From AI Experiments to P&L Impact: Ram Sukumar on Making AI Work in Manufacturing

Manufacturers had loads of data. The more difficult question is whether that data can help plants make better decisions, and ultimately, improve financial performance.

That difference is key to Ram Sukumar's approach to artificial intelligence.

Sukumar is Director of Operational Excellence at HEXPOL Compounding, where he supports continuous improvement initiatives across 13 plants in the United States and Mexico. With nearly 25 years of experience spanning rubber manufacturing, research, quality and continuous improvement, his interest in AI comes from a practical manufacturing challenge: companies have enormous amounts of operational data but often struggle to turn it into timely, actionable decisions.

In a conversation with Nishkam Batta, Editor-in-Chief of HonestAI Magazine, Sukumar discussed how manufacturers can evaluate AI, connect investments to P&L outcomes, scale pilots across plants and uncover the hidden costs buried inside manufacturing operations.

AI Should Solve a Manufacturing Problem

For Sukumar, adopting AI simply because the technology is advancing is not a strategy.

HEXPOL has been exploring AI across several areas, including connecting testing and manufacturing data, safety applications, forecasting, inventory optimization and process intelligence. But Sukumar emphasized that the objective is not AI adoption itself. The technology has to improve decision-making and support business priorities such as EBIT and inventory optimization.

That principle also shapes how he thinks about ROI.

Consider a production line running at 2,000 pounds per hour. If AI-supported process improvements help increase that to 3,000 pounds per hour, the value should be measured through the resulting business impact.

Does higher throughput reduce overtime or labor costs? Does it lower product cost and improve margins? Can the additional capacity support more sales? And does the change increase or decrease equipment wear?

Sukumar identified capacity, product margin and equipment value as an important dimensions, when evaluating such improvements.

The underlying message is straightforward: manufacturing AI eventually needs to connect to the P&L.

Start Small, Prove It and Then Scale It

The path from an AI idea to enterprise deployment does not have to begin with a company-wide rollout.

Sukumar described a data-correlation initiative being piloted at one plant, with the potential to expand across the 13 sites he supports if successful. The process starts with cleaning and validating the data, followed by helping people understand and interpret it. More advanced capabilities such as querying the data and performing correlation or regression analysis come later.

Process intelligence is following a similar philosophy: begin with a small number of plants, understand whether the approach works and control the initial investment before scaling.

For manufacturers overwhelmed by the number of possible AI use cases, this offers a practical model: prove value in a constrained environment before attempting enterprise-wide transformation.

Vendor Selection Goes Beyond the AI Demo

Choosing the right technology partner is another critical part of the equation.

Sukumar explained that vendor evaluation is not simply about finding the lowest-cost option. The team considers whether the technology can solve the business case, the vendor's previous experience, feedback from existing clients, implementation costs and the experience of the project manager supporting the engagement.

The pilot is particularly important.

Cost still matters because the investment ultimately needs to produce an acceptable return. Sukumar said manufacturing equipment investments typically look toward a two-to-three-year ROI range and raised the question of applying a similar discipline to AI and technology investments.

Making the Hidden Factory Visible

One of the most compelling opportunities Sukumar sees for AI is making manufacturing's hidden costs easier to identify.

Quality losses, scrap, downtime, excess inventory, freight costs and poor planning can all affect financial performance. The problem is that the operational activity and its financial consequence are not always visible together.

Sukumar described opportunities to use manufacturing data to identify patterns behind quality issues and move from reactive investigation toward earlier insight. He also discussed using process intelligence around downtime and throughput, as well as applying forecasting and optimization to inventory decisions such as minimum order quantities and safety stocks.

The longer-term opportunity is to connect these operational signals more directly with financial outcomes.

Scrap, downtime, inventory, freight and planning ultimately connect to general ledger accounts and the P&L. Bringing that information closer to real time could help plant management understand the financial consequences of operational decisions without requiring them to work through the finance layer first.

Prioritize AI Like Any Other Improvement Initiative

Manufacturers may identify dozens of potential AI projects. They cannot execute all of them simultaneously.

Sukumar's approach is grounded in a familiar continuous-improvement concept: an ease-and-impact matrix.

Projects are evaluated based on implementation difficulty, required resources, expected ROI, training requirements and change-management needs. That matters particularly in lean manufacturing organizations where plant teams cannot realistically execute 15 or 20 major initiatives simultaneously.

AI therefore does not need an entirely separate investment philosophy. It can be evaluated using many of the same disciplines manufacturers already use for operational improvement.

The Technology Will Change. The Business Problem Should Lead.

AI is evolving faster than most annual planning cycles. Sukumar acknowledged the difficulty of keeping pace with that change, but his response is not to chase every new model or capability.

Instead, the organization can remain focused on the underlying problems: sales growth, manufacturing throughput, quality and scrap reduction, or inventory optimization.

The specific AI model matters less than whether the solution addresses one of those problems cost-effectively and aligns with the company's long-term strategy. That may be one of the most useful lessons for manufacturers navigating the current AI market.

The objective is not to become the factory using the most AI. It is to build a factory where better data and better intelligence translate into better operational and financial decisions.

For manufacturing leaders, the question is therefore becoming less about "Where can we use AI?"

It is increasingly:

Bio Block

About the Interviewer