Most sales organizations treat discovery as an event: a 30-minute call, a set of questions, a form in the CRM that turns green when the fields are filled. The data does not support that framing. Discovery is not a call. It is the accumulated state of what you can prove about a buyer's problem, their decision, and their risk of changing. The call is just where some of that proving happens.

The gap is wide. RAIN Group's buyer research found that only 26% of buyers say sellers are good at uncovering their needs, and only about a quarter rate sellers as good listeners. That is not a training problem at the margins. It is the majority verdict on the single stage that determines whether everything downstream is built on something real.

The working definition worth adopting: discovery is the process of establishing, with evidence rather than assertion, that a buyer has a problem worth solving, a path to deciding, and a tolerable risk in changing.

Notice what that definition excludes. It does not include qualifying the buyer for your convenience. It does not include demonstrating expertise. It does not include building rapport, though rapport is a useful by-product. Those are things sellers want. Discovery is about what has to be true for a purchase to happen at all, and about finding out whether it is true early enough that the answer is still cheap.

The reason this matters more now than it did five years ago is that the failure mode has shifted. Deals used to be lost to competitors. Increasingly they are lost to nothing. Matthew Dixon and Ted McKenna's analysis of 2.5 million recorded sales conversations, published as The JOLT Effect in 2022, attributed 40-60% of lost deals to customer indecision rather than to a preference for the status quo or a competitor's win. Their data showed 87% of deals carried medium to high levels of buyer indecision. Where indecision was medium, win rates ran around 30%. Where it was high, win rates collapsed to 6%.

Indecision is a discovery failure with a delay built in. The buyer's uncertainty was present during discovery. It simply was not surfaced, because the seller was asking about pain and budget rather than about what would make this person afraid to sign.

Conversation analytics has produced the most cited numbers in modern sales, and they deserve a careful reading rather than a repost.

Gong's analysis of 519,291 B2B sales conversations found that discovery calls in which the seller asked 11-14 questions had the highest success rates, with returns flattening beyond that range. The same dataset found that covering three to four distinct customer problems produced the highest likelihood of advancing the deal, and that going past four actually reduced the buyer's sense of urgency. Top performers spread their questions across the call. Weaker performers front-loaded them, working through a checklist in a way buyers experience as an interrogation. Talk-to-listen ratios for top performers sat near 46:54.

Then Gong's March 2025 analysis of 326,000 calls of at least ten minutes reported something that looks contradictory: sellers who closed averaged 15-16 questions, while sellers who lost asked closer to 20, and closed-won calls averaged 57% seller talk time against 62% for losses.

The honest reading is not that one study is wrong. It is that question count is a symptom, not a lever. Both datasets point the same direction on the thing that actually varies: sellers who lose ask more questions and talk more, because they are hunting rather than following a thread. A seller who has found the real problem asks fewer, better, deeper questions and then gets quiet. A seller who has not found it keeps asking, because each answer fails to close the loop. Coaching a rep to "ask 12 questions" optimizes the symptom. It produces a checklist, which is the exact behavior the data associates with losing.

The older research says the same thing with different instruments. Neil Rackham's work at Huthwaite, built on observation of more than 35,000 sales calls, found that in larger and more complex sales it was not situation questions or even problem questions that separated successful sellers, but implication questions (what the problem costs, and to whom) and need-payoff questions (what solving it would be worth). Situation questions serve the seller. Implication questions serve the buyer's own understanding of their position. Thirty-five thousand calls and half a million calls, separated by four decades of technology, converge on the point that depth beats coverage.

Here is a reusable structure the Sales Agent can apply to any opportunity. Think of discovery as a ledger with four columns. A deal is only as real as its emptiest column.

Column one: the gap. The distance between what is happening now and what the buyer needs to be happening, stated in the buyer's own numbers. Not "inefficient onboarding" but "onboarding takes 19 days against a 7-day commitment we made to enterprise customers." A gap you cannot state in the buyer's units is a gap you have not found. Gong's three-to-four-problem finding applies here: you are looking for a small number of gaps that matter, not an inventory.

Column two: the cost of inaction. What the gap costs per quarter, and, critically, who personally absorbs that cost. This is Rackham's implication question doing its work. A cost with no owner does not generate urgency, because organizations tolerate diffuse pain indefinitely. The question that fills this column is rarely "what does this cost you." It is closer to "who notices when this goes wrong, and what happens to them."

Column three: the decision path. Who decides, by what process, against what alternatives, on what timeline, and where the disagreement lives. Gartner's May 2025 survey of 632 B2B buyers, fielded in August and September 2024, found that 74% of buying teams exhibit unhealthy conflict during the decision process, meaning conflicting objectives, disagreement on the right action, or being overruled from outside the group. Buying groups that reached consensus were 2.5 times more likely to report a high-quality deal. The same research found that content tailored to the buying group raised consensus by about 20%, while content personalized to individuals reduced it by 59%, apparently by reinforcing each member's existing position. That is a strong argument against the instinct to run separate, bespoke conversations with each stakeholder. Discovery in a group context means finding the disagreement, not routing around it.

Column four: the risk of change. What could go wrong for this buyer, organizationally and personally, if they proceed. The JOLT research identified three drivers of indecision: choice overload (too many valid options to distinguish), information overload (more material than the buyer can reconcile), and expectations overload (doubt that the promised outcome is achievable here). Each of these is discoverable. None of them is discovered by a question about budget. Notably, the JOLT data found that when sellers responded to indecision by increasing fear-based messaging about the cost of doing nothing, deals became more likely to be lost, in 84% of such interactions. Pressing harder on column two is the standard response to a column four problem, and it makes things worse.

A ledger with columns one and two filled and columns three and four empty describes most of the pipeline that dies in "no decision." It looks healthy in the CRM because CRMs are built to record pain and budget.

There is a temptation to treat discovery as unchanged since Rackham. The buyer's information position has changed enough to matter.

Gartner's research on sense making, based on a survey of more than 1,000 B2B customers, found that sellers who helped customers make sense of conflicting, abundant information outperformed those who supplied more information or simply asserted a position, with roughly 80% of sense-making sellers closing high-quality, low-regret deals. The scarce good is not information. It is confidence in interpreting it.

That has intensified. Gartner's survey of 645 B2B buyers, conducted in August and September 2025 and released in May 2026, found that 45% of buyers used generative AI during a recent purchase, mostly for vendor and product research, and that buyers consulted an average of seven information sources. Buyers reported near-equal distrust of both channels: 51% were concerned about misleading information from generative AI, 49% about misleading information from sales reps. And yet 69% said they turn to sales reps to validate AI-generated insights, even while 67% said they would prefer a purchase process without a rep at all.

Read those together and the modern discovery job becomes clearer. The buyer arrives with material, some of it wrong, much of it generic, and no reliable way to tell which parts apply to their situation. They do not want a rep. They want a way to check their thinking. A discovery conversation that adds information competes with an infinite supply of it. A discovery conversation that helps a buyer test what they already believe against evidence from comparable situations is doing something the buyer cannot get elsewhere.

Practically: ask what they have already concluded and where they got it. "What have you already looked at, and what did it leave you unsure about" is a better opening than any situation question, because it surfaces the buyer's actual model and its weak points in one move.

Two evidence-backed points on developing this.

First, adaptiveness beats attitude. George Franke and Jeong-Eun Park's meta-analysis in the Journal of Marketing Research (2006), pooling 155 samples covering more than 31,000 salespeople, found that adaptive selling behavior improved self-rated, manager-rated, and objective performance, whereas customer orientation as a disposition improved only self-rated performance. Wanting to serve the customer is not the same skill as changing your approach based on what you are hearing. Discovery training that works on scripts and question banks builds the wrong thing. Training that builds the ability to abandon a plan mid-call builds the right one.

Second, follow-up questions are underrated by the people asking them. Karen Huang, Michael Yeomans, Alison Wood Brooks, Julia Minson and Francesca Gino, writing in the Journal of Personality and Social Psychology (2017, volume 113, issue 3), found across studies of live conversations that people who asked more questions, and particularly more follow-up questions, were better liked, and that participants systematically failed to anticipate this. The mechanism was perceived responsiveness. In a discovery context the practical implication is direct: the second question about the same answer is worth more than the first question about a new topic, and sellers consistently underweight it. This is the mechanism behind Gong's finding that top performers distribute questions rather than front-loading them. Distributed questions are usually follow-ups. Front-loaded questions are usually new topics.

Run discovery against the ledger, not against a form. Before the next step is scheduled, state out loud which of the four columns is thinnest and design the next conversation around it. A deal with a strong gap and an unknown decision path is not a strong deal that needs a proposal. It is a weak deal that needs a different conversation.

Discover with the group, not around it. Given that group-level relevance raised consensus while individual personalization cut it, the highest-value discovery move in a multi-stakeholder deal is usually to get two people who disagree into the same conversation and surface the disagreement rather than to run parallel private tracks.

Treat indecision as a discovery question, not a closing problem. When a deal stalls late, the missing work is almost always in column four, and the instinct to increase urgency actively damages it.

Write the recap. Sending back a written statement of the gap, the cost, the decision path, and the open risks, and asking the buyer to correct it, converts discovery from something you believe into something they have confirmed. It is also the cheapest possible test of whether you understood.

Finally, resist counting questions. The number is a readout, not a target. If a rep needs 20 questions, the problem is that question three did not go deep enough.