How to Use Fireflies Conversation Analytics to Coach Reps
Most teams install Fireflies, get transcripts, and never open the analytics tab. The analytics are the part that changes rep behaviour — a talk-ratio number on a dashboard does more than the 40-page transcript nobody reads.
The catch: conversation metrics are easy to measure and easy to misread. Talk ratio below 50% isn’t automatically good. Here’s how the numbers work and which ones deserve a weekly review.
What does Fireflies analytics actually measure?
Fireflies computes per-meeting metrics from the transcript and speaker labels. The ones that hold up:
- Talk ratio — percentage of speaking time per participant. Derived from transcript timestamps, so it’s reliable if speaker labels are right.
- Sentiment — positive/neutral/negative per speaker, per sentence. Crude, but trend direction is informative.
- Monologue detection — the longest uninterrupted stretch from one speaker. Anything over 3 minutes in a discovery call is a signal.
- Keyword tracking — custom phrases such as competitor names or “budget”, counted across calls.
- Question rate — questions per 10 minutes from each speaker.
The ones to be sceptical of: overall sentiment scores (they drift with transcript quality), and anything derived from topic detection without your custom vocabulary configured.
Set up keyword tracking first. It’s under Settings → Conversation Intelligence, where you add a list of phrases. Put in your competitors, and “pricing”, “contract”, “security review”, “integration”. Three weeks later you have a real answer to “how often do we hear about Competitor X”, which nobody has.
How should you read talk ratio?
The common advice is “talk less than 50%”. It’s directionally right and too blunt.
Discovery calls: a rep above 60% is pitching instead of asking. That’s a coaching conversation.
Demo calls: 50-65% is normal. Someone has to drive the screen.
Pricing negotiation: reps typically talk 70%+ and that’s fine — you’re stating terms.
So compare within call type, not globally. Fireflies doesn’t segment by call type automatically, but you can tag meetings (Meeting → Add to channel) or filter by meeting title pattern, and read the metric per channel.
The much more useful metric is longest monologue. A rep who talks 45% overall but delivers a 6-minute uninterrupted pitch has the same problem as one at 65% overall, and it’s far easier to point at in a 1:1. “Here are 40 seconds where you answered a question the buyer hadn’t asked yet” lands better than a percentage.
How do you turn metrics into a coaching loop?
A weekly loop that survives: 20 minutes, one rep, one metric, one clip.
- Pick a metric for the week — not all of them. Question rate, or monologue length.
- Pull the outlier call. Sort the rep’s calls by the metric. Pick the worst one that isn’t already known-bad.
- Send a 60-second clip, not a transcript. Fireflies lets you share a timestamped link; a shared link to 14:32-15:40 of the actual call outperforms 2,000 words of summary.
- Agree one change. “Ask about timeline before showing pricing.” One thing.
- Check next week whether the metric moved.
The failure mode is doing this monthly with a spreadsheet of 12 metrics. Reps disengage because it reads as surveillance. One metric, one clip, one change, every week, is a different message.
What’s the fastest way to compare competitors?
Keyword tracking gives counts. To get context, use the search on the stored transcripts: search a competitor name and you get every mention across the org with a timestamp and a link.
Build a quarterly competitor sheet from it:
| Competitor | Mentions | Meetings | Common objection | Where we lose |
|---|---|---|---|---|
| Competitor A | 34 | 22 | “already have a contract” | Renewal timing |
| Competitor B | 19 | 15 | “cheaper for small teams” | Seat pricing |
| Competitor C | 8 | 6 | “no one’s heard of you” | Social proof |
The “where we lose” column can’t be extracted automatically — it comes from reading the flagged mentions. But filtering to 34 mentions instead of 400 calls is the difference between doing the analysis and not.
Practical warning: keyword matching is literal. “Competitor B” won’t match “B’s pricing” or a mis-transcribed variant. Add 2-3 spelling variants per competitor, and check the tokenisation by searching a phrase you know appears.
What should you not do with these metrics?
Don’t rank reps by sentiment. It measures transcript quality and speaking style more than deal outcomes. A rep with a flat delivery on a happy call scores worse than a warm-sounding rep losing the deal.
Don’t put talk ratio on a leaderboard. It gets gamed instantly — reps go quiet, buyers fill the silence with less information, and calls get worse.
Don’t benchmark against Fireflies’ suggested averages. They’re aggregated across all customers and skew toward high-volume inside sales. Your enterprise deals will look wrong against them.
Don’t review analytics without watching a call. The number tells you where to look. It never tells you why.
What’s the actual adoption problem?
Transcript tools get abandoned when nobody’s job includes reading them. Two things fix that:
Assign an owner. One person (sales manager or enablement) whose job includes a weekly pass. Without a named owner, analytics rot.
Tie it to something already reviewed. If there’s a pipeline review on Mondays, add one slide of conversation data to it. Tools survive when they attach to an existing ritual, not when they create a new one.
Also check Settings → Team → Permissions. By default, reps can see each other’s transcripts. Some teams thrive on that; others have a quiet mutiny. Decide deliberately before someone finds out by accident.
Quick Summary
- Fireflies analytics track talk ratio, monologue length, question rate, sentiment and custom keywords.
- Configure keyword tracking with competitor names first — it answers a question nobody can currently answer.
- Read talk ratio within call type; the longest monologue is a better coaching target than the average.
- Coach with one metric, one 60-second clip and one change per week.
- Don’t rank reps on sentiment or put talk ratio on a leaderboard — both get gamed.
- Analytics get abandoned without a named owner and an existing meeting to attach to.
