The first is point AI solutions. Narrow solutions are often built to solve isolated tasks in a  single machine and a single defect type. These solutions never connect to the operating system of the plant. The value then stays locked in a proof-of-concept with no path to the broader operation and cannot scale.

The second is that demonstrations succeed in one machine or one line and then stall when you try to extend them across sites and shifts. Nothing was built for scaling these processes, from a multi-vintage fleet of machines to underlying data infrastructure, the model retraining pipelines, and the operator interfaces. What worked in a controlled pilot environment will break when you introduce the full complexity of multiple lines, shifts and real production variability.

The third, automation that isn't connected to MES, ERP and PLM cannot close the loop on real business outcomes. Enterprises need an operating layer that orchestrates the entire stack.

Q: What can be done to mitigate those setbacks that keep them from scaling up?

A: The starting point is leading with business outcomes and not a one-size-fits-all technology solution. Define the future-state vision and work backward to identify which use cases deliver the most economic value. That changes the conversation entirely. Instead of selling a solution, you're now selling a transformation and then backing it up with engineering excellence.

The structural piece that matters most is what we call Value Drops. A consulting engagement should be designed in combination with a series of implementation projects, each one delivering tangible, measurable value to the customer at regular intervals rather than a large investment upfront with value arriving at the end of a long consulting and implementation cycle. In practice this means an 8–12 week pilot that proves a specific business outcome and validates ROI, then refinement and adoption, then productization, then multi-site rollout. Each stage funding and justifying the next. This way, customers can see what they're getting before they commit to what comes after.

The twin-workstream model we used with LMW, a global leader in textile machinery and tools, is a good example of this in practice. The strategy revolved around setting direction i.e. benchmarking competitors, mapping the use case portfolio, defining the roadmap across data, technology and talent, while implementation was already building and deploying in parallel. Value wasn't withheld until the strategy was finished.

Enterprise integration has to be a design principle from the start. Connecting ERP, MES, PLM and edge systems into a single fabric is what allows the AI to actually close the loop with the systems, rather than surface insights no one acts on. Human-at-the-helm governance means transparent, auditable AI where operators stay in control and builds institutional trust.

 Q: Tell us about your work with this textile company in India. What kind of advantages did the company see from the partnership?

A: LMW (Lakshmi Machine Works) is one of the world's leading manufacturers of textile spinning machinery, covering the complete yarn-spinning process from blow room to ring frames. They hold roughly 84% market share in India's textile machinery segment. LMW wanted to solve customers’ problems like wastage rates, unplanned downtime, quality issues and skilled-labor scarcity by integrating machine architecture with actionable intelligence.

What we built focused on three problems that LMW's customers felt most acutely.

The first was yarn quality. Achieving a target yarn quality requires getting the machine settings and the raw material mix exactly right—historically something that depended heavily on the knowledge of experienced operators. We built a machine learning system that predicts yarn quality metrics (CSP, RKM, U%, IPI) from machine settings and raw material inputs and works in reverse: given a target quality, recommend the optimal settings and material combination.

The second was alarm reliability on the draw frame—a critical machine in the spinning process that determines the consistency of the fiber fed into ring frames. The draw frame was generating a high volume of alarms, the majority of which turned out to be nuisance stops like the machine halting for a condition that didn't actually require intervention. Each unnecessary stop meant lost production time and an operator walking across the floor to reset a machine. We deployed an ML model at the edge, integrated with the PLC via OPC-UA, that classifies alarm severity in real time and sends the appropriate command based on what the model predicts. Nuisance machine stops reduced significantly.

The third was the Intelligent Service System. Service engineers in the field were spending hours on average to resolve a machine issue, working largely from tribal knowledge and escalation chains. We built a GenAI-powered RAG (Retrieval-Augmented Generation) system called Yukta, deployed on LMW's AWS cloud, that gives engineers instant access to answers drawn from thousands of historical service tickets, DEMS data, LMW manuals, and WhatsApp service exchanges. Rather than starting from scratch on each ticket, the engineer asks a question and the system retrieves the most relevant past cases and solution paths. This enabled service consistency and improved first-time-fix ratio, lower Mean Time to Repair across the board.

The AI-enabled textile machines were co-launched with us at the global textile industry's flagship trade event, ITMA Asia + CITME in Singapore in 2026.