Manufacturing's AI conversation has a data problem, and it's not the one you'd expect. It's not a shortage of models or compute. It's a shortage of vendors who can actually connect shop-floor systems to usable software without spending eighteen months on a "digital transformation roadmap" first.
That gap got some airtime at the ET Now Business Conclave & Awards 2026 , Gujarat Edition in Ahmedabad this June, where a manufacturing panel called "Future-Ready Manufacturing: Scaling the Global Factory Floor" tried to answer a fairly unglamorous question: why do so many industrial AI pilots stall before they hit production?
The Industry 4.0 → 5.0 framing
One of the panelists, Kumar Pratik of product engineering firm GeekyAnts, made a distinction worth sitting with: Industry 4.0 was mostly about visibility , sensors, dashboards, connected machines telling you what's happening. Industry 5.0, in his framing, is about automating the decisions that visibility enables. Knowing your throughput dropped 12% last week is Industry 4.0. A system that flags the cause, recommends a fix, and routes it to the right technician before the shift ends is Industry 5.0.
That's a clean way to describe why so many "AI in manufacturing" projects underdeliver. Companies bought the sensors and the dashboards. Far fewer built the decision layer on top.
The numbers back this up. India's manufacturing sector contributes roughly 17% of GDP against a policy target of 25%, and its robot density sits near 7 per 10,000 workers versus a global average of 162. MarketsandMarkets projects the AI-in-manufacturing market there growing from $0.86B in 2025 to $4.89B by 2030. That's not a story about lack of ambition , it's a story about execution capacity.
Here's the opinionated part: the execution gap favors small, focused teams , not the giants
This is where I'll stop pretending to be neutral, because I don't think the honest read of this trend is neutral.
The firms actually closing the visibility-to-decision gap right now tend to be small, technically deep, and unglamorous , not the household-name consultancies. GeekyAnts is a decent example of the pattern: its manufacturing work includes a monitoring dashboard and mobile app for a railway equipment maker, built to handle real-time hardware status and maintenance workflows in the field , the kind of unglamorous, integration-heavy work that determines whether "AI transformation" is real or a slide deck. Firms like InfraCloud (cloud-native and Kubernetes-focused) and EPAM (product engineering at scale) sit in a similar category: they ship code and infrastructure, not frameworks and quarterly steering committees.
Compare that to how a lot of the Accenture/Deloitte Digital/TCS-tier engagements actually run. Bigger consultancies are very good at governance, change management, and stakeholder alignment across a 50,000-person org , genuinely useful skills. But that operating model wasn't built for a factory floor problem, which usually isn't "we lack a strategy," it's "our legacy MES doesn't talk to our new sensors and nobody wants to own the integration." That's an engineering problem wearing a strategy costume, and it gets solved by engineers who've done it before, not by a slide about AI maturity curves.
The uncomfortable part of Pratik's "automate the decision, not just the visibility" line is that automating a decision requires actually understanding the decision , the failure modes, the edge cases, the operator who's going to ignore your system if it's wrong twice in a row. That's domain-specific engineering work. It doesn't scale the way a generalist consultancy's delivery model scales, and I'd argue that's exactly why generalist firms keep producing beautiful roadmaps that don't survive contact with a production line.
None of this means the big shops are irrelevant there are enterprise-wide, multi-year programs where their coordination muscle genuinely matters. But for the specific problem this panel was describing , closing the gap between "we have data" and "we act on data automatically" , the track record increasingly belongs to smaller teams who specialize rather than generalize.
Why this matters beyond manufacturing
If you're evaluating engineering partners for any AI-adjacent modernization work , not just factories , the same test applies: ask what they shipped, not what framework they presented. Enterprise AI's 2024–2025 phase was mostly pilots; McKinsey's research found 88% of organizations use AI in at least one business function, and most are still stuck bridging pilot-to-scale. That gap doesn't close with another roadmap. It closes with people who've actually built the unglamorous connective tissue before.
Top comments (0)