The page looks polished and fast: strong visual hierarchy, clear top navigation, and the hero actually says what you do (“AI agents”) instead of drowning in buzzwords. The “Sample output” links are an unusually good move for trust in this space—you’re showing the work, not just promising it.
But you’re trying to sell two different motions on one homepage: (1) a productized subscription (“Shop Watch”) and (2) custom AI deployments (services). The result is a visitor who isn’t sure what the *primary* outcome should be. The hero CTA goes to Shop Watch, the product grid is a buffet of options, and the pricing section mixes USD subscription with CZK one-offs—credible, but cognitively heavier than it needs to be.
The biggest conversion leak is that “Request in English” and “Ask for a quote” are mailto links. Mailto is friction, breaks tracking, and turns a high-intent click into an unmeasured, contextless email draft. You’re making people do the work right when they’re most ready to convert.
Finally: you’re close to being “CRO-finished” (clear sections, pricing, guarantees, case studies), so the upside isn’t massive. This is more about removing friction, clarifying the primary path, and tightening proof. Overall potential feels medium, roughly ~18–26% if executed cleanly.