Event streaming vs request-response for real-time dashboards

we have an analytics dashboard that currently polls an api every 5 seconds to update metrics. it's generating a fair amount of load, and the product team wants a more 'real-time' feel. we're considering moving to a more event-streaming model. options include websockets, server-sent events (sse), graphql subscriptions, or even bridging kafka topics directly to websockets. my question is, how much real-time do users actually need for a dashboard? for data that's only stale by 5-10 seconds, is the overhead and complexity of a full-blown event streaming solution truly justified? or is polling, perhaps at a slightly higher frequency, actually 'good enough' for most business users? when does the cost-benefit analysis tip towards full event streaming for dashboards versus optimized polling?