of factory robots, but the spending is running ahead of the skills and systems needed to make it pay.
02
On the shop floor, the limiting factor is usually a slow manual station and a production number nobody can fully trust, not the capability of the machines.
03
Before approving the next line, audit where output actually leaks, station by station, and insist the production record come from sensors rather than from what a supervisor writes down.
A plant owner buys an assembly line rated to make 100 parts an hour, switches it on, and watches it turn out fewer. The supplier sends its own engineers, who run the very same machines, and the line hits its number. Nothing mechanical has changed between the two runs. The only thing different is the person standing at the station.
Himanshu Jadhav has seen this many times. He runs the Indian arm of Jendamark, an automation firm that builds assembly lines and the software to run them for carmakers and parts suppliers in India and Europe. It is the part of the automation story the headline numbers tend to miss.
The figures are striking on their own. India installed a record 9,100 industrial robots in 2024, a 7% rise that made it the sixth-largest market, with the automotive sector accounting for 45% of the total (IFR, World Robotics 2025). EY India expects generative AI to increase productivity in vehicle production and assembly by 35%-37% by 2030 (EY India, The AIdea of India, 2025). The sector already accounts for about 7% of GDP. It supports some 32 million jobs, directly and indirectly, according to figures from NITI Aayog cited by the Center for Social and Economic Progress (CSEP), a New Delhi think tank, in a study it published this year.
What robot adoption figures cannot show is whether those machines are actually raising output. A new CSEP study of India’s auto-component industry suggests that, in many cases, they are not. The bottlenecks lie less with the robots than with workforce readiness, weak data systems and a widening gap between firms that can invest in automation and those that cannot.
Automation Is Splitting India’s Parts Makers
Within that 45% automotive share, the real divide is in the supply chain. CSEP’s fieldwork shows that automation is concentrated among Tier-1 suppliers that run high-volume lines and must meet strict quality standards set by global carmakers. These firms use robots for welding, forging, material handling, and inspection, not just to reduce labor dependence but to improve consistency, traceability, safety and cycle times. Smaller MSMEs automate more selectively, usually in response to labor shortages, absenteeism or a specific shop-floor bottleneck.
That national average hides two different industries. CSEP’s researchers spent time inside three factories in the Gurugram-Manesar belt outside Delhi. One was a large Tier-1 supplier with overseas customers, one a mid-sized firm selling to domestic carmakers, and one a small specialist making low-volume, high-variety parts.
The three plants behaved nothing alike. The Tier-1 ran robots on welding, forging and inspection because its export customers demand tight tolerances and full traceability, and automation is the price of staying in their supply chains. The specialist could not automate even where it wanted to. It makes hundreds of different parts in monthly runs of 5,000 to 10,000 units, and the time lost resetting a robot for each short batch eats the savings. The mid-sized firm automated one station at a time, usually to cover a job that is unpleasant or a shift that workers tend to skip.
About 80% of the component firms in the Delhi-Haryana cluster sit in the lower two tiers, CSEP found. That makes the regional challenge clear. Clusters with more large Tier-1 suppliers are better placed to absorb automation, while MSME-heavy clusters risk falling behind unless training, data systems and policy support catch up. So the national robot count tells a chief executive less than it seems to. It shows that firms are buying machines, not whether those machines are delivering higher output.
Research highlight
The core evidence is CSEP’s April 2026 study of automation in India’s auto-component sector. It draws on shop-floor observations and interviews with management and union leaders at three Gurugram-Manesar factories: a Tier-1 exporter, a mid-size MSME, and a small specialist. The study adds data from ACMA, SIAM and the Annual Survey of Industries. Robot figures are from IFR’s World Robotics 2025.
The Slowest Station Is Often the Manual One
Go back to the line that fell short. Its stations are timed to pass work down the line in sequence, so when one station is slow, every station after it waits. The line’s speed is set by its slowest station. Jadhav puts rough numbers to it. An automated welding cell might run at 95% of its designed cycle rate through a shift, and a semi-automatic station at about 90%. A station where someone loads a part by hand, checks it and passes it on tends to manage 70% to 75%. Buying faster robots does nothing to lift that, because the bottleneck sits at the hand station, not the machines.
Those stations lag for ordinary human reasons. The same job is done a little differently by each worker, and a regular is quicker than someone drafted in to cover an absence. Over a shift, those small differences add up to units that were never made. On a European line that is 70% automated, the manual share is small enough to manage. On an Indian line, still close to half-manual, it is the main event.
India’s labor structure makes it worse. Much of the component shop floor runs on contract workers hired through agencies, who move between jobs often and miss shifts more than permanent staff, which is part of why owners turn to robots in the first place, CSEP found. The draft Labour Codes of 2025 and their provision for fixed-term employment might calm that churn. Skill is the harder limit. An automated line does not need stronger workers so much as different ones, able to read a dashboard, follow cycle-time logic and troubleshoot a cell that has halted. India’s Industrial Training Institutes and polytechnics still produce too few of them (CSEP, 2026). A firm can buy the robots and then discover that no one on the floor can keep them fed and running
“The machines haven’t changed and they don’t behave differently for our engineers versus the customer’s operators. The only real variable is the human operator.”
Himanshu Jadhav, CEO, Jendamark India
AI’s First Job Is an Honest Production Number
AI has a real use on a shop floor, though a narrower one than most sales pitches imply. What a plant needs from it is simple to state. While a shift is still running, a manager should be able to see why a line stopped, where defects are slipping through, and whether the day’s reported output matches what the machines actually made.
Jadhav’s firm sells software for this category of work. Its system, Odin, is one of several products on the market and helps illustrate what such tools are designed to do. One module uses cameras to monitor the line and stop it when an operator skips a step, fits the wrong part or works without required safety gear. The more important feature is duller. The system builds its production record from the sensors, the machines and the programmable controllers rather than from anything a person keys in. If someone tries to edit that record, it keeps the original and flags the change to the plant head.
A lot of Indian plants still run on figures a supervisor notes by hand and a report stitched together after the shift, which leaves room for optimism and for worse. When the production log cannot be quietly adjusted, the arguments that happen above the shop floor change, because everyone is working from the same number. Jadhav’s claim for the system is mundane and important. A plant head can ask it why yesterday came up short or who needs retraining, rather than reconstructing it from memory. Until that underlying record is trustworthy, every dashboard built on top of it is guessing.
4 Layers Separate the Plants That Pull Ahead
The firms that get a return on automation tend to build in a set order, and it is an easy order to break. Picture four layers, each resting on the one beneath it. Call it the Shop-Floor Trust Stack.
- Machine reliability. Whether the equipment runs, holds its cycle time and stays up.
- Human consistency. Whether operators are trained, follow the process and vary little from one to the next.
- Data integrity. Whether the production record comes from sensors and logs every change.
- Decision speed. How fast a manager can trace a problem to its cause and act.
The small specialist in the CSEP study, when it can afford automation at all, usually buys the first layer and stops. The result is a partly automated line still run on guesswork, with weak tracking of work and little usable production data. The Tier-1 exporter, by contrast, builds all four layers into one system, helping explain why it keeps pulling away from the MSMEs that supply it. The gap is not only about money. It is also about the three layers that rarely appear on a purchase order, and therefore rarely get built.
Implications by Role
C-Suite. Stop treating a new line as a capital-expenditure item and start treating it as a productivity system. Before signing off on machines or AI, ask for a station-by-station read on where manual variance, missing data and skill shortages are costing output. Make that review a condition of the budget from the next planning cycle.
Functional and plant leaders. Stop treating machine uptime as the headline metric. Run a weekly review that ties stoppages, defects and lost units to the specific stations and shifts that caused them. Find the manual stations capping the line and retrain against them before ordering the next robot.
Boards and governance. Press management to separate what was spent on automation from what it returned, in throughput, quality and safety, not in robots installed. Treat the integrity of AI-generated production data as a control issue, with the seriousness given to the numbers in the accounts.
The Work That Is Left
Come back to that line making fewer parts for its owner than for the engineers who built it. The machines were never the problem. What the owner could not see was which station, which operator and which unrecorded decision was costing the units, and no robot reports that on its own.
The suppliers that pull ahead over the next decade will be the ones that can tell, while a shift is still running and without a fight over whose number is right, exactly where output is leaking. For most suppliers, the robots are already arriving. What remains missing is the less glamorous investment in training, measurement and trustworthy data that decides whether the machines pay for themselves.
MIT Sloan Management Review’s AI Research Forum will make its India debut later this year, bringing together enterprise leaders, researchers, and practitioners to examine how autonomous AI is moving from experimentation to governed deployment at scale. To speak, partner, or attend, register here.
