A senior AE at a Series B portfolio company sent me her last 20 discovery calls last month. She wanted me to score them. Closed-won rate on those 20 calls: 15%. Industry benchmark for that segment is closer to 28%. She is a strong rep. Her decks are tight, her demos are clean, her email follow-ups are better than mine. And she was losing deals on the first 30 minutes of every cycle.
The pattern was identical across all 20 recordings. She asked solid questions. The buyer answered. She moved to the next question. She never came back. By minute 22 she was screen-sharing the product. By minute 35 the buyer had stopped engaging.
This is the discovery problem in B2B right now. Reps know the frameworks. They have the question lists. They have AI tools that transcribe everything and remind them what MEDDPICC stands for. And win rates on first-meeting-to-closed-won are still hovering around 17 to 22 percent for most SaaS sellers I work with. Something is broken at the discovery layer, and it is not the question list.
I want to walk through what is actually going wrong in 2026, what the data from Gong and Chorus says about top performers, and the question patterns that move deals versus the ones that just fill the calendar.
What the call recordings actually show
Gong analyzed 519,000 discovery calls and the gap between top performers and average reps is sharp.
A few things stand out from that data.
Top reps talk less. Not by a small margin. Average reps own about two thirds of the airtime on discovery, which means the buyer is reacting to the rep instead of the rep reacting to the buyer. The 46/54 split that wins is not a soft preference. It is the operating point where the rep is actually learning something instead of presenting.
Top reps ask more questions but stop at 14. Past 14, close rates drop. Buyers feel interrogated. This is the line between diagnostic conversation and intake form, and most reps either fall short of 8 questions or push past 18 trying to fill out their MEDDPICC sheet.
Calls under 30 minutes are 28 percent less likely to advance. This is counterintuitive because reps think they are respecting the buyer's time. They are mostly cutting off the part of the conversation where pain becomes specific.
Monologues over 76 seconds from the rep cut close rate by about 15 percent. Monologues over 90 seconds from the buyer increase close rate. Read that twice. The single biggest signal of a real opportunity is the buyer talking uninterrupted for over a minute and a half. If you cannot get that in your first call, you do not have a discovery problem. You have a relevance problem.
Why the frameworks are not saving you
I have written about MEDDIC and how most teams break it. The same problem exists for every popular framework. Reps treat the framework as a script. They run through the letters. They fill in the boxes after the call. The deal desk is happy and the buyer is gone.
BANT was built in 1968 by IBM. The B was always supposed to be an outcome of value-building, not a gate at minute 8. SPIN selling has the right structure but reps skip the Implication questions, which are the actual revenue driver, and over-index on Situation questions that read like a survey. MEDDPICC is a deal review tool. It is not a discovery method. Chris Orlob has been pretty direct about that distinction and he is right.
A framework is what you fill in after a good call. It is not what you run during one.
If your discovery sounds like the framework, you are interviewing. Buyers do not buy from interviewers. They buy from people who know their job better than they do.
The reason this matters in 2026 specifically is that AI tools make the framework problem worse, not better. Gong and Avoma will fill in your MEDDPICC sheet from the transcript. Salesloft Rhythm will tell you which fields are missing. The temptation is to chase the missing fields on the next call instead of running an actual diagnostic conversation. I have watched reps spend 8 minutes asking who the economic buyer is because their CRM had an empty field. The buyer notices.
The peer-level diagnostic
The best discovery calls I review do not feel like sales calls. They feel like a strategy consultant talking to a CFO. The rep walks in with a hypothesis based on the company's stage, segment, signals, and 10 minutes of LinkedIn research. The first question is not "tell me about your business." The first question is "I read that you went from 12 reps to 30 last quarter and we usually see this exact pattern when that happens. Is that what you are seeing?"
The buyer answers in two ways. Yes, that is exactly what we are seeing. Or no, our problem is actually X. Either answer is gold. You either confirmed your hypothesis and now you are ten minutes ahead, or you have a real problem to chase.
This is what peer-level means. You are not asking permission to ask questions. You are doing the diagnostic the buyer's own ops team should be doing.
To do that you need three things before the call. You need the company's actual stage and trajectory, not a guess. You need a hypothesis about what their top two pains are based on similar customers you have seen. And you need three questions that test that hypothesis from different angles. If you cannot do that in 30 minutes of prep, your ICP is too broad. Fix that first.
The questions that actually move deals
Here are the question patterns I see in the calls that close. None of these are mine. I have stolen them from John Barrows, Josh Braun, Chris Orlob, and a handful of AEs I respect.
Walk me through what happens if you do nothing for the next 12 months. This is the cost-of-inaction question. Most reps ask it as "what happens if you do not solve this?" which is leading and the buyer will give you a polite answer. The 12-month frame forces the buyer to actually project the cost. Listen for hesitation. Hesitation means they have not actually thought about it, which means this is not a real priority.
Who else is feeling this pain and how do they describe it? This is the multi-threading question disguised as a discovery question. You need to know who else cares before you ask for an intro. The HubSpot data shows deals with 4 or more stakeholders engaged close at 3x the rate of single-threaded deals. You earn the second stakeholder by asking about them, not by demanding access.
What have you already tried and why didn't it stick? This is the most underused question in B2B sales. Buyers always have a graveyard of failed initiatives, internal projects, and tools they bought and never used. If you know what is in the graveyard, you know exactly what objections you will face in procurement and what stories your champion will need to tell internally.
If we built a business case, who would need to sign off besides you? Not "who is the decision maker." That question gets a defensive answer. The "if we built a business case" frame puts you on the same side of the table. You are co-building. The list of names you get back is your real account map.
What would have to be true for this to be a top three priority next quarter? This is the deal-or-no-deal question. The answer tells you whether you are looking at a real opportunity or a research call. If the answer is vague, you have a research call. If the answer is specific (board offsite next month, comp plan resets in Q3, churn target tied to Series C narrative), you have a deal.
The questions in the right column produce CRM-grade artifacts. After the call, your champion has a paragraph on the cost of inaction, a list of stakeholders, a real account map, an objection map from past failures, and a clear read on priority. That is the difference between a discovery call and an interview.
The AI paradox in 2026
About 81 percent of revenue teams now use conversation intelligence according to the Pavilion Pulse Q1 2026 report. Gong, Chorus, Avoma, Fireflies. Every call gets transcribed, summarized, scored, and pushed back into the CRM.
The intent is good. The effect is mixed.
Top reps in 2026 still listen actively in the moment. They take handwritten notes on a notepad while the buyer is talking. They ask follow-up questions based on tone and hesitation, not just words. The AI is a backup, not the system of record for what actually happened.
Average reps have started outsourcing the listening. They half-pay-attention, run their question list, and trust Gong to catch what they miss. The result is calls where the rep never asks a real follow-up. The buyer says "we tried this two years ago and it failed" and the rep says "got it" and moves on. The AI captures that the buyer mentioned a previous attempt. It does not capture that the rep had a chance to ask why it failed and skipped it. That follow-up was the moment that would have moved the deal.
The fix is mechanical. Tell yourself you will not look at the screen during the call. Use the AI for the post-call MEDDPICC fill, the call summary email, and the deal-review preparation. Do not let it run cover for poor listening.
How we use AI for discovery prep, not discovery itself
We run a fairly opinionated stack at Ziel Lab. Discovery prep happens in Clay and n8n. The call itself is human. Post-call analysis goes back through Gong or Avoma into HubSpot.
The thing that makes this work is the order. The framework lives at the end, not the beginning. The rep is not running through MEDDPICC during the call. The rep is running a peer-level diagnostic, taking notes, and then letting the AI organize the data afterward.
If you flip that order and ask the rep to fill out MEDDPICC fields during the call, you will hit the 14-question wall, the buyer will feel interrogated, and your win rate will sit where it has been sitting.
What buyers actually complain about
I run a small group of B2B buyers who agree to give me anonymous feedback on calls they take. The patterns are consistent.
First, premature pitching. About 67 percent of lost deals show product mention in the first 9 minutes per Chorus 2024 data. Buyers say it differently. They say "I felt like the rep had a script and I was just there to confirm the script." The premature pitch is what tells the buyer this is not a peer conversation.
Second, no acknowledgment of internal politics. Buyers know there is a reason their company has not solved this. The rep needs to surface that reason. If the rep skips it, the buyer assumes the rep does not understand the actual barrier and the deal will stall later.
Third, generic recap emails. Half of buyers told me they have stopped reading recap emails because they are template-driven and do not reflect what was discussed. A specific recap email that quotes the buyer's actual language and surfaces the consequences of inaction is rare and powerful. AI can help draft these but the rep has to inject the specifics.
Fourth, "happy ears." Josh Braun's data is that 73 percent of calls that reps mark as "great" never advance. The buyer was being polite. The rep heard interest. The deal sat in the pipeline for two quarters and died. We see the same pattern when we run pipeline cleanup engagements. Roughly a third of the pipeline is happy-ears deals from one bad discovery call.
Want us to score your last 20 discovery calls?
We do this as a free 30-minute audit. You send us recordings, we send you the talk ratios, the missed follow-ups, and the three changes that would lift your win rate the most.
Book a discovery audit →What to fix on Monday morning
If you are running a sales team and you want to do one thing on Monday, do this. Pull the last 20 closed-lost deals where the AE marked the discovery call as "great." Listen to the first 15 minutes of each call. Count the number of times the rep asked a real follow-up question after the buyer said something specific. Not "tell me more" but a follow-up that referenced what the buyer just said.
If the count is under 3 per call, you have a discovery problem and not a closing problem. The fix is not a new framework. The fix is teaching reps to listen, build a hypothesis, and run a peer-level diagnostic. The questions matter less than the listening behind them.
You can build the prep stack to make this faster and you can use AI for the post-call analysis. The actual call has to be a human paying attention to another human. That is still the most important 30 minutes in B2B sales.
FAQ
How long should a B2B discovery call be?
Chorus.ai data puts the sweet spot at 39 to 46 minutes. Anything under 30 minutes is correlated with a 28 percent lower advance rate. Anything over 60 minutes starts to feel like an interrogation unless the buyer is genuinely engaged. Book 45 minutes by default and let the buyer choose how much of it to use.
How many questions should a rep ask in discovery?
11 to 14 is the range Gong identified for top performers. Below 8, you have not done enough diagnostic work. Above 14, buyers report feeling interrogated and close rates drop. Quality of follow-up matters more than raw question count.
What is the right talk-to-listen ratio?
46 percent rep, 54 percent buyer is the closed-won range. Average reps run 68 percent rep talk time. If you are not actively measuring this with Gong or Chorus, you are guessing. Most reps think they listen more than they do.
Should I use BANT, MEDDIC, or SPIN?
The framework matters less than how you use it. Run a peer-level diagnostic conversation and let AI fill in MEDDPICC after the call. If you run the framework as a script during the call, you will hit the question wall and the buyer will check out. The framework is for the deal review, not the discovery conversation.
How do I know if a discovery call was actually good?
Three signals. The buyer talked uninterrupted for over 90 seconds at least once. You left the call with the names of at least 2 other stakeholders. You can write a recap email that quotes the buyer's specific language about the cost of doing nothing. If you have all three, you have a real opportunity. If you have one or zero, you have a polite call that will die in the pipeline.