June, 2026

2 min Read

WHEN THE MACHINE IS NOT THE PROBLEM

04 Jun 2026

Drawing from his experience in automotive chassis engineering and insights from an MBA at XLRI, B Sai Krishna highlights why 70–80% of manufacturing AI pilots fail to scale: a profound talent and trust gap. AI models, like a tyre degradation predictor he witnessed, often stall because data scientists lack domain intuition, while field technicians distrust the data. True operational transformation requires "AI-enabled engineers" who understand physical systems, like camber gain and tyre slip, rather than generic data scientists. Success lies in a strategic sequence: Borrow vendor solutions to start, Buy specialist talent to accelerate, and Build internal capabilities to truly own the technology.

I spent four years tuning suspension geometry and analysing tyre wear patterns. Three months into MBA, I’m realising the industry’s most critical misalignment isn’t in the steering column.

During my final year at an automotive OEM, our chassis team was conducting trials for integrating an AI-based tyre degradation prediction module into the vehicle validation cycle. The system read real-time load transfer data, suspension deflection inputs, and road surface classification to predict tyre wear progression across different drive cycles. From a pure engineering standpoint, it was sophisticated work — the kind that makes a chassis validation report genuinely interesting to write.

Eight months after the system went live as a pilot, a former colleague informed me that the tyre technicians conducting durability tests at the proving ground were still manually pulling wheels every 500 kilometres and making their own wear assessments. Not because the model’s predictions were wrong. Because nobody had built any trust between the output on the screen and the person crouching next to the wheel with a depth gauge.

I wrote it off as a last-mile adoption issue and didn’t think about it again. Now, sitting inside XLRI and attending classes on Operations Strategy and Organisational Behaviour, I finally have the complete framework for what I actually witnessed: a talent gap that no technology budget can realign.

And it is everywhere. Studies show 70–80% of manufacturers have piloted AI, yet fewer than one in four scale it into real operations. Automotive was supposed to lead this transformation. In too many places, it is still spinning its wheels.

The Gap That No Wheel Alignment Procedure Can Fix

In chassis development, I learned to read a vehicle dynamically — how suspension kinematics change under cornering loads, what a tyre’s contact patch deformation tells you about road input frequency, and why a specific spring rate change produces an entirely different steering feel depending on the anti-roll bar stiffness at the other axle. That understanding doesn’t come from textbooks. It accumulates across thousands of kilometres of test driving, durability loops, and suspension tuning iterations.

What I couldn’t do was take six months of K&C rig data, combined with proving ground telemetry, and build a predictive model for bush fatigue failure across a vehicle platform. That requires a completely different set of skills — and the industry’s uncomfortable reality is that almost nobody sits at that intersection.

The global AI talent gap stands at 3–4 million professionals. In automotive manufacturing, people who truly bridge engineering knowledge and AI account for less than 10–15% of the workforce. A data scientist can process suspension deflection data at scale, but they won’t inherently know that the rear trailing arm bush on that platform runs hot under high-scrub loading, or that tyre pressure variation across a temperature cycle will corrupt the load cell readings on the K&C rig unless accounted for in preprocessing.

What the industry needs — and I mean this both as a technical observation and honestly as a career insight — is AI-enabled engineers. Not data scientists learning suspension theory from a Wikipedia page. Engineers who understand camber gain curves, tyre slip angles, and chassis compliance can work meaningfully with the models processing it all.

That distinction completely reframes who you hire, how you train, and what you actually expect AI to deliver.

Build, Buy, Borrow — From Someone Who Has Chased Tyre Wear Across Three Continents

The framework reads cleanly on a whiteboard. It looks considerably more complicated when you’ve spent time at Nürburgring durability loops or high-speed oval sessions at Idiada.

Build — developing AI capability within your existing engineering team — is the option that makes immediate intuitive sense to anyone who has done real chassis development. The institutional knowledge inside a senior suspension engineer is extraordinary and almost entirely informal. Why does a specific damper valving setup that works perfectly on Belgian pavé produce shimmy on American highway expansion joints? How a tyre’s self-aligning torque characteristic changes the steering feel calibration target for a given front suspension geometry. Which bushing compliance signature in the data indicates the beginning of fatigue degradation versus normal rubber settling?

No externally hired data scientist arrives with that. And on the test track — much like on the plant floor — credibility is earned through demonstrated understanding of the physical system. A solution built and endorsed by the chassis team gets used. An identically capable solution delivered by an outside team without that context gets quietly set aside after three months.

The honest constraint: meaningful upskilling takes 6–12 months. Companies invest $1,000–$3,000 per employee annually on digital capability programs, and only 30–40% of those programs translate into genuine deployment. For a chassis programme running against a fixed Job One date, that timeline creates real pressure.

Buy — hiring specialist AI talent directly into automotive engineering environments — becomes necessary when the technical gap is simply too wide to bridge internally within a programme timeline. Automated road surface classification using suspension response signatures. Digital twins for virtual suspension tuning across full vehicle load cases. Predictive tyre compound degradation modelling across mixed-surface duty cycles. These require people who have actually built these systems before.

The failure pattern I now recognise immediately from development experience: technically strong AI hires who produce impressive work that doesn’t survive contact with the test track. The model didn’t account for the fact that tyre temperature gradients across the contact patch behave non-linearly under combined cornering and braking loads. Or that the wheel force transducer data from winter testing in Sweden has a systematic offset that the test team compensates for manually and never formally documents. I could have walked any incoming data scientist through a dozen such caveats on our platform within the first week — knowledge that isn’t in any dataset, any test report, or any onboarding document.

Pairing AI hires with experienced chassis engineers from day one isn’t an optional best practice. In suspension development specifically, where physical intuition about system behaviour is half the job, it is the only way bought talent actually delivers.

Borrow — vendor platforms, Tier-1 partner AI solutions, consulting firms — is where most manufacturers begin, and it is a rational starting point. Around 60–70% of initial AI pilots are externally led. For an OEM building internal justification before committing to transformation at scale, a proven vendor solution reduces risk and shortens the path to demonstrable results.

The vulnerability I understand from chassis development: vendor AI is calibrated for generalised automotive use cases. Suspension and tyre behaviour are platform-specific to a degree that generic solutions genuinely struggle with. A predictive model trained on industry-average suspension data will miss the specific wear behaviour of your specific geometry running your specific tyre compound on your specific target markets’ road surfaces. Unlike a well-documented damper specification, the accumulated model intelligence doesn’t stay behind in an engineering drawing when the vendor relationship ends.

The Sequence That Actually Produces Results

The manufacturers pulling meaningfully ahead aren’t choosing between these three options. They are sequencing them with intention.

Borrow first — deploy a partner or vendor solution, prove the concept, and get the engineering team comfortable reading AI outputs alongside their own physical assessments. Buy next — bring in specialists to build on top of real test data and close the performance gap that any generic model inevitably leaves open. Build over time — develop internal engineers who own, interpret, and evolve the models across platform generations and new market requirements.

One automotive components manufacturer following this sequence improved OEE by 10–15%. No single phase produced that outcome. The sequence did. A workable talent composition to aim for: 50% upskilled internal engineers, 20–25% hired AI specialists, 25–30% external partners. Borrow to start. Buy to accelerate. Build to own.

What the Proving Ground Taught Me: The Classroom Is Confirming

In chassis development, we had a fundamental diagnostic principle: never evaluate a component in isolation. A tyre doesn’t wear abnormally because of the tyre. It wears abnormally because of the geometry, the loading, the compliance, and a dozen upstream decisions that all converge at the contact patch.

The same systems thinking applies directly here. AI isn’t underperforming in manufacturing because the algorithms are weak. It is underperforming because the talent system around those algorithms — the people, the trust, the domain integration — hasn’t been engineered with anywhere near the rigour of the product itself.

I came into chassis engineering believing that getting the physics right was the hardest part of the job.

I came into MBA understanding that it was never the hardest part.

B SAI KRISHNA is a PGD-GM student at XLRI Jamshedpur