Event streaming vs request-response for real-time dashboards

we have a critical analytics dashboard that currently polls an api every 5 seconds to update charts and metrics. it's for internal use, but the data is pretty time-sensitive, and execs sometimes complain about it not being 'real-time enough.' we're exploring moving away from polling to a more push-based model. options on the table include websockets, server-sent events (sse), or even graphql subscriptions. another idea floated was bridging kafka topics directly to a websocket service. the main question is, how much real-time do we *really* need? is a 5-10 second stale data point truly a problem for an internal dashboard, or are we over-engineering for a perceived issue? i'm curious if others have gone down this path for internal dashboards. what was the justification for moving to event streaming, and what were the benefits vs. the added complexity? sometimes i feel like polling, while unsexy, is surprisingly robust and easier to operate at a certain scale, especially if the data staleness tolerance is reasonably high. thoughts?