By Héctor C. Moncada Díaz
The Chatbase founder has spent three years watching enterprise teams buy AI support platforms that look impressive in demos and frustrate customers in production. He has a specific theory about why, and it goes well beyond any single feature.
A company can have a 70 percent deflection rate and still be losing customers over its support experience. Yasser Elsaid, founder and CEO of Chatbase, noticed this pattern early, and it convinced him that the industry has been organizing itself around the wrong unit of analysis. Most platforms are built and sold as a collection of features: a smarter chatbot here, a better handoff there, a faster routing engine somewhere else. Elsaid’s view is that none of those features matter much in isolation. What matters is whether AI and human agents function as a single operating system for support, or as two separate systems that occasionally talk to each other.
He arrived at that distinction through direct experience, not analysis. He ran customer support for his own product out of a personal Gmail inbox in the months after Chatbase launched in 2023. As the platform scaled to thousands of business customers across more than 80 countries, what he kept noticing was not any single broken feature. It was a broken relationship between two halves of the support operation that were never designed to work as one thing. A customer would email about a problem the AI had already addressed in chat the day before, and there would be no record of it. The AI did its part. The humans did theirs. The seams between them were where almost everything went wrong.
What Elsaid has built his platform around is the idea that AI agents and human agents should not be thought of as two systems with a handoff between them. They should be thought of as one system with two types of operators. The AI has full visibility into what a human agent has done in a conversation. A human agent has full visibility into what the AI has already tried. Neither one is working from a partial picture, and neither one is waiting for information to be handed off from the other. That is the principle. Everything else, including how transitions between AI and human happen, is just one expression of it.
The handoff moment is the most visible place this shows up, since it is the point where a customer can tell whether the system feels unified or fragmented. But Elsaid treats it as one symptom among several, not the design itself. The same principle governs a customer who opens a chat on Monday and calls about the same issue on Thursday, and finds the agent already knows what happened. It governs how a human supervisor sets a rule once, such as auto-approving refunds under a certain amount, instead of reviewing each one individually. And it governs whether a ticket’s full history survives a move from email to voice to a live agent, or starts to thin out with every transfer. Different mechanisms, same underlying design.
“Anything else that claimed to have a big impact on churn that was not about improving the product was probably more of a waste of time,” Elsaid has said. The comment was made about retention strategy specifically, but it reflects the same belief that runs through how he thinks about the whole support system: that the unified architecture is the product, and most of what looks like a retention or efficiency problem is actually a symptom of that architecture being incomplete.
The reason this is hard to demonstrate, in his view, is that a single feature is easy to show in a demo and a unified operating model is not. A vendor can run a clean, scripted conversation in a sales call and it will look identical whether the system behind it is unified or stitched together from separate tools. What is much harder to show, because it only becomes visible over months of real use, is whether the AI and the human team are still operating off the same information after the fortieth conversation, the fifth channel switch, the first edge case nobody scripted for. That is the gap Elsaid believes separates platforms that perform well in a demo from platforms that hold up over years of enterprise deployment.
What Elsaid is arguing for, ultimately, is a different way of evaluating AI support platforms altogether. Not a checklist of individual capabilities, but a single question: does this platform operate as one coherent system, or as a set of features that happen to be sold together? He believes enterprise buyers evaluating long-term platform strategy are starting to ask that question more directly, and that vendors who built around features rather than around a unified model will find that distinction harder to paper over as deployments mature.
Elsaid does not talk about disrupting the industry. He talks about building a system where the distinction between AI work and human work disappears for the customer, and where the architecture, not any single feature, is what determines whether that happens. Three years in, that orientation has produced more than 10,000 business customers and $10 million in annual recurring revenue, without outside funding. He would probably resist the idea that the numbers prove the thesis. He would say the customers are still deciding whether the system feels like one thing or two. He seems confident about which way that goes.
