Unlocking Enterprise Efficiency with AI

What we talked about.

In this episode of the Sunny Ray Show, host Sunny Ray sits down with Chetan Saundankar, Founder and CEO of Coditation, a boutique enterprise AI, data engineering and product engineering company founded in Pune, India in 2016. Chetan also leads Plant360.ai, an industrial AI platform focused on unlocking legacy engineering data. He explains how Coditation helps enterprises move from AI experimentation to measurable business outcomes, drawing on his prior role as CTO of a healthcare startup acquired in 2015. The conversation covers why legacy code and old engineering drawings represent accumulated business knowledge rather than just technical debt, how enterprises can avoid runaway AI costs from unchecked chatbot usage, and why the last five percent of AI accuracy determines whether a project succeeds or fails. Chetan also details how Plant360 digitizes decades old engineering drawings into structured knowledge graphs, delivering documented efficiency gains, and where he sees genuine AI value emerging for enterprise clients today versus where the hype still outpaces results.

Coditation founder Chetan Saundankar on turning enterprise data chaos and legacy knowledge into secure, actionable AI outcomes.

The questions, and the answers.

What are you and Coditation building, and why should the world care?

I run two companies, Coditation and Plant360. Coditation is a boutique AI focused services company. We help customers adopt AI, essentially taking them from experimentation to outcomes. We work with enterprises and mid market companies to help their execs deliver on the AI promises they've made to their boards. Plant360 is our industrial AI product, the tech and platform layer specifically for industrial enterprises like oil and gas.

You started Coditation in Pune in 2016, well before the current AI wave. What made data and AI engineering the right bet then?

Before Coditation I ran two product companies, one in healthcare data starting in 2015. I saw a gap between what technology offered, back then it was machine learning, not AI, and the last mile layer needed to convert that technology into real business outcomes. Enterprises were investing heavily in data but lacked that handholding. That gap is exactly what became Coditation.

What's the most expensive AI mistake you've watched an enterprise make?

One financial institution built a copilot AI agent for their network operations team, but employees started using the chatbot like an internal ChatGPT, feeding it gigabyte sized files for unrelated questions. Their token costs ballooned so high, in the single digit millions of dollars, that they had to roll it back entirely. It shows what happens when you don't design for unintended consequences.

Plant360 digitizes engineering drawings into a knowledge graph. Why start there in the industrial stack rather than anywhere else?

We were forced into it. Industrial enterprises have decades of legacy drawings, are extremely data hungry, and are losing tribal knowledge as experienced staff retire. Our first project involved auto cad files and PDFs you can't run anything on. We realized brownfield data was still locked, so unlocking that knowledge became essential before we could move further up the stack, which is what we're doing now.

You cite efficiency gains from digitization, like 70 percent. Who measures that and what does it actually count?

We measure it against the real human effort it previously took, design engineers or draft engineers doing this work manually, and compare that baseline to results after deploying Plant360. On average we're north of 70 percent efficiency improvement with real clients. Even with very old hand drawn drawings from the 1950s and 60s, we've never seen results below 50 percent.

Where is AI genuinely creating value for your clients right now, and where is it still theater?

AI today is mostly about efficiency and productivity, automating human centric processes. Machine learning still matters for predictive use cases like forecasting and smart scheduling. The biggest gains from large language models come from unlocking knowledge buried in files, conversations and wikis to power business processes, along with software engineering. Clients do need a one to three year horizon to see real ROI.

You say an agent at 95 percent accuracy is a liability unless you know what the other 5 percent looks like. How do you find that out?

Software has always been deterministic, but AI outputs are inherently non deterministic, so you must design around that unknown gap. You can build smart tools to make human review efficient, or design agents with specific constraints. If you jump into a project without accounting for that 5 to 7 percent gap upfront, it usually leads to failures down the line.