You will spend twenty minutes explaining. He will spend one sentence telling you the truth. Only one of those was worth the meeting, and it is not the one you prepared.
On the urge to establish credibility. You believe the explaining is groundwork. It is not. It is you managing your own discomfort at being unknown in the room, and the buyer is paying for it in the only currency he brought, which is time.
Nobody has ever bought because the category was explained well to them.
He says, offhand, twelve minutes in: we are trying to get onboarding time down this quarter.
There. That is the whole call. A budget exists because of that sentence, a person is measured on it, and it was said sideways while you were talking about something else. What did you do with it? You matched it to a feature. You said a number. You made it about the thing you sell, and the moment closed.
You did not miss the signal. You heard it, and the hearing is what made you start talking.
Ask what it costs him. Not what it costs in general, what it costs him, in money, this year. The question is uncomfortable and the discomfort is the entire reason it works. A problem no one can price is a preference, and preferences do not survive a budget review.
You skip it because you are afraid the answer is small. If it is small, you have learned something worth more than the deal.
Then ask what happens if he does nothing. Watch what happens to his face. That is the question, and you have been avoiding it for years.
The do-nothing question is the one most often skipped and the one that predicts most. We have never seen a deal close well where nobody could answer it. We have seen plenty of deals with enthusiasm, a champion, and a scheduled next step die on exactly that gap.
Now the thing that should be strange and somehow is not.
Ask five managers in your company what good discovery is. You will get five answers. They will overlap enough that nobody notices they differ, and differ enough that a rep coached by two of them is being pulled apart.
Ask for the document. There is no document. There is a methodology someone licensed, which is a qualification checklist wearing discovery's clothes. There is a deck from a kickoff two years ago. There is whatever the current manager believes, which came from whoever managed them.
And when that manager leaves, the next one does not replace the belief. They add to it, because replacing it would mean telling the team the last two years were wasted. Do this three times and you have sediment. Layers of half-systems with no through-line, and a team that has quietly learned the coaching changes every eighteen months, so none of it is worth taking seriously.
With nothing written down, what can you hand a new rep? Only the visible behaviour of whoever is currently good. Shadow her, she is excellent at this.
So he copies the pause. He copies the question order. He copies help me understand. And it is dead in his mouth, because the thing that makes it work was never in the behaviour.
Our theory, and we cannot prove it. What she is doing is holding a hypothesis about his situation and revising it live. The questions are downstream of the hypothesis. The pause is not technique, it is her thinking.
If that is right, then teaching the questions is teaching the exhaust rather than the engine, and every hour spent drilling question phrasing is an hour spent on the wrong end of the thing. It would also mean discovery is mostly a research skill wearing a conversation costume, which would reorganise where a sales team spends its money.
We are perhaps sixty percent convinced. Not enough to tell you to move a budget.
Two reps give a bad discovery call. The first does not know what good is. Nobody told him. He is doing his honest best with what he was given.
The second knows exactly what good is, could describe it to you over a drink, and abandoned all of it nine minutes in when the buyer pushed back and his pulse went up.
On the recording these are the same call. They are opposite problems, and the coaching that helps one entrenches the other.
Coach the second on the framework and you are teaching a man something he already knows, which reads as contempt. Coach the first on composure and you are asking him to stay calm while doing a thing he cannot do.
Both nod. Both agree to try harder. Neither improves, and you write in the review that he needs more reps.
The cheapest test anyone has: ask for the plan in writing before the call. One cannot write it. The other writes a good one and does not follow it.
It costs nothing and almost nobody does it, because asking for a written plan feels like bureaucracy right up until you understand it is a diagnostic.
People we respect say the framework already is the standard, and the problem is simply that reps do not follow it. We think that is wrong, and we hold it loosely.
A qualification framework tells you which facts to collect. It says nothing about how to run the twenty minutes in which a human being decides whether to tell you the truth. Those are different skills. Only one of them has ever been written down, and we have all been pretending that counts as both.
On the present moment
The tools got better this year, and cheaper, and they will do it again next year. Opus 5 reads a million tokens of context at five dollars a million in and twenty-five out. Sonnet 5 costs less than that. Cowork records the call. Something reads every conversation your team has, all of them, not the four a manager sampled.
This is genuinely new and worth taking seriously. It is also worth being precise about what it changed.
What was invisible became visible. That is the whole of it. Nothing about the conversation itself got easier.
Every individual call in that pattern sounds fine. Enthusiasm about a customer problem is not a performance issue. The monologue only exists as a pattern, at volume, across a quarter, which means for the entire history of selling it has been unobservable. A manager listening to four calls a month was never going to find it.
So the machine can now show you the shape of your own habit. It cannot make you ask the question you have been avoiding since 2019.
One team measuring their own recordings found somewhere between 60 and 80 percent of calls ending one-and-done while the calendar looked healthy throughout. That is their number from their recordings. Not a benchmark, not reproduced by us, and you should not repeat it as though it were an industry figure.
The part worth repeating is how they found out, which was by having something read all of it.
Kimi K3 shipped as open weights at the end of July: 2.8 trillion parameters, 104 billion active, native vision, a million tokens of context. Frontier-class capability you can download and run on your own infrastructure, which is a different kind of event from another price cut.
Read their own abstract before you get excited, though. It says plainly that K3 still trails the strongest proprietary models. The benchmark chart everyone is sharing is the vendor's chart, from the vendor's report, on the vendor's eval suite. That is not an accusation, it is just what a technical report is, and it is the part the commentary keeps dropping.
Then it reportedly sold out. Moonshot stopped taking new subscribers rather than quietly throttling people already paying, and existing users saw the service speed up the same day. Demand outran compute. Worth sitting with, because most planning in this category assumes capability rises while price falls, smoothly, forever. Sold out is not what that curve looks like.
Take the price of reading a call to zero. Every rep still has to sit in the silence after the hard question. Take context to infinity. Someone still has to decide whether this buyer is worth the quarter.
The tools compound. The thing they are pointed at does not. Every generation of sellers has been told their generation of tooling was the end of the craft, and every one of them found out it moved the work rather than removing it.
Here is what you can actually control. Not whether he buys. Not whether procurement moves in August. Not whether a competitor undercuts you the week before signature.
Whether you did the reading. Whether you asked the expensive question. Whether you told him the truth when the easy thing was available. That is the whole list, and it has not changed in forty years, and no release notes are coming for it.
Two things worth actually running
Here is what I know about {account}: {paste}
Do not write me questions yet.
First: tell me what I do not know that I would
need to know to run a useful discovery call.
Be specific about the gaps.
Then ask me which of those I can fill before the
call, and which I have to ask him live.My notes or the transcript: {paste}
Find the moment he mentioned an initiative or a
problem, and tell me what I said in the next
30 seconds.
Then: which of these did I actually ask, quoting
where, or NOT ASKED.
- what are you using today
- where does it break
- what does it cost you
- what changes if it is fixed
- what happens if you do nothing
Do not open with what went well.A model reading your transcript sees words. It does not see whether he trusted you. We have watched it score a mechanical five-question interrogation above a call where the rep asked three things clumsily and the buyer opened all the way up, and the second one closed.
We suspect call scoring quietly optimises reps toward interrogation, because what is measurable about a call is the questions asked, while what matters is whether the person opposite decided to be honest with you.
If that is true, a team scoring every call on question coverage gets rising scores and flat win rates, which is roughly what the last few years of enablement spend look like from outside. We would like to be wrong about this one.
The role-by-role version of where this sits in a week is in this guide to using Claude across a sales workflow, and the mechanics of turning any of it into something the team repeats are in packaging repeated work as reusable skills.
What we have not worked out
We are not going to tie these off. They are the arguments we are having with ourselves, and we would rather be corrected than consistent.
Is discovery a questioning skill at all, or a research skill that surfaces as questions?If it is research, then almost every enablement budget is allocated backwards, and the training industry is selling the visible half of the skill.
Does reviewing your own calls with a model make you better, or only more self-conscious on the next one?We do not know. It matters enormously, because self-review is the cheapest intervention available and we are recommending it above.
Can the thing that makes discovery good survive being written down?Every standard, including any we would write, is a compression of something a good rep does fluidly. Compression discards. We cannot tell whether what gets discarded is decorative or structural.
How much of what we call skill here is just novelty, and does it decay as buyers learn the pattern?Buyers have sat through a decade of bad discovery. Some now answer all five questions on autopilot, correctly, while telling you nothing true at all.
If the models keep getting cheaper and better at this rate, does any of the above hold in three years, or are we writing down the customs of a trade that is about to change shape?This is the one we argue about most. Cheap, patient, and never bored is a real advantage over an anxious human on a Thursday afternoon. It is also the claim every vendor in the category is making, which makes us suspicious of ourselves for finding it plausible.
If you have evidence on any of it, we want it. What changes our minds gets recorded in the log.