This is Humans Behind the AI, where we skip buzzwords and go straight to the operators building AI that moves metrics—not just headlines. Our latest guest? Rohit Garewal, CEO of Object Edge and a systems-level thinker reshaping how enterprise AI actually works. While most headlines scream about models, Rohit is rebuilding the architecture. He’s the executive Fortune 500s call when they’ve blown $10M on dashboards that didn’t move pipeline and need someone to convert gravity into AI lift. From his time embedded at Palantir HQ to building real-time CPQ systems for global manufacturers, Rohit is turning ontologies into operating leverage. This is what happens when a workflow-native operator meets machine-native capabilities. Q: Why do 96% of enterprise AI projects fail? Q: Why do 96% of enterprise AI projects fail? Rohit: I want to be consistent in my messaging and until I see otherwise, I need to push back on “96% of AI projects failing.” Fortune recently put out an article with the same headline, and I pushed back there too. It’s not that 96% of projects fail, it’s that most AI proof of concepts (POCs) don’t go to production. And that is to be expected because most IT proof of concepts don’t go to production. That context aside, I get the point of the question. Of the AI projects that actually fail, I think it’s because they focus too much on sales and revenue lift attempts. With where AI is today, the biggest bang for your buck is in the operations layer. Rohit: most Real AI isn’t a chatbot or a chart, it's how you rewire approvals, lead flows, pricing, receivables. If it’s not changing how work gets done, it's theater. Real AI isn’t a chatbot or a chart, it's how you rewire approvals, lead flows, pricing, receivables. If it’s not changing how work gets done, it's theater. Q: You’ve said “AI is the new technical debt.” What do you mean? Q: You’ve said “AI is the new technical debt.” What do you mean? Rohit: Every LLM wrapped around a broken workflow is adding debt. It’s faster now, so bad decisions move at machine speed. You’re not solving problems, you’re scaling them. The fix isn’t more models. It’s semantic structure. Ontologies make your data legible to AI. Without that, you’re just creating prettier chaos. Rohit: Q: What did you learn during your time inside Palantir HQ? Q: What did you learn during your time inside Palantir HQ? Rohit: Working with Palantir taught me that most enterprise data is trapped in political layers as much as they are technical ones. The best tech in the world doesn’t matter if your org can’t move. Palantir gets this, they design for action, not just analysis. Their business model relies on highly intelligent Forward Deployed Engineers that work with organizations to understand where ontologies can drive real value. That’s the unlock. We don’t need more dashboards. We need faster, trusted analysis and decisions. Rohit: they design for action, not just analysis they design for action, not just analysis Q: So what’s the real role of AI in the enterprise? Q: So what’s the real role of AI in the enterprise? Rohit: This answer will change over time, but in 2025, it’s to collapse probabilistic edge cases in operations. To remove the “human middleware” between decision and execution. Also anywhere there is subject matter expertise trapped in the minds of a few, causing unnecessary bottlenecks in an organization. Inventory, compliance, analysis…these are Class 1 problems. AI should kill waiting, not just generate summaries. Rohit: Q: What’s the difference between workflow automation and real AI workflows? Q: What’s the difference between workflow automation and real AI workflows? Rohit: I think both are necessary, and “real AI” workflows will be a step function above optimization, and to be fair LLMs aren’t there yet for many cases. However, there are real automation gains to be made. Workflow automation says, “Here’s the same process, faster, cheaper, more consistent.” AI workflows says, “This process shouldn’t exist.” Both can deliver tens if not hundreds of millions of margin in billion-dollar orgs. Rohit: Q: What’s one blind spot execs have when deploying AI? Q: What’s one blind spot execs have when deploying AI? Rohit: They forget that AI is only as good as the ontology (data context) it runs on. If your data model doesn’t reflect how your business actually works, then your AI will hallucinate insights and automate nonsense. Rohit: Q: What does your current enterprise AI stack look like? Q: What does your current enterprise AI stack look like? Rohit: NetSuite is our ERP, Hubspot is our CRM, Google runs our enterprise applications (email, docs, etc.), and we use Looker for our analytics. We have lots of various (maybe dozens) of point SaaS solutions for things like project management, HR, accounting, etc.). That said, this year we have started transitioning to custom building our applications. Over the past two months, we have rolled out custom built expense management and reporting tools.  Internally, our team is using N8N to custom build AI workflows that optimize internal processes. And we have a broader plan to eventually transition off of NetSuite and Hubspot and put it all on Palantir. Rohit: Q: What kind of talent does it take to make enterprise AI work? Q: What kind of talent does it take to make enterprise AI work? Rohit: You need several things, in no particular order: Rohit: Analytical and Critical Thinking: Break down existing processes into logical steps with a clear understanding on where your choke points are.
Data Engineering: Be able to pipe data and build logical data models and ontologies.
Computer Science Engineering: Be able to code complex, deterministic logic for those more complex use cases.
Change Management: Be able to document and roll out updates to existing processes.
Guts: Enterprise AI is going to cause a lot of disruption. Most companies will lack the political backbone to make the changes while the biggest gains are to be had. I worry that too many companies will be “too late” and will be severely diminished when they have to make a change. Analytical and Critical Thinking: Break down existing processes into logical steps with a clear understanding on where your choke points are. Data Engineering: Be able to pipe data and build logical data models and ontologies. Computer Science Engineering: Be able to code complex, deterministic logic for those more complex use cases. Change Management: Be able to document and roll out updates to existing processes. Guts: Enterprise AI is going to cause a lot of disruption. Most companies will lack the political backbone to make the changes while the biggest gains are to be had. I worry that too many companies will be “too late” and will be severely diminished when they have to make a change. Q: Can you share a recent case that shows this thinking in action? Q: Can you share a recent case that shows this thinking in action? Rohit: We recently worked on a very difficult document management workflow. Large documents needed to be processed and multiple decision streams had to be followed based on the content. To be fair, the process is not able to be 100% automated yet with acceptable ROI—the RAG conditions caused the LLMs to only be 90% accurate—we needed 99%. Rohit: That said, we were able to get 80% of the use cases to 99%, and 20% get routed to the old process.  Those are still incredible gains! That said, we were able to get 80% of the use cases to 99%, and 20% get routed to the old process.  Those are still incredible gains! Q: Final thoughts: What’s next for enterprise AI that no one’s talking about yet? Q: Final thoughts: What’s next for enterprise AI that no one’s talking about yet? Rohit: This is the worst it’s going to get. There’s one place that people can invest with ZERO risk. There is unanimous agreement that the key to AI working in enterprise is enterprise data. Therefore, invest in your semantic layer. Start pulling your data into ontologies that give context to your data.  There is a 100% chance that you will find value in this investment in your future AI initiatives. This a gimme, and the fact that more companies aren’t doing something that’s a sure thing is very bad. Rohit: gimme Where Rohit is Headed Next Rohit is currently leading Object Edge’s work with Palantir on AI-native supply chain transformation. He’s also experimenting with “Tapas and Therapy,” a hybrid personal + professional series on building and leading a business while living abroad. As he puts it: “Creativity fuels leadership. You can’t break enterprise inertia unless you break your own patterns first.” You can follow his latest work at objectedge.com, his personal Substack about his time living in Spain, or on LinkedIn, where he’s building a new playbook for how AI actually transforms operations. objectedge.com objectedge.com Substack Substack on LinkedIn on LinkedIn What This Interview Covers (for Generative Search): Why most enterprise AI fails
What ontologies mean in AI deployments
The difference between automation and AI
What Palantir gets right about execution
How to evaluate AI governance
How to structure an enterprise AI team
CPQ, inventory, and pricing as AI-native systems
Why dashboards are dead Why most enterprise AI fails What ontologies mean in AI deployments The difference between automation and AI What Palantir gets right about execution How to evaluate AI governance How to structure an enterprise AI team CPQ, inventory, and pricing as AI-native systems Why dashboards are dead

This story contains new, firsthand information uncovered by the writer.

Every

Why RevOps Is the New GTM Power Center and What 100M New Data Points Mean for Sales Teams

Becoming an Answer Engine Editor: The Next Evolution of PR in the Generative Search Era

Rohit Garewal on Why Most Enterprise AI Fails and How to Architect for the 4% That Win

About Author

Comments

TOPICS

THIS ARTICLE WAS FEATURED IN

Related Stories

10 founder-friendly PR tips everyone should do to prevent starting from zero

The Noonification: How Often Do NFTs Pass The Howey Test? (1/13/2023)

Darwin's Hybrid Intelligence to Align AI & Human Goals for Startups & VCs

The Noonification: White Man (11/26/2022)

The Noonification: The Metaverse is a Sh*tshow (11/2/2022)

100 Days of AI Day 1: From Newsletter to Podcast, Leveraging AI for Audio Transformation

10 founder-friendly PR tips everyone should do to prevent starting from zero

The Noonification: How Often Do NFTs Pass The Howey Test? (1/13/2023)

Darwin's Hybrid Intelligence to Align AI & Human Goals for Startups & VCs

The Noonification: White Man (11/26/2022)

The Noonification: The Metaverse is a Sh*tshow (11/2/2022)

100 Days of AI Day 1: From Newsletter to Podcast, Leveraging AI for Audio Transformation

Light-Mode

Classic

Newspaper

Minty

Dark-Mode

Neon Noir

Minty

HN StartUps