Beyond the POC: AWS’s Playbook for Enterprise AI Success
The AI Forecast: Data and AI in the Cloud EraShow Notes
Most AI pilots never make it past the demo phase. AWS Machine Learning Lead Praveen Jayakumar has seen plenty of promising AI projects get stuck between a successful demo and production. Teams often define what success looks like without deciding what failure looks like, leaving underperforming projects alive long after they should have been shut down. As Praveen puts it, they become “zombie” AI projects. In this episode of The AI Forecast, Paul Muller sits down with Praveen Jayakumar, who leads Machine Learning Solution Architecture for Amazon Web Services (AWS) across Asia Pacific and Japan, to explore how enterprises can give AI projects a realistic path to production. Praveen shares what he’s learned working with organizations on machine learning and generative AI deployments, including why teams should build for production while they’re still proving the concept. Observability and evaluation become particularly important as models change, costs scale, and AI systems begin interacting with enterprise data. Paul and Praveen talk about:
Why promising AI proofs of concept stall before production How to define success and kill criteria before an AI project begins Why observability should be built into the proof of concept How evaluation frameworks make it easier to test different AI models Why the most powerful frontier model may be the wrong choice How model selection affects the unit economics of enterprise AI
Recorded at EVOLVE26 Singapore, this episode offers a practical look at what happens after the AI demo, when teams have to decide what’s ready to scale and what’s better left behind. If you’re responsible for enterprise AI, this episode will help you build a path to production and keep zombie projects from consuming resources long after their expiration date. Stay in touch with Praveen: Praveen Jayakumar on LinkedIn: https://www.linkedin.com/in/pjpraveen/ +++ Like and subscribe to The AI Forecast, sponsored by Cloudera, to follow our EVOLVE26 Singapore series and stay up to date on the latest conversations about enterprise data and AI.
More from The AI Forecast: Data and AI in the Cloud Era
One developer burned through 300 million AI tokens in a single week. The payoff was far less impressive. For Todd Trevillion, it's a striking example of what can happen when AI adoption outpaces the controls needed to manage it. And with autonomous AI agents beginning to make decisions on behalf of businesses, the stakes are getting higher. Recorded at EVOLVE26 Singapore, this episode of The AI Forecast follows Paul Muller as he sits down with Todd Trevillion, technology entrepreneur and co-founder of Citadel Edge, to explore whether enterprises are truly ready for agentic AI and what it takes to deploy the technology responsibly. Working with organizations in highly regulated industries, Todd has seen how quickly AI ambitions can run into operational reality. He questions whether enterprises are prepared to give autonomous agents greater control when the consequences of their actions can be difficult to predict. Paul and Todd dive into:
The AI worked in the sandbox. Then it met the real world. David Hardoon learned that lesson during an ATM optimization project designed to improve cash-delivery routes. Once deployed, the system collided with traffic and unpredictable schedules, which are the realities that face the people doing the work. Recorded at EVOLVE26 Singapore, this episode of The AI Forecast brings Paul Muller together with David Hardoon, Head of Advanced AI at Accenture, to explore why the human side of AI transformation can determine whether promising technology ever delivers value in production. Paul and David discuss:
Your AI agent has credentials. Now, what happens when it goes off script? An agent can take action and change course before a traditional security review even begins. For Alex Bovee, identity is only the beginning of the security conversation. In this episode of The AI Forecast, Paul Muller sits down with Alex Bovee, co-founder and CEO of C1 and longtime cybersecurity and identity leader, to explore how autonomous AI agents are changing enterprise security. Alex argues that zero trust was built for a world where people remained behind the keyboard. AI agents introduce non-deterministic behavior and can relentlessly optimize toward a goal, sometimes finding routes their creators never anticipated. Paul and Alex’s discussion examines:
An AI project can deliver a great prediction and still fail the business. Rajesh Kelvalkar has seen it happen firsthand. In one mining project, the model worked, but the team discovered that the downstream business process wasn’t configured to act consistently on its recommendations. The technology had done its job, but the transformation hadn’t. Recorded at EVOLVE26 Singapore, this episode of The AI Forecast brings Paul Muller together with Rajesh Kelvalkar, Senior Advisor at Tech Data APJ, to explore what more than 20 years of transformation work has taught him about putting AI into practice. Rajesh argues that organizations often approach transformation as a technology initiative, even though much of the work happens elsewhere in the business. Clear ownership and measurable outcomes determine whether an AI pilot becomes part of operations or disappears when the project team moves on. Paul and Rajesh explore:
After 23 years in machine learning and data science, Cao Hong has seen plenty of AI projects come and go. His measure of success is simple: did it create value for the business? Recorded at EVOLVE26 Singapore, this episode of The AI Forecast brings Paul Muller together with Cao Hong, Principal Director of Data Apps at NCS, to explore how organizations can turn AI experimentation into measurable business impact. That becomes harder as enthusiasm for AI pushes expectations higher. Cao Hong argues that organizations need to define the outcome they’re working toward early and then understand how AI will fit into how the business actually operates. Having worked across research and commercial AI projects, Cao Hong has seen how differently the two environments operate. Research can explore problems whose value may be realized years from now. In business, AI ultimately has to earn its investment. That means connecting the technology to a clear use case and establishing how its performance will translate into business value. Paul and Cao Hong explore: