Cracking the Agentic AI Code: What Enterprises Can Learn from Successful Payment and Revenue-Boosting Innovations

The rise of artificial intelligence is no longer news - it’s a reality. Yet, as more companies rush to deploy AI agents into their systems, many find themselves stumbling and failing. The blame doesn’t solely lay on their AI models but more so on their lack of preparatory groundwork. During VB Transform 2025, high-ranking leaders in the tech industry offered valuable insights derived from successful AI deployments on a grand scale.

Moderated by Joanne Chen, a general partner at Foundation Capital, the talk panel included seasoned executives like Shawn Malhotra, Rocket Companies’ CTO, Shailesh Nalawadi, Sendbird’s Head of Product, and Thys Waanders, Cognigy’s SVP of AI Transformation. Together, they divulged a critical revelation about successful AI adoption - Companies that prioritize building evaluation and orchestration infrastructure first are more likely to succeed, while those hastily charging towards production with intricate models often falter.

Rocket Companies’ Malhotra provided compelling evidence of AI’s cost-cutting prowess, sharing how an engineer’s two-day effort in crafting a simple agent resulted in saving a whopping million dollars per annum. Cognigy’s Waanders placed emphasis on the value of automating customer interaction, which resulted in significant decrease in the average handling time per call.

On the other hand, AI proves its worth as a revenue-generator. Malhotra mentioned that due to faster query responses and improved customer experiences, there’s been an uptick in conversion rates. Nalawadi added the value of proactive consumer interaction, unleashing new revenue avenues before the customer even perceives an issue.

Despite the apparent ROI opportunities of AI deployment, considerable challenges remain, particularly concerning production deployments. One poignant concern is the lack of evaluation infrastructure prior to the creation of AI agents. Traditional software testing approaches prove to be ineffective for AI agents, hence the unequivocal need for new and dedicated testing mechanisms. Building an evaluation infrastructure in advance and treating it as a ‘unit test’ for your AI agent system is critical.

To conclude, the value derived from AI agents goes beyond mere cost efficiency. It drives productivity, customer satisfaction, and revenue generation. However, such success comes with a caveat - The right infrastructure must be in place for AI to work its magic. That’s a lesson that’s needless to learn the hard way.