How AI is Transforming Manufacturing beyond the Factory Floor
Rahul Garg, Founder and CEO, Moglix, and Credlix in an interaction with Mary Janifha Evangeline. X, Editor, India Manufacturing Review shared his perspective on the growing role of AI in manufacturing, explaining why its greatest opportunities may now extend beyond the factory floor. He discussed how AI is reshaping supply chains, procurement, logistics, and demand forecasting while creating measurable business value across operations. He also highlighted AI’s role in enabling faster, more informed strategic decision-making across finance, sales, customer service, and other enterprise functions, offering insights into how manufacturers can move from traditional automation toward more intelligent, data-driven operations and more.
Why do you believe AI's biggest manufacturing opportunity now lies beyond the factory floor rather than inside production facilities?
The framing is right, though I would put one qualification on it. The factory floor is not finished. It is simply further along.
The distinction I find more useful is that a shop floor problem is a local problem. Every plant has its own layout, its own machines, and its own process history. What you build for one line rarely transfers to the next line, and almost never to another company. The gain is real and it stays where it was created.
Outside the factory, many challenges are shared across the industry. Supplier delivery performance, lead times, steel grade price fluctuations, and the actual sourcing cost of a specification are common industry-wide concerns rather than issues limited to a single plant.
That changes what can be built. Inside the plant you build a tool for yourself. Outside it you can build shared reference data that improves as more of the sector uses it, and every organisation reading from it sees further than it could alone.
There is a size argument as well. Sourcing, inventory, logistics and working capital account for 60 to 70 per cent of manufacturing cost. But the compounding is the more interesting part.
How is AI transforming supply chains, procurement, logistics, and demand forecasting to create greater business value than traditional factory automation?
I will share my observations. Factory automation makes an existing process faster. Supply chain AI changes what the decision is.
Take procurement. A buyer historically chose from suppliers they knew, at prices they had last seen, on lead times they assumed. A system drawing on transaction history can suggest an alternate material, flag a supplier whose delivery performance has been slipping for two months, and benchmark a quoted price against what the same item traded at last week. None of that is automation. It is a different quality of decision.
The reason the value tends to be larger than shop floor gains is where it lands. Throughput improvements show up in the profit and loss statement. Supply chain improvements show up on the balance sheet. Inventory released is cash released. A shorter sourcing cycle is working capital freed.
For a manufacturer running on thin margins, that distinction matters more than any efficiency percentage.
What role does AI play in helping manufacturers make faster strategic decisions across finance, sales, customer service, and enterprise operations?
Most decisions in a manufacturing company are not slow because leaders are indecisive. They are slow because the information sits in six systems and three people's heads.
In finance, the most useful AI applications are often unglamorous. Spend visibility across categories and plants can give finance teams a clearer view of where money is being committed and spent. Cash forecasting can become more accurate when models incorporate actual payment behaviour alongside contractual payment terms and historical patterns. The benefit is better visibility into near-term liquidity and working capital, allowing finance teams to make decisions based on current operating data rather than six-week-old numbers. In sales, quoting is the obvious one. An industrial quote involves specification, availability, landed cost and credit exposure. Assembling that manually takes days, by which time the customer has often moved on.
In customer service, most enterprise queries are status questions. Where is my order. Why was it short shipped. Those are answerable from data the company already holds.
I would add a caution. Speed is only useful if the judgment behind it holds. The safer approach should be to let the system prepare the decision and keep a person accountable for taking it. Recommendation first, approval second.
The failure I have seen most often is not a bad recommendation. It is an organisation that gradually stops asking whether the recommendation is right.
Also Read: Can India Lead the World in Intelligent Machine Manufacturing?
How can manufacturers use AI to build more resilient and agile supply chains amid global disruptions, geopolitical risks, and changing customer demand?
The last eighteen months have made this case better than any consultant could. Tariff positions moved. Shipping routes through West Asia became unreliable. Input costs changed for reasons that had nothing to do with demand.
Resilience used to mean holding more inventory. That is an expensive form of insurance, and it protects against the wrong risk.
What helps is optionality. Knowing, before you need it, which alternate suppliers are qualified for a given specification, what their capacity looks like, and what switching would cost in time and money. That is a data problem more than a sourcing problem.
Two things should be in place. Visibility past tier one, because disruption usually begins at tier two or three, where most manufacturers have no view at all. And a live view of qualified alternates rather than a list refreshed once a year.
For instance, on our own platform, supplier onboarding time has come down by roughly 80 per cent. I would be careful about what that means. It does not shorten technical qualification, which has its own timelines and should. It removes the administrative delay sitting in front of qualification, which is often where the weeks are actually lost.
Beyond predictive maintenance and robotics, which AI applications are currently delivering the highest ROI for manufacturing organizations, and why?
Several stand out, and they share a common shape.
Sourcing and negotiation support include price benchmarking, alternate material suggestions, auction design. In our deployments this has delivered procurement cost optimization of 4.5 to 6 per cent, with savings on request for quotation close to 10 per cent. On large industrial spend, that is material.
Demand forecasting and inventory positioning: Modest accuracy gains produce disproportionate inventory reduction, because safety stock is set by forecast error rather than by demand.
Warranty and quality claims analysis: Claims data tells you which supplier, which batch and which design choice is costing money in the field. Most manufacturers hold it and read it as a cost line rather than as a signal.
Document intelligence: Contracts, invoices, specifications, test certificates. Manufacturing runs on documents nobody reads carefully until something goes wrong.
Supplier risk: Most companies still discover a supplier problem when the shipment fails to arrive. The reason these deliver is structural. Each is a bounded decision, taken often, with a measurable outcome soon afterwards. A system improves fastest where the feedback is quick and honest.
Predictive maintenance works for the same reason. It simply got there first, because sensors produced clean data before enterprise systems did.
How are AI-powered digital twins, enterprise analytics, and real-time data intelligence reshaping decision-making outside the production environment?
The interesting shift is that digital twins have moved from modelling a machine to modelling a network.
A supply network twin lets a company ask questions it previously answered by intuition. What happens to landed cost if a duty changes? Which plants stop if a single tier two supplier fails? What a 20 per cent demand shift does to working capital across three quarters? These were quarterly exercises done in spreadsheets by whoever was available. They are becoming continuous, and that changes who asks them.
Real-time intelligence changes the rhythm more than the answer. When a company reviews supplier performance once a year, it manages by exception and by memory. When the same view refreshes daily, the conversation with the supplier changes character. It becomes corrective rather than punitive, and suppliers respond differently.
I would offer one caution. A twin is only as honest as the data beneath it. Material master data in most manufacturing companies is inconsistent across plants. The same item carries four codes and three descriptions. Cleaning that up is tedious and unrewarding, and it is where these programmes quietly succeed or fail.
The simulation is the visible part. The data discipline is the actual work.
Also Read: Unlocking Manufacturing Growth with Innovation
What are the biggest challenges manufacturers’ faces when scaling AI across enterprise functions, and how can they overcome data, talent, and governance barriers?
Most programmes do not fail technically. They fail at the join between the model and the way work actually runs.
On data, the constraint is rarely volume. It is consistency. Item codes, supplier master records, unit conversions, plant level variation. Companies underestimate this and then blame the model.
The integration question follows closely. A manufacturer runs on SAP or Oracle and will not replace it. Anything expecting to sit outside the system of record and still be adopted is optimistic. The work is in reading from and writing back to what is already there.
On talent, the shortage is not data scientists. It is people who understand both the function and the system. A procurement head who can specify what good looks like is worth more to a programme than another engineer.
On governance, the requirement is traceability. If a system recommends a supplier or clears an invoice, someone should be able to see why, and someone should remain accountable for the outcome. Audit trails and clear approval rights are not compliance overhead. They are what makes adoption possible in regulated sectors.
Programs should start where the process is already measured. Ambition without a baseline rarely survives the first review.
How should manufacturing leaders measure the business impact of AI initiatives beyond productivity, including customer experience, sustainability, profitability, and innovation?
The first discipline should be to stop counting adoption. Number of users, number of use cases, number of pilots. None of it says whether anything improved.
Four families of measures have been more useful in our experience.
Cash flow: Inventory days, cash conversion cycle, working capital released. This is where supply chain AI shows up first, and it is a number finance can verify.
Reliability: On time in full delivery, forecast error, order fill rate, compliance against procurement policy. Customers feel these before they feel anything else.
Decision speed: How long a purchase requisition takes to become an order. How long a supplier takes to qualify. How often a decision has to be reversed. That last one is the most revealing and the least tracked.
Sustainability: Emissions per unit shipped, packaging intensity, and the completeness of supplier disclosure data. Most companies cannot report the last one honestly, which is itself a finding worth acting on.
On innovation, the measure I would suggest is time from a design change to a supplier able to deliver it.
One rule should hold across all of them. Savings should be reconciled by finance against booked values. Self-reported AI savings rarely survive an audit.
With generative AI rapidly evolving, what new opportunities do you foresee for product design, supplier collaboration, knowledge management, and workforce enablement?
Product design is where I would look first, though not for the reason usually given. Generative design tools are impressive, but the larger prize is designing for what can actually be sourced. Roughly 70 to 80 per cent of a product's cost is locked at the design stage. If a designer can see availability, lead time and landed cost of a component while choosing it, the decision improves before it becomes expensive to change.
Knowledge management is the most underrated. A great deal of manufacturing know-how sits with people close to retirement. Quality histories, failure patterns, the reason a particular supplier was dropped in 2011. These systems are good at making that searchable in plain language, and the alternative is losing it entirely.
Supplier collaboration benefits more quietly. Specification clarification, technical queries across languages, quotation responses from suppliers with limited digital capability. In India this matters more than it might elsewhere, given how much of the supplier base is small.
Workforce enablement comes down to time to competence. A new buyer has traditionally taken about a year to become genuinely useful.
One caveat. In industrial settings, generated output should be treated as a draft requiring verification, not as an answer.
About Rahul Garg,Founder and CEO, Moglix, and Credlix:
Rahul Garg, Founder and CEO, Moglix, and Credlix is also the Chair of the CII National Unicorn Forum, Chair of the CII Special Task Force on Advanced Manufacturing, and Co-Chair of the CII Manufacturing Council for 2026-27, placing him at the center of India's push to build manufacturing and technology sovereignty for the next decade. He chairs the YPO Delhi Chapter and maintains a close working relationship with DPIIT and the Ministry of Commerce.
