Hey everyone, I've been using Fireflies.ai across a few different teams now—some in Salesforce, some in HubSpot—and I keep staring at those "talk time" percentages in the meeting summaries. I get the basic idea, but I'm trying to figure out how to actually *use* this data beyond just a neat stat.
Like, if a sales call shows the AE at 60% talk time and the prospect at 40%, is that "good"? I've heard you want more prospect talk time in discovery, but then in a demo, maybe the rep should be higher? And what about internal syncs where one person is at 70%? Is that a monologue or just an effective update?
I'm curious how others are interpreting these numbers. Are you setting benchmarks for different meeting types? Have you found it useful for coaching reps on their balance, or maybe even spotting meetings that went off the rails? Also, does it count silence or crosstalk, or is it pretty smart about attributing who's actually speaking?
Trying to move past just observing the metric and into making it actionable for revops and team workflows. Any stories or tips would be awesome.
Yeah, that's exactly the right question. That 60/40 split is a classic example - it's not inherently good or bad without the meeting context. For a discovery call, I'd want the prospect talking way more than 40%, probably more like 70%. But for a technical demo where I'm explaining a complex feature, 60% might be totally appropriate.
The real power for me has been tracking the trend for individual reps. If someone's talk time spikes to 80% on a call that was supposed to be a discovery, that's a red flag for a monologue. It's a great, quick filter to find calls that might need a coaching review. I've also heard it helps spot if a prospect disengaged - their talk time often plummets midway through a bad demo.
On your last point, from my experience, the AI is pretty good at filtering out crosstalk and long pauses, but it's not perfect. I'd take the percentages as a strong directional signal, not a precise scientific measurement. Have you tried comparing the talk time metric against the actual call transcript to see how well it aligns?
The context is everything, but you're chasing a metric that's easily gamed. Someone can dominate the clock with open-ended questions that force the prospect to give long-winded answers, skewing the percentage in a way that looks 'good' on paper but was a terrible conversation. The software just measures air time, not quality or intent.
I'd be more worried about what you're paying for this 'insight.' How much does it actually cost to get that breakdown per seat, and what tangible coaching outcome have you seen? If a rep needs an AI to tell them they monologued for 70% of a discovery call, you've got bigger problems.
Also, good luck if you have a fast-talking prospect or someone with a thick accent. In my experience, the attribution gets messy, and then you're making decisions based on flawed data. Ever audited a call it scored versus your own ears? The discrepancies can be laughable.
—DW
You're asking the right questions. The percentages are a diagnostic starting point, not a verdict. The actionable step is to define those meeting-type benchmarks you mentioned, then use the deviation as a flag for review.
For example, we built a simple framework for our AEs:
- Discovery Call Target: Prospect 65-75%, AE 25-35%
- Demo/Presentation Target: AE 55-70%, Prospect 30-45%
- Internal Sync: No firm benchmark, but a consistent 70%+ from one person triggers a check-in on meeting structure.
This turns the stat from an observation into a filter. When a discovery call shows a 50/50 split, that's my queue to pull the transcript and look for why - was the prospect giving short answers, or was the rep explaining too much? It's saved our coaches hours previously spent randomly sampling calls.
On your last point about attribution, in my testing, Fireflies is reasonably accurate on clear audio for attributing speech. It does struggle with crosstalk, often assigning it to the last clear speaker, and prolonged silence isn't factored into the speaking percentages. The real value is in the aggregate trend, not any single call's precision.
Method over hype
Targets and frameworks. So you've now added a new KPI to manage and argue over. The minute you put a "target range" on a discovery call, your reps will start performing to the metric, not the conversation. They'll drag out pauses to hit the 65% prospect talk time, or worse, let the prospect ramble off-topic just to fill their quota.
You're trading one form of inefficiency for another. Instead of randomly sampling calls, you're now manually reviewing every call that falls outside arbitrary bands. Did you actually get hours back, or just shift the work?
And aggregate trend is only valuable if the underlying data is sound. If it struggles with crosstalk and accents, your trend is just a prettier version of a flawed signal.
Just saying.