Walk into almost any manufacturing conference today and you’ll hear the same pitch: smart cameras on the line, predictive maintenance dashboards, talk of autonomous factories just around the corner. Yet according to Rajat Srivastava, Founder, CEO and CPO of Df-OS, most of this activity sits on a foundation that isn’t ready to bear its weight — trusted, connected operational data.
In a wide-ranging conversation for Express Computer’s Factory of the Future series, Srivastava argues that the real divide in Indian manufacturing isn’t between companies that use AI and those that don’t. It’s between factories that have quietly done the unglamorous work of unifying their data, and those that have simply bolted AI tools onto processes that remain as fragmented as ever.
Automation Isn’t Readiness
The AI-ready label gets thrown around loosely, but Srivastava draws a sharp line around what actually qualifies. Genuinely AI-ready factories, he says, don’t begin with technology — they begin with an outcome, like cutting changeover losses or reducing customer returns, and work backward.
“AI-ready factories have done the unglamorous work of making their operational data accessible, reliable and connected to decisions,” Srivastava says. “They don’t start with ‘let’s do AI’; they start with clear outcomes like reducing changeover losses or cutting customer returns.”
On the ground, he explains, this means unifying data across machines, processes and people “instead of leaving ERP, MES and SCADA in separate islands.” It also means investing in data governance — “defining owners, standards and quality checks so that KPIs mean the same thing across lines and plants.” Just as importantly, he adds, these factories “embed AI into workflows: recommendations flow into planning, maintenance, quality approval and shop-floor routines, and impact is measured.”
Contrast that with what he calls the more common pattern — predictive maintenance here, a smart-camera inspection system there. “These tools may generate useful insights, but they rarely influence bigger operational decisions like production schedules, staffing, or supplier planning,” he says. “AI-ready factories connect data and decision-making into one integrated system, while others are simply adding AI tools on top of disconnected processes.”
The Foundation Beneath the Hype
Ask Srivastava about the excitement surrounding AI agents, digital twins and fully autonomous plants, and he doesn’t dismiss the technology — he questions the sequencing. “The excitement around AI agents, digital twins and autonomous factories is understandable,” he says, “but these ideas assume something most plants don’t yet have: clean, contextualised and timely data across machines, processes and people.”
“What we often see is advanced projects built on weak foundations,” he continues. “Digital twins run on incomplete process maps and inconsistent cycle-time data, turning into static visualisations instead of decision engines.” Because data is scattered across separate systems, he adds, “AI agents struggle to understand the relationships between production events, work orders, and human actions.” Even pilots that succeed can stumble later — “AI models often perform well in pilot projects but struggle to scale because data is recorded differently across production lines.”
His prescription is deceptively simple: “standard definitions for key operational elements such as orders, batches, stations, and operators, consistent data capture across the factory, and a shared operational timeline that connects every event and decision.” Until that groundwork exists, he warns, “AI programs risk solving the wrong problems or delivering brittle solutions.” His advice to manufacturers riding the current wave of enthusiasm: “use the excitement around agents and twins to fund the boring but essential work of getting factory data right.”
Why “Real-Time” Dashboards Often Aren’t
Most large and mid-sized Indian manufacturers already run ERP, MES and SCADA systems, along with an array of shop-floor applications. So why does a genuine real-time view of operations remain elusive?
“Indian manufacturers often have the right [systems]: ERP, MES, SCADA, quality systems,” Srivastava says, “yet still lack a true real-time view of operations. The core issue is fragmentation. ERP knows the plan, MES knows execution, SCADA knows machine state, but nothing stitches these together into one live narrative per product, order or line.”
Quality checks, rework decisions and material substitutions, he notes, are “captured manually or uploaded in batches.” The result: “Many dashboards claim to show real-time operations, but they’re often displaying outdated information.” It gets messier still when “the same product [has] different IDs across ERP, MES, and quality systems, making it difficult to track products accurately or measure live OEE.” Add legacy machines that “weren’t built for streaming data or open integration,” and connection projects that are “under-scoped, delayed or treated as one-off IT tasks rather than strategic infrastructure,” and the outcome, he says, is “multiple partial truths instead of a single operational reality.”
Defining “Trusted Data”
If there’s a phrase Srivastava returns to throughout the conversation, it’s “trusted data.” On a factory floor, he says, that means “operators, planners and leadership can treat data as a faithful record of what physically happened, down to the unit, minute and station.” He breaks it into three properties: accuracy, context and consistency.
“Accuracy means quantities, events and timestamps match reality; scrap is recorded when it occurs, not approximated later,” he explains. “Context means each data point is tied to product, process step, material lot, tool and operator, so you can answer who, what, where, when and how for any anomaly. Consistency means KPIs like first-pass yield or cycle time are calculated the same way across lines and plants.”
Getting there is harder than it sounds. “Information is collected from multiple sources — machines, operators, and processes — and is often recorded inconsistently,” he says. Compounding the problem, “teams are typically incentivized to maximize output rather than maintain data quality, [so] gaps and inaccuracies are common.” The consequence: “AI can produce confident recommendations based on incomplete or incorrect information. For manufacturing AI to deliver real value, trustworthy data must come first.”
The Human Layer Machines Can’t See
Industry 4.0 conversations tend to fixate on machine data, but Srivastava is emphatic that the human and process layer matters just as much, particularly in brownfield factories that mix manual and automated stations.
“Machine data tells you what a machine did; process and human-generated data explain why the factory behaved the way it did,” Srivastava says. “In brownfield plants, this becomes crucial.”
Real bottlenecks, he argues, “are often procedural: changeover rules, inspection sequences, manual approvals, material handling routes. None of that lives in PLC tags.” Operators, meanwhile, “reorder tasks, improvise setups or adjust quality thresholds,” and “these micro-decisions frequently dominate cycle time, scrap and customer complaints.” Ignore that layer, he warns, and “in older lines that mix manual and automated stations, ignoring human and process data makes the operation look more stable than it truly is.”
Capturing work-instruction execution, checklists and rework reasons in structured form, he says, “allows AI to see patterns beyond ‘machine down’ or ‘temperature high.’ It reveals systematic process deviations, training gaps and workflow design issues.” Looking forward, he’s blunt about where the edge will come from: “the competitive edge will increasingly come from how well [factories] digitise and analyse the human and process layer, not just the machines.”
When AI Sees What Dashboards Miss
Srivastava points to a case from “a high-mix electronics environment” as proof of what’s possible once data is genuinely connected. “Conventional dashboards showed acceptable first-pass yield and machine uptime. On paper, the plant looked healthy. Yet warranty returns for certain variants were higher than expected.”
“By combining machine data, operator inputs, and traceability records, AI was able to uncover patterns that traditional analysis missed,” he says. Instead of relying on overall averages, “it examined how product variants, operators, shifts, and manual process steps interacted.” What it found: “a manual cleaning task was being performed differently by certain operators on specific shifts, but only for a particular group of products.” Because standard KPIs averaged performance across everyone, “this issue remained hidden.”
The fix was almost anticlimactic. “Once the cleaning process was standardized and consistently monitored, warranty returns dropped significantly, even though overall production metrics showed little change.” As Srivastava sums it up: “when AI has access to connected data from machines, processes, and people, it can reveal hidden operational issues that standard dashboards often overlook.”
Closing the Expectations Gap
Manufacturers frequently expect AI to deliver an immediate productivity jump. “The biggest gap lies between the expectation that AI will ‘automatically boost productivity’ and the reality that AI is a decision-support capability, not a magic switch,” Srivastava says.
“Factories often expect predictive models to deliver savings in weeks, but it takes months to clean, connect and stabilise data pipelines,” he explains. There’s also too much focus on “algorithms, platforms” and too little on “changing behaviour.” AI “may highlight optimal sequencing, maintenance windows or inspection regimes, but if planners, supervisors and operators don’t adjust how they work, nothing improves.”
He also flags the classic pilot-to-scale trap: “a single line pilot, with intense support and curated data, looks impressive. Scaling to multiple lines and plants exposes inconsistent data definitions, process variations and change-management challenges.” His conclusion is pointed: “Manufacturers that treat AI as a continuous capability, iteratively improving data, models and workflows[,] see durable gains. Those treating it as a one-off automation project experience short-lived wins and growing scepticism.”
Traceability as a Competitive Weapon
“Digital traceability is evolving from a compliance checkbox into a strategic advantage,” Srivastava says. In electronics, automotive and FMCG, “granular traceability enables faster and more surgical recalls: instead of pulling entire batches, companies can isolate the affected units and processes, dramatically reducing cost and brand damage.”
It also opens doors to premium markets, since “customers and regulators increasingly demand proof of origin, ethical sourcing and full product genealogy,” and “suppliers able to provide this evidence gain preferred-vendor status and better margins.” Beyond risk management, he sees the same data becoming “fuel for optimization” — feeding “predictive quality, smarter line balancing and supplier performance analytics.” His closing line on the topic: “the ability to say ‘we know exactly what happened to every unit we shipped, and can act within hours’ is now a differentiator.”
Mid-Sized Manufacturers Aren’t Locked Out
With thousands of mid-sized manufacturers operating on limited digital budgets, is the “Factory of the Future” destined to remain a large-enterprise privilege? Srivastava says no. “Building an AI-enabled factory is less about the size of the investment and more about the order of execution,” he says. “For mid-sized Indian manufacturers, a clear roadmap and disciplined implementation matter far more than enterprise-scale spending.”
Rather than attempting a full plant transformation, he recommends mid-sized manufacturers “focus on high-value, narrow problems: reducing complaints for a key product, stabilising output on a critical line or cutting energy losses.” Cloud-based tools “can sit on top of existing ERP/MES, avoiding massive upfront investments,” provided “the priority should be interoperability and data standards rather than monolithic platforms.” He also points to an underrated advantage: “mid-sized manufacturers often have an advantage, they operate across fewer sites, have faster decision-making, and can standardize processes more easily.” As he puts it, “the factories of the future will be built by companies that manage their data and operations well, not simply those with the biggest budgets.”
The Factory Operating System of 2030
Asked to look ahead, Srivastava predicts the end of the dozens-of-disconnected-applications era. “By 2030, it’s unlikely factories will still be managed through dozens of disconnected applications,” he says. “We are moving toward a unified operational intelligence layer; a kind of factory operating system that sits above ERP, MES, SCADA and niche apps.”
This layer, in his vision, “will maintain a live, canonical model of the factory: products, lines, assets, processes and people connected in a single data graph. It will orchestrate workflows across quality, maintenance, planning and changeovers, using AI for recommendations but remaining explainable and overrideable by humans.”
Crucially, he says, “it will provide a common ‘language’ for AI agents, digital twins and human users, so new tools plug into the same operational reality instead of creating fresh silos.” Existing applications won’t disappear, “but they will feel more like services under an overarching operational brain.” His closing line doubles as advice to the industry at large: “Manufacturers that treat this unified intelligence layer as critical infrastructure and not just analytics, will be best positioned to scale AI, respond to disruptions and innovate faster.”