# Post-call coaching arrives about a week too late Conversation intelligence tells a rep what they got wrong after the deal has already moved. The useful window is four seconds wide, and it is during the call. Every conversation intelligence tool on the market runs the same loop: record the call, transcribe it, analyse it, and show the rep what they could have done better. The analysis is often correct. It is also delivered several days after the buyer already decided how the conversation felt. ## The window is four seconds wide Watch where a deal actually turns. A buyer says something like _"we don't want to switch right now"_, and the next four seconds decide the rest of the call. The rep either says something specific about that buyer's situation, or they say the thing every vendor says. A scorecard delivered on Friday cannot reach into Tuesday. ## Why the gap persists The technical reason is straightforward: analysing a call properly requires the whole call. The useful signal in conversation intelligence is talk ratio, topic coverage and competitor mentions, none of it computable until the conversation is complete. Helping during the call is a different problem. It needs the buyer's last sentence, everything already known about their company, and an answer within about two seconds. That is a retrieval and latency problem, not an analytics one, and it is why the two categories look nothing alike under the hood. ## What has to be true to close it Three things, all of them unglamorous: 1. **The context has to already be loaded.** If the model starts researching the account when the objection lands, it has lost. 2. **The output has to be the sentence.** A bullet list is not usable at speaking speed. The rep still has to compose, and composing is the thing they did not have time for. 3. **It has to sound like the rep.** A line the seller cannot say in their own voice will not get said at all. None of that replaces post-call analysis. It is a different job, at a different moment, for a different person.