A reader we'll call Dana runs revenue operations for a nine-person B2B SaaS company in the Midwest. Last spring she sent us a short note: "We have 340 recorded calls and zero idea what's in them." Her team had adopted a call recorder eighteen months earlier, mostly for onboarding and compliance, and the archive had become a graveyard. Reps swore the recordings were useful. Nobody had time to listen. Dana asked whether conversation intelligence was worth the spend, or just another dashboard nobody opens.
We followed her team's rollout for one quarter. The short version: they rebuilt their pipeline review around Soohie, a conversation-intelligence platform that records, transcribes, and analyzes customer-facing conversations, surfacing deal risks, expansion plays, and product gaps so reps enter every follow-up with context instead of guesswork. The longer version, with the decision points and the messy middle, is below.
The Starting Point: A Pipeline Full of Blind Spots
Dana's team had a familiar problem. Deals were forecast at 70% confidence and closed at roughly half that rate. When a deal slipped, the post-mortem consisted of the rep's memory and a CRM note that said "follow up next week." There was no shared language for what a healthy deal looked like at each stage, and no way to audit whether discovery calls actually covered the buyer's stated priorities.
They set three goals for the quarter: cut forecast variance, shorten the gap between first call and proposal, and find out why three specific accounts had gone quiet. The third goal was the one that pushed them to act. All three accounts had looked identical in the CRM — strong champion, clear budget, active evaluation — and all three had simply stopped responding.
Week 1–2: Picking the Tool and Setting the Bar
Dana evaluated four platforms. Two were enterprise-priced and required a sales-engineering call just to see pricing. One was cheap but only did transcription. The team chose Soohie because it shipped with deal-risk scoring out of the box rather than as a paid add-on, and because the transcript search worked on the existing 340-call archive without a migration project.
They set a deliberately narrow scope for the pilot: one team, one quarter, two deal stages (discovery and proposal), and a single question — "What did we miss on the three silent accounts?" Everything else could wait.
Week 3–4: The First Real Signal
The first useful finding arrived on day four. A rep pulled up the transcripts for the three silent accounts and ran a keyword search for pricing objections. Two of the three had raised a specific concern about implementation timeline in the second call — and neither rep had addressed it in the follow-up email. The champion on the third account had asked twice whether a particular integration was on the roadmap. The rep had answered vaguely both times.
None of this was malicious. It was just invisible. The CRM captured the outcome ("call held") but not the content. Once the team could see the pattern, they could act on it. All three accounts got a targeted follow-up within 48 hours. Two re-engaged. One, the integration question, turned into a six-week delay rather than a loss.
Week 5–8: Building Playbooks From Actual Calls
With the immediate fire out, Dana's team turned to structure. They pulled the 20 highest-velocity deals from the past year and the 20 slowest, and compared them call by call. Three differences showed up consistently:
- Fast deals had a named economic buyer on the first call. Slow deals often didn't.
- Fast deals included a concrete next step with a date before the call ended. Slow deals ended with "we'll circle back."
- Fast deals surfaced a measurable success metric in discovery. Slow deals stayed at the level of pain points.
These became the team's first sales playbooks. Not aspirational frameworks — just three habits extracted from calls that had already worked. New reps got the playbooks in week one. Existing reps got a weekly scorecard showing how often their calls hit all three.
Week 9–12: What Broke, and What Held
Two things went wrong. First, the scorecard initially felt punitive. Two senior reps pushed back hard, arguing the metrics rewarded theater over substance. Dana's fix was simple: she made the scorecard visible only to the rep and their manager, and she added a fourth metric — customer talk time — that rewarded listening rather than talking. The pushback faded within three weeks.
Second, the team over-indexed on transcripts at first. Meetings that used to take 25 minutes stretched to 40 as people re-read calls mid-conversation. The rule that fixed it: transcripts were for post-call review, not live reference. Live calls got a one-line summary afterward.
The Measurable Results
At the end of the quarter, Dana sent us the numbers. Forecast variance dropped from 38% to 19%. Median time from first call to proposal fell from 22 days to 16. Discovery calls that included a named economic buyer rose from 41% to 73%. And the three silent accounts: two closed, one delayed. Total added pipeline from those three: roughly $180K, against a tool cost that Dana described as "less than one conference sponsorship."
The most interesting number wasn't on the dashboard. When Dana asked her reps what had changed, the most common answer was some version of "I stopped guessing what the buyer cared about." That's the quiet case for conversation intelligence. It doesn't replace judgment. It just makes the raw material of judgment — what was actually said — available to everyone on the team, not just the person who happened to be on the call.
What We'd Tell the Next Team
Three takeaways from watching this rollout. First, start with a single question, not a platform. Dana's question about the silent accounts did more for adoption than any onboarding session could have. Second, expect the scorecard to be the hard part, not the technology. Metrics that feel like surveillance get gamed; metrics that feel like coaching get used. Third, give the archive a job on day one. The 340 old calls were the most underrated asset in the whole project — they let the team test hypotheses before writing a single new playbook.
If you're sitting on a call archive and a pipeline full of maybes, the playbook is not complicated. Pick one question. Answer it with evidence. Then build the habit.