They Had the Budget. They Had the Ambition. They Almost Wasted Both.
Kavitha had a number in her head for three years.
Not a revenue target. Not a headcount figure. A specific, stubborn number, the percentage of production capacity her textile plant in Coimbatore was leaving on the table every single month. She had calculated it herself, late one night, cross-referencing delivery delays against machine utilization reports. The number was 17%.
Seventeen percent of capacity. Gone. Not because the machines were old. Not because the team was incompetent. Just gone. Somewhere between the weaving floor and the finished goods warehouse, the plant was bleeding time in ways nobody could fully explain or trace.
In 2023, the board gave her the budget to fix it. The instruction was simple: bring this plant into the future.
Kavitha knew exactly what that meant. AI. She had seen what it was doing to factories in Tiruppur. She had read the case studies. She had sat through three industry conferences where the word “smart factory” appeared in every second sentence. She was ready.
She called the vendors.
Three Pitches. Three Promises. One Very Expensive Mistake Waiting to Happen.
The first vendor arrived in a chauffeured car with a team of four and a presentation that opened with aerial footage of an automated factory in Germany. Fully robotic assembly. Computer vision at every station. A control room that looked like mission control at ISRO. The numbers were extraordinary. The price was more extraordinary. Implementation timeline: 22 months minimum.
Kavitha moved on.
The second vendor was local, leaner, and more grounded. They proposed an AI demand forecasting tool align production schedules to order pipelines, reduce excess inventory, improve lead times. Sensible idea. But when Kavitha asked how it integrated with their ERP system heavily customized over a decade, practically a living organism at this point, the vendor’s lead consultant paused for four seconds too long before answering.
She moved on again.
The third vendor felt different. Smaller team. Sharper questions. They proposed a computer vision quality control system for the weaving floor, cameras that could detect fabric defects in real time, flagging rejections before they traveled three steps downstream to finishing. Concrete. Visible. Fast to deploy. The vendor promised measurable results within 90 days.
Kavitha liked this one. She told her assistant to schedule a contract review.
She was six days away from signing when an old industry contact called her with one piece of advice: before you commit, speak to Arjun.
The Man Who Refused to Talk About Technology
Arjun ran a small AI consulting practice out of Chennai. No chauffeured car. No aerial footage. He showed up with a notebook and two questions.
What is actually costing you the most right now? And how do you know?
Kavitha answered the first question confidently. Quality rejections at the weaving stage. Hence the computer vision pitch.
Arjun wrote something down. Then asked her to walk him through a normal production week, not the highlights, the texture. The daily friction. The decisions that happened too slowly. The information that existed somewhere in the plant but never reached the person who needed it at the moment they needed it.
Kavitha talked for two hours.
What surfaced from that conversation had nothing to do with weaving defects.
The Bottleneck Nobody Had Named
Three steps downstream from the weaving floor sat the dyeing unit.
Six weaving lines fed into it. Each line produced batches with different thread counts, different fiber compositions, different order specifications. Getting those batches through the dye vats in the right sequence, minimizing color changeover time, reducing chemical waste, balancing vat availability against order urgency was one of the most complex daily optimization problems in the entire plant.
It was being solved every morning by one man.
Rangan. 24 years on the floor. He arrived at 5:30 AM, drank his tea, looked at the day’s batch list, and constructed a scheduling sequence entirely from memory and instinct. No spreadsheet. No system. No documentation. Just two decades of pattern recognition running quietly behind his eyes.
He was brilliant at it. The floor team called him the calculator not because he used one, but because he never needed to.
He was also 61 years old.
And when Arjun asked Kavitha what happened to dyeing unit scheduling when Rangan took leave, she went quiet for a moment.
Chaos, she said finally. We just manage.
Arjun put his pen down.
There is your 17%, he said.
The Pitch Nobody Would Have Made
Arjun’s recommendation landed like a quiet shock.
Don’t buy the computer vision system. Not yet. Start with Rangan.
Spend eight weeks documenting every scheduling decision he makes. Build a dataset from three years of dyeing records, batch sequences, changeover durations, chemical consumption, vat utilization, output quality. Use that foundation to build an agentic AI scheduling system that encodes his logic, optimizes across every variable simultaneously, and generates a daily plan that any operator on any shift can execute.
The goal: make sure 24 years of irreplaceable expertise doesn’t retire when Rangan does. And make it available to the plant 24 hours a day, not just when he’s standing on the floor.
Kavitha pushed back. It felt smaller than what she had imagined. Less dramatic than computer vision cameras and real-time defect detection. Harder to present to the board as a transformation.
Arjun didn’t flinch. The computer vision system gives you a 4% reduction in weave rejections. The scheduling system gives you 11 to 14% more throughput from the same floor, the same machines, the same team, and the same budget you already have. You tell me which number the board wants to hear in 12 months.
She made the call.
Eight Weeks With Rangan
What followed was the least glamorous phase of the project and the most important one.
An analyst joined Rangan every morning at 5:30. Watched him work. Asked why after every decision. Logged variables nobody had ever written down, the way he prioritized vat availability against color family sequencing, how he factored in downstream finishing timelines, why certain batch combinations were chemically incompatible in ways that weren’t in any manual.
Rangan was suspicious for the first two weeks. By the third week, he was explaining his reasoning unprompted, filling in context the analyst hadn’t thought to ask for.
By week six, he told Arjun: I think you’re trying to put my brain in a computer.
Arjun said: we’re trying to make sure your brain doesn’t disappear when you go home.
Rangan thought about that for a moment. Then said: fine. But I want to check its work every morning.
He still does.
What the Numbers Said
The AI scheduling system went live four months after that first conversation.
It generates an optimized daily dyeing plan at 5 AM every morning before the first shift walks in. Operators can override it. Rangan reviews it daily and adjusts when his instinct says the system has missed something. Every adjustment feeds back into the model.
Six months in: dyeing cycle time down 14%. Chemical waste reduced by 9%. On-time delivery to export partners climbed from 79% to 94%.
The 17% Kavitha had been staring at for three years came down to 6%.
The computer vision system, by the way, is now in phase two planning. On its own timeline, for its own reasons, solving its own problem. Not as a substitute for the harder work. As an addition to it.
What Kavitha Knows Now That She Didn’t Know Then
She still gets vendor pitches. She handles them differently.
Before any vendor gets past their opening slide, she asks one question: have you looked at our operations, or are you showing me something you built for someone else’s problem?
Most of them fumble it.
AI in manufacturing is not a product category. It is a diagnosis followed by a design followed by a deployment in that order, always. The factories getting it right are not the ones with the biggest budgets or the most sophisticated technology. They are the ones that asked the right questions before they bought anything.
Kavitha almost bought the wrong future. A two-hour conversation changed the sequence.
That sequence is everything.
Evvo Technology begins every engagement with your operations, not our catalogue. Because the most expensive AI mistake in manufacturing isn’t a failed deployment, it’s a perfectly executed solution to the wrong problem. Let’s find the right one.
Sometimes, the most important part of sequence is knowing where AI should not be the answer at all. We wrote about exactly that moment in our next piece, read The Consultant Who Told Us Not to Use AI to see what that diagnosis looks like from the other side.
