Try OnsaMost sales managers listen to less than 5% of their team’s calls. The rest? Feedback comes from memory, gut feel, and whatever the rep chose to mention in the pipeline review.
Meanwhile, somewhere between 10% and 40% of inbound calls are being mishandled — prospects ready to buy, getting brushed off by reps who are tired, distracted, or just don’t know better. One company we worked with found that a single mishandled call had cost them $30K in customer acquisition spend. The prospect was ready to visit. The rep said “we don’t take appointments.”
The data is sitting right there — in every recorded call, every transcript, every awkward silence. The question isn’t whether to analyze it. It’s how deep to go.
We’ve found that sales call intelligence breaks down into five levels. Most teams never get past the first. The ones that do find revenue they didn’t know they were leaving on the table.
This is the simplest level, and it works from day one.
Set up automated detection for problem calls — calls where a qualified prospect got a bad experience. A rep who didn’t ask for next steps. A prospect who mentioned budget and timeline but got sent a generic PDF instead of a meeting invite. A call that ended with “I’ll follow up” and never did.
The numbers are consistent across teams we’ve analyzed: 10% to 40% of targeted calls have some form of handling failure. Not bad intent — just human error at scale. Reps get tired. They forget the discovery framework at 4 PM on a Friday. They get stuck on one objection and fumble the rest.
Here’s where it gets interesting. Of those mishandled calls, roughly a third can be recovered with a same-day follow-up. A quick call back: “I wanted to make sure we addressed your question about pricing — here’s what I should have mentioned.” That’s a 5-15% uplift in revenue just from recovering waste that already happened.
The ROI math is straightforward. If you’re spending $30K-40K per qualified inbound call (marketing spend divided by call volume), and 20% are mishandled, and you can recover a third of those — the numbers add up fast.
What to detect at Level 1:
— Calls where a qualified prospect didn’t get a next step scheduled
— Calls shorter than 2 minutes (likely a brush-off)
— Calls where competitor was mentioned but no differentiation was offered
— Calls where pricing was discussed but no follow-up was sent within 24 hours
Most sales leaders have a mental model of how busy their team is. They expect 40-50 targeted conversations per day.
The real number, consistently, is 3-4 targeted conversations per day. The rest is admin, internal meetings, CRM updates, and calls that go to voicemail.
This isn’t a lazy team problem. It’s a visibility problem. When you actually measure conversation output — not “time in seat” or “calls dialed” but productive, targeted conversations — the gap between expectation and reality is almost always shocking.
And the fix is remarkably simple. Making the utilization data visible and reviewing it weekly lifts productivity without any coaching, training, or process change. People start working differently when they can see their own numbers. The same principle applies to pipeline analysis — making hidden patterns visible changes behavior.
What to measure at Level 2:
— Targeted conversations per rep per day (not dials — conversations)
— Time-to-first-response on inbound calls
— Callback rate on missed calls
— Productive hours vs. admin hours ratio
This is where most “call coaching” tools stop. But the difference between generic coaching and data-driven coaching is enormous.
Generic coaching sounds like: “You need to ask better discovery questions.”
Data-driven coaching sounds like: “In 8 of your 12 calls last week, you skipped the Implication questions after identifying the problem. Here’s what happened — in the 4 calls where you did ask Implications, 3 resulted in a next step. In the 8 where you didn’t, only 1 did.”
The data makes the feedback specific, non-personal, and hard to argue with. You’re not saying “you’re bad at discovery.” You’re saying “here’s a pattern, here’s what it costs, and here’s what to do differently.”
From a pilot we ran — 25 deals, 108 calls, two SDRs — the single strongest predictor of a successful outcome was discovery quality. When the discovery score was 3 or higher (out of 5), the rep was 3 times more likely to book a meeting. Pain points were surfaced 2.5 times more often in those calls.
Here’s what that looks like in practice. A good coaching call:
SDR: “We have slots on Monday. Morning at 11 and afternoon at 3. I think 11 works — shall I book it?” Collected WhatsApp and email for confirmation.
And a bad one — same team, same week:
Prospect: “We signed with another vendor.” SDR: “But I’ve been chasing you for 5 months! You promised!”
The second rep defaulted to guilt-tripping instead of objection handling. Without data, this pattern would have stayed invisible — the rep filed it as “prospect went cold” and moved on. With call scoring, the Head of Sales could point to 4 similar calls that week and say: “Here’s the pattern. Here’s what you do instead.”
The coaching implication was clear: don’t tell the reps to “sell better.” Tell them exactly which questions they’re skipping and show them the calls where they did it right. One useful trick: generate anonymized “bad call” scenarios from real patterns for training exercises, but use actual recordings for the “best call” highlights — the rep who nailed it gets recognition instead of shame.
What to build at Level 3:
— Weekly call scorecards per rep (not individual call grading — pattern summaries)
— Specific skill gaps: discovery, objection handling, closing, follow-up
— “Best call of the week” highlights — let good calls become training material
— Comparison against the team’s top performer (anonymized if needed)
One important note about talk-to-listen ratio: 60/40 is the benchmark (prospect talks 60%, rep talks 40%). In practice, the best discovery calls start with the prospect talking 80%+, then the rep pitches, then the prospect comes back with questions. If your rep is at 70% talk time, the problem isn’t that they talk too much — it’s that they’re not creating space for the prospect to reveal their situation.
Here’s the shift that separates Level 3 from Level 4: stop reviewing calls one by one.
At Level 3, you listen to a call and give feedback. At Level 4, you analyze all calls from the past two weeks and extract patterns across the entire team.
This changes the kind of insights you find:
Zombie deal threshold. In our pilot data, deals that required more than 7 calls without reaching a meeting were almost never going to convert. After 7 calls, the probability of success dropped to near zero. That’s not an intuition — it’s a pattern across 108 calls. The actionable decision: auto-escalate any deal past 7 calls to the Head of Sales for a kill-or-escalate decision.
Competitive intelligence. When you aggregate what prospects say about competitors across dozens of calls, you get a real-time competitive landscape that no analyst report can match. “DataGuard keeps coming up in Healthcare deals” or “Prospects in manufacturing consistently cite integration problems with their current vendor.” With 6-7 SDRs, this intelligence gets lost in individual heads. With AI aggregation, it flows to the team weekly.
Product gaps. In our pilot, three pain points kept surfacing across calls: “stuck on spreadsheets,” “bad vendor support,” and “poor QuickBooks integration.” Each one was invisible at the individual call level — no single rep would flag it. But when AI aggregates across all calls, product priorities start writing themselves.
Seasonal patterns. Budget cycles, hiring seasons, and industry events create predictable waves in call content and conversion rates.
What to build at Level 4:
— Bi-weekly pattern reports (not daily — you need enough calls for signal)
— Competitive mention tracking with sentiment
— Product feature request aggregation
— Conversion rate by call pattern (discovery depth, talk ratio, objection type)
This is the level most teams never reach, but it’s where the compounding value lives.
Take everything from Levels 1-4 — the alerts, the utilization data, the coaching patterns, the competitive intelligence — and auto-generate a living knowledge base.
This knowledge base contains:
How your company actually sells. Not the playbook someone wrote 18 months ago — the real patterns from your best calls this quarter. What phrases work. What objections come up. How the top performers handle the “we’re happy with our current vendor” response. What the listen-to-talk ratio looks like in calls that convert.
Training material that writes itself. New hire onboarding goes from “shadow someone for two weeks” to “here are the 50 best calls from last quarter, categorized by situation and technique.” The knowledge base becomes the training program.
The foundation for AI assistants. If you want to build a real-time call assistant — one that suggests questions during the call, flags when discovery is shallow, or prompts the rep to ask about budget — the knowledge base from Level 5 is what you train it on. Without it, you’re building on generic frameworks that don’t match your market.
A Gong subscription runs roughly $50K per quarter for a 10-person team. At that price, you get a UI with transcripts and some analytics. What you don’t get is a knowledge base that’s actually tailored to your sales process, your market, and your competitive landscape. The value isn’t in the recording — it’s in what you extract from it.
When you start scoring calls (Level 3+), you need a framework. Three dominate the market, and they serve different purposes:
MEDDPICC — works best for enterprise deals ($50K+, 3+ month cycles, buying committees). It’s a checklist: did the rep identify the Economic Buyer? The Champion? The Paper Process? Easy for AI to score because it’s binary — either you covered it or you didn’t.
SPIN — works best when discovery quality is the bottleneck. Situation → Problem → Implication → Need-Payoff. The magic is in the I and N questions — where the prospect realizes the cost of inaction. If your reps are “premature pitchers” who jump to the solution before understanding the problem, start here.
Challenger — works best in commoditized markets with informed buyers. The rep teaches something the prospect didn’t know, tailors it to their specific situation, and takes control of the process. Hardest for AI to evaluate because it requires behavioral analysis, not just checklist matching.
The practical recommendation: if you’re not sure, start with SPIN for discovery evaluation and layer in MEDDPICC for deal qualification tracking. Use Challenger principles for how your reps position themselves, but don’t try to score it computationally until your pipeline is mature.
And here’s the insight that matters most: sometimes there’s no difference between winning and losing calls on methodology scores. If your AI analysis shows that won deals and lost deals look identical on call quality, the bottleneck isn’t in how your team sells — it’s in who marketing is bringing to the call. That’s a lead quality problem, not a coaching problem.
Here’s what a real pilot looks like, end to end.
The setup: A B2B company with two SDRs whose job was booking meetings and demos. The Head of Sales was spending roughly 4 hours per week on selective call listening — maybe 10-15 calls out of 100+. Feedback was based on whatever calls he happened to catch.
The pipeline: CRM recordings → download → transcription (AssemblyAI, $0.15/hour) → AI scoring with SPIN framework → pattern analysis across all calls → executive report. Total setup time: one week from CRM access to full report.
The data: 25 closed deals (half won, half lost), 108 calls total, ~70 long enough for meaningful analysis. Each call scored on discovery depth, pain identification, next steps, and talk-to-listen ratio.
What the data showed:
Discovery was everything. When discovery scored 3+ out of 5, the rep was 3x more likely to book a meeting. Pain points surfaced 2.5x more often. Lost deals almost never got past surface-level questions — the rep jumped to pitching before understanding the problem. This single insight redirected the entire coaching strategy for the quarter.
The zombie threshold was 7 calls. Deals that required more than 7 calls without reaching a meeting had near-zero conversion probability. Before this analysis, several open deals had 10-12 calls logged — the reps kept trying because nobody told them to stop. The Head of Sales implemented a simple rule: any deal past 7 calls gets escalated for a kill-or-escalate decision.
Hidden competitive intelligence. Aggregating what prospects said about competitors across all 108 calls produced a competitive landscape that no analyst report could match. Three recurring pain points — spreadsheets, bad vendor support, poor QuickBooks integration — gave the product team immediate feature priorities.
The coaching got specific. Instead of “you need to ask better questions,” the Head of Sales could say: “In 8 of your 12 calls last week, you skipped Implication questions. In the 4 where you asked them, 3 resulted in next steps.” The data made the feedback non-personal and actionable.
The before/after: 4 hours per week of random listening → an automated weekly report covering every call, with specific patterns, coaching recommendations, and deals flagged for review. The Head of Sales now spends 30 minutes reviewing the report instead of 4 hours sampling calls — and catches more problems, not fewer.
You don’t need a $50K tool to start. Here’s how to run a meaningful pilot in one week:
Day 1-2: Collect. Pull 20-25 closed deals (half won, half lost). Download the call recordings. If you don’t have recordings, start recording today and revisit this in a month.
Day 3: Transcribe. Use any speech-to-text service. AssemblyAI is $0.15/hour with good accuracy. Privacy matters here — make sure your provider lets you delete transcripts and doesn’t use your data for training.
Day 4: Score. Feed transcripts to an AI model with your chosen framework. Start with SPIN — it’s the most universally applicable. Score each call on: discovery depth (1-5), pain identification (1-5), next step clarity (1-5), and talk-to-listen ratio.
Day 5: Compare. Split results by won/lost. Where are the gaps? If discovery quality is the strongest predictor (it usually is), that’s your coaching focus for the next quarter.
Day 6-7: Build alerts. Set up Level 1 — flag calls where a qualified prospect didn’t get a next step. This alone will start recovering revenue within the first week.
The key is not to overthink it. You’ll iterate the framework, refine the scoring, and add dimensions over time. But starting with 5 basic questions and 25 calls will give you more signal than most teams get from months of ad hoc listening.
How many calls do I need for meaningful analysis?
For individual coaching (Level 3), 10-15 calls per rep per week is enough to spot patterns. For team-wide pattern analysis (Level 4), you want at least 50-100 calls over 2-4 weeks. For knowledge base generation (Level 5), 200+ calls over a quarter gives you a comprehensive foundation. Start small — even 25 calls from a pilot will reveal surprising patterns.
Won’t reps resist being monitored?
Frame it as coaching, not surveillance. Share the analysis with reps — show them their own patterns and improvement trajectory. Use “best call of the week” highlights so the system rewards good behavior, not just flags bad. The teams that do this well create a culture where reps ask for their own call reviews, because the feedback is specific and actionable, not punitive.
Can AI really evaluate sales calls accurately?
For structured frameworks like MEDDPICC (checklist-based), AI accuracy is high — 85%+ agreement with expert evaluators. For nuanced frameworks like Challenger (behavioral), it’s improving fast with reasoning models but still needs human calibration. The practical approach: have AI score everything, then spot-check 3-5 calls per week manually to keep the system honest.
What about privacy and compliance?
This is critical. Choose transcription providers that allow data deletion, don’t train on your recordings, and comply with your jurisdiction’s recording consent laws (one-party vs. two-party consent varies by state/country). Always inform prospects that calls may be recorded. Never store raw audio longer than needed — transcribe, analyze, then delete the recording.
How does this compare to buying Gong or Chorus?
Gong and Chorus are excellent products if you have $50K+/quarter budget and want a polished UI with team-wide dashboards. The trade-off: you’re locked into their analysis framework, CRM integration costs extra, and the prompt/scoring logic isn’t customizable. Building your own pipeline (even a basic one) gives you full control over what gets scored, how, and where the results live. Many teams start with a custom pipeline for the first 6 months, then evaluate whether a packaged solution adds enough value to justify the cost.
I’m Bayram, founder of Onsa.ai. We build AI agents for sales research, lead qualification, and pipeline analysis. Call intelligence is part of how we help teams find revenue they’re leaving on the table. Try Onsa.ai to see what your pipeline is hiding.