Introduction
Where AI in Manufacturing Stands
Data Quality Before Dashboards
Why Decision Speed Matters Most
Digital Twins: Expectation Versus Reality
Preparing the Workforce Alongside Technology
How AI and Experts Collaborate
Predictive Maintenance and AI-Supported Safety
Cybersecurity Risks in AI Adoption
Key Takeaways
We spent the day at the Manufacturing 5.0 Conference & Awards, hosted by International Business Conferences (IBC). Senior leaders from manufacturing, FMCG, government freight infrastructure, and large-scale plant operations discussed what is actually working with AI in their factories, and what remains a future plan rather than a present reality. A recurring theme was where the next few years of risk is likely to come from as India’s manufacturing sector continues to adopt these technologies.
Several points from the day stood out to us. We have summarized them below, reflecting upon where AI in manufacturing genuinely stands today, rather than where it is often described as standing.
Where AI in Manufacturing Actually Stands Today
Looking past the general enthusiasm around the subject, the picture that emerged through the day was fairly consistent. India’s manufacturing sector, which contributes close to 17% of the country’s GDP and continues to grow, is somewhere between early trials and full-scale adoption of AI. Almost every leader present described at least one AI deployment already running in production, but very few described a factory that was fully run by AI in any complete sense. The overall pattern looked something like this:
| Common description | What is actually in place |
|---|---|
| “Digital twin running the plant” | A 3D visualization layered on top of operations, used mainly for presentations |
| “AI making decisions” | AI surfacing information; people still making the final decision |
| “Full predictive maintenance” | Automatic maintenance alerts limited to specific, high-value equipment |
| “Dashboards as intelligence” | Dashboards showing what already happened; forecasting still has to be built separately |
| “Factory systems are already secure” | These systems were mainly protected by not being connected to the internet, which is no longer the case |
AI in Manufacturing Starts with Data Quality, Not Dashboards
One paint manufacturing leader opened with the numbers behind his challenge — roughly 350 raw materials, six plants across six regions, more than 300 products, and over 600 stock-keeping units. Consistency was the hard part that came with that scale.
A customer in Chennai expects the same paint quality as a customer in Delhi, even when raw material supply is disrupted, and with that many inputs and locations feeding into production, quality could easily drift from plant to plant and batch to batch.
The shift his team made:
| Before | After |
|---|---|
| A dashboard that reported quality after a batch was already produced | A system built directly into the process that monitors raw material quality at every stage |
| Deviations discovered once the finished product failed a check | Equipment speed and settings adjust automatically the moment a deviation shows up |
| Maintenance on a fixed monthly schedule, regardless of actual equipment condition | Maintenance requests generated automatically the instant a machine falls out of calibration |
The two systems run on the same underlying idea: a machine drifting toward failure gets flagged with the same immediacy as a batch of paint drifting out of spec. Neither one waits for a scheduled check-in to catch the problem.
Why Decision Speed Matters More Than the Technology Itself
An FMCG leader highlighted that AI-driven automation for managing spare parts inventory—recalculating reorder levels and vendor selection—demonstrates that the true value of manufacturing AI lies in faster, less expensive decisions rather than the technology itself. With Deloitte estimating unplanned downtime costs manufacturers $50 billion annually, leaders are investing in decision speed to prevent costly operational delays.

Because this massive cost concentrates at the moment decisions arrive too late, effective solutions must target decision points directly rather than merely improving visibility, a principle that also drives in-plant logistics automation for gate, bay, and inventory management. Ultimately, focusing on faster decision execution, rather than just visibility, is critical to reducing operational losses. You can read the full analysis in the original text.
Digital Twin Technology in Manufacturing: Expectation Versus Current Reality
This part of the discussion was notably direct.
Asked whether digital twin technology is a practical tool today or still mostly a concept, one leader from a government freight infrastructure organization gave a candid answer: what most companies currently call a digital twin is, in practice, a 3D visualization. A true digital twin, one that maps each machine’s usage patterns and cycle time, simulates wear over time, predicts failure points, and continually updates itself using live sensor data, requires maintaining two connected systems at once. Most organizations have not reached that stage yet.
This was not meant as a dismissal of the concept, but as a distinction between how digital twin technology is often described and how it is currently implemented. For anyone evaluating a vendor’s claims about digital twin capability, it is a reasonable question to ask which of the two versions is actually being offered.
The structural difference between the two is easier to see than to explain:

Preparing the Workforce Alongside the Technology
A second theme ran through much of the discussion: in many organizations, the technology is currently ahead of the people expected to use it, and several leaders acknowledged this directly.
One site head, asked what skills the next generation of engineers will need five years from now, did not begin with technical or programming skills. His list, in order, was as follows:
- Problem-solving fundamentals: AI cannot replace human engineering judgment; you still need to understand why a machine behaves the way it does.
- Data literacy: Factories generate massive volumes of data, making data interpretation—rather than just collection—the essential skill.
- Applied AI understanding: You don’t need to build AI models. You just need to know which tool fits the problem and how it drives business outcomes like efficiency or cost.
- Cross-department collaboration: Isolated engineering functions are losing impact; meaningful improvements happen where departments intersect.
- Ongoing learning: With tech cycles lasting roughly 15 months, the ultimate competitive advantage belongs to those who learn to use new tools quickly and responsibly.
His closing observation was direct: five years from now, if the same question is put to five different engineers, the answers are likely to be similar, because the available tools will be much the same across companies. What will differ is how quickly each organization’s people learned to use those tools effectively.
How AI and Experienced Staff Are Expected to Work Together
A plant head overseeing a large operation addressed a common concern directly: will AI eventually replace experienced staff, including operators with decades of experience? His answer was no.
This was presented not simply as reassurance, but as an operating principle: AI is not intended to act as the decision-maker. It is intended to support the decision. The experienced operator continues to make the final call.
His practical observation on adoption was the most useful part of the discussion: an experienced operator will not trust a new system simply because a sensor produces a reading. Gaining that trust requires involving experienced staff from the early design stage, and allowing them to observe the system’s recommendations prove accurate over several cycles. Confidence in the system is built gradually, through repeated and consistent results, rather than through a single demonstration.
The practical consequence for anyone planning a rollout is that this trust-building period needs to be scheduled and budgeted as its own phase, with its own timeline, rather than treated as a delay to work around. A system that is technically ready but has skipped this stage tends to get switched off or ignored by the people meant to rely on it, regardless of how accurate its recommendations turn out to be. The adoption curve, in other words, is set by the pace of trust, not the pace of installation.
Predictive Maintenance and AI-Supported Safety in Manufacturing
This part of the discussion was particularly relevant to the work we do.
A recurring point across several speakers: most organizations already have records of lost-time incidents, but this data is typically unstructured and inconsistent, which limits its usefulness for prediction. Moving from having incident data to actually preventing incidents using AI generally requires several elements working together:
- Monitoring and behavior analysis focused on near-misses and accountability, not only on past incident records
- AI-enabled cameras and drones performing the kind of continuous observation a human safety team cannot maintain at scale
- A single, unified safety dashboard in place of several separate systems
- Higher-risk areas, such as zones near conveyor belts, monitored through overspeed and harsh-braking alerts rather than manual checks
- Underlying data security, since a camera network is only as reliable as the systems carrying and storing its data
One point worth noting: AI could potentially take on part of the role of a safety auditor, continuously verifying that detection and suppression systems are functioning correctly, rather than discovering gaps only after an incident occurs. This is closely related to the work we do in near-miss detection, behavior-based alerting, and camera-based risk scoring for transport yards and highways rather than factory floors. It was a useful confirmation to hear a paint manufacturer and a freight infrastructure organization describe a similar approach, arrived at independently, in a different industry.
Looking back across everything covered so far, four unrelated parts of a factory, and four unrelated speakers, kept describing the same underlying architecture without ever naming it as one:

Cybersecurity Risks as Manufacturing Adopts More AI
The final part of the discussion addressed a topic that was, unexpectedly, treated as the most serious concern of the day.
The point made was as follows: operational technology, the systems that control physical equipment on a factory floor, was traditionally considered secure mainly because it was not connected to the internet. As AI and connectivity are increasingly integrated into these systems, that assumption no longer holds. The concern is no longer limited to opportunistic hackers; it now includes state-sponsored actors with the resources and patience to pursue long-term objectives, including the ability to interfere with a supply chain at the manufacturing stage itself, before any warning signs appear in standard security monitoring.

Key Takeaways on AI in Manufacturing
For anyone evaluating a new AI investment on their own factory floor, the following questions came up repeatedly throughout the day:
Across every speaker, the pattern was the same: the system needs to prove itself in stages before people are asked to rely on it.
A system that only reports what already happened is different from one that can indicate what is likely to happen next.
AI can recommend a course of action, but people continue to make the decision. Nearly every speaker made this point without being asked.
Data literacy and a working understanding of AI were treated as necessary investments, not optional training sessions.
The discussion on cybersecurity made clear that using AI and understanding the risks that come with using AI are two separate matters, and both need to be addressed.
Across every industry represented at the event, the same underlying discipline applied. That is the central takeaway from the day, and a useful question to return to whenever AI is presented as a solution before the actual problem has been clearly defined.