A discovery call produces one thing worth keeping: a record of what the buyer said, sorted by how well each statement can be checked. Rapport, timing, the shape of the offer and whether a proposal is worth writing are all downstream of that record. They are not separate achievements of the call; they are conclusions the record either supports or does not.
This is why the questions matter more than the conversation. A call that produces a warm impression and four unattributable sentences has produced nothing that survives the week. A call that produces thirteen answers, seven of which can be handed to a colleague as a document, has produced a working file. The difference is not charisma. It is whether the questions were built to leave evidence behind.
The eighteen questions in this piece are grouped by what each group is trying to establish: situation, pain, consequence, decision process, budget authority and timing. None of them asks for a commitment. None is designed to produce a yes. There are no closing questions in the set, because a closing question asks for a decision the call has not yet earned the information to support — and a buyer who says yes to a question they cannot answer with evidence has given the seller a fact that will not survive contact with their own organisation.
What the call has to establish
Six things, and they are ordered because the order is the logic. Situation comes before pain, because a question about what hurts is unanswerable to somebody who has not yet described how the work runs. Pain comes before consequence, because frequency has to be established before cost can be attached to it. Consequence comes before the decision process, because a cost nobody is accountable for does not become a decision. The decision process comes before budget authority, because authority is a property of a decision rather than of a person — one director may be able to commit $10,000 alone and need a committee for $100,000. Budget authority comes before timing, because a date is real only if the person who can commit the money has attached themselves to it.
Each group has an answer that closes it and an answer that leaves it open. Table 1 sets them out. The right-hand column is the more useful one: it is a list of the sentences that sound like answers and are not.
Table 1. The six groups of discovery questions, what each establishes, and the answer that closes or leaves it open. Constructed for this piece from the six groups the discipline requires; the fourth column follows the sales role’s rule that no figure enters a draft without a document behind it.
| Group | What the group establishes | The answer that closes it | The answer that leaves it open |
|---|---|---|---|
| Situation | How the work is done today: the steps, the people, the systems, the documents | A named artefact: the current form, a system export, the process map | “Mostly email” — an architecture described from memory |
| Pain | Which step fails, how often, and what failing looks like in practice | A dated case with a count behind it | “A bit of a nightmare” — a feeling with no instance |
| Consequence | What the failure costs and who carries it | The mechanism — rework, credit, penalty, a lost account — and the person who has to explain it | “It costs us a lot” — a cost with no owner and no number |
| Decision process | Who decides, what each of them needs to see, and what route a decision takes here | The committee, the criteria each member applies, a precedent | “I’m the decision-maker” — one person’s account of a process with three participants |
| Budget authority | Whose money, which line, whether it is allocated, and the threshold above which it moves | The budget line and its status, or the name of the person who holds it | “Budget isn’t an issue” — the answer that ends the enquiry it was meant to open |
| Timing | What has to be true before a start, and which date traces to a document | A renewal notice, an audit date, a board calendar, a season | “Q1 sometime” — a quarter, not a date |
Read the fourth column as a list of the sentences that most often close a discovery call early while seeming to answer the question. Each of them is a courtesy: the buyer is being helpful, giving the seller something to write down. The discipline is to write it down in the class it belongs to and then ask the next question in the group rather than treating the courtesy as a closed item.
The eighteen questions
Table 2 is the artefact. Three questions per group, eighteen in total, each with the evidence the answer has to leave behind. The set is written to be tailored rather than recited: the sector vocabulary changes, the systems named in the situation group change, and a question about a renewal notice becomes a question about a framework agreement in the public sector. What does not change is the structure, because the structure is what makes the record checkable afterwards.
Table 2. The question set: eighteen questions in six groups, with the evidence each answer has to leave behind.
| Group | The question | What the answer has to leave behind |
|---|---|---|
| Situation | Walk me through how this is done today, from the request to the result. | The steps in the buyer’s own order, and the artefact each step produces |
| Situation | Who touches it between the request and the result, and what does each of them do? | Names and roles, with the handoffs that fail most often marked |
| Situation | What does it run through, and which of those systems can export? | The systems named, and whether evidence can be produced from them |
| Pain | Where does it break most often, and what does breaking look like? | One case described end to end, in the buyer’s words |
| Pain | How many times in the last six months, and who noticed first? | A count the buyer owns, and the route by which it reached them |
| Pain | What have you already tried, and what happened to it? | The list of attempts and the reason each stopped — the graveyard you should not propose from |
| Consequence | When it breaks, what happens next — to the customer, to the money, to the person who explains it? | The mechanism of cost, in sequence |
| Consequence | Who is accountable when it goes wrong, and what does accountability cost them? | A named role, and what that role has to do when the failure lands |
| Consequence | If nothing changes for another twelve months, what does that look like? | Whether the status quo is tolerable, described by the buyer rather than argued by you |
| Decision process | Who else would have to agree, and what would each of them need to see? | The committee, and the criterion each member applies |
| Decision process | What has been bought here in the last two years that resembled this, and how did that decision run? | A precedent, with its route and its duration |
| Decision process | What would have to happen for this to be decided by nobody at all? | The no-decision path, named — the route that leaves no trace in a pipeline |
| Budget authority | Whose budget would this come from, and is it allocated this year? | The line, its holder, and its status |
| Budget authority | Above what amount does this become somebody else’s decision? | The threshold, and who sits above it |
| Budget authority | What did you spend on the closest equivalent, and what was it compared against? | The last comparable figure, and the alternatives it was judged against |
| Timing | What has to be true before you can start, and which of those is not true today? | The gating condition, isolated |
| Timing | Is there a date that is not yours — a renewal, an audit, a season — that fixes this? | An external clock, and the document that carries it |
| Timing | Which of the dates you have given me has a document behind it? | The document, by name, or the admission that there is none |
Two properties of the set are deliberate and worth stating, because they are what distinguish it from a script. The first is that every question is answerable by the wrong person. There is no question in the list that only the decision-maker can answer and no question that a well-briefed colleague cannot take away and return with a document. A script that depends on reaching one person dies the first time that person is on holiday. The second is that every question changes something. The answer to a question in the situation group alters which systems the offer has to fit; the answer to the timing group alters whether a proposal is written this month or after a renewal. A question whose answer cannot change either the draft or the decision does not belong in the set, however interesting the answer would be.
What is absent matters as much. No question asks the buyer to confirm that a problem exists so that the conversation may continue. None is phrased to produce agreement — there is no “wouldn’t it be useful if” construction in the table, because a question that can only be answered one way produces a sentence that cannot be used. And none asks for a decision. The set ends in the same place a good case ends: with the conditions under which the buyer would decide, not with a request that they decide now.
Worked example: eighteen questions, twenty-three asked, seven answers with a document
The example is constructed for this piece. Larkfield Facilities is a mid-sized facilities operator. On 6 October 2026 the operations director and the finance manager take a call of forty-seven minutes. Three calls are spent before the meeting: one to read the brief the buyer sent, one to read the account record from the previous year, and one to draft the question set in the sector’s vocabulary — the systems named in the situation group, the renewal date the account record already carries, and the two questions the previous year’s note left open.
On the call, eighteen questions are asked once and five follow-ups are asked of answers that were vague on the first pass. Twenty-three questions in forty-seven minutes is two minutes a question, which is the real budget on a call of that length: the forty-seven minutes hold twenty-three questions only if the answers stay short, and every answer that runs to five minutes spends two that belonged to another question.
Table 3 sorts the answers by class. The class is not a judgement about the buyer’s honesty; it is a statement about who else could check the answer without asking the buyer again.
Table 3. One call’s eighteen questions, sorted by whether a third party could check the answer. Constructed for this piece; the counts describe one call and are not a benchmark.
| Class of answer | Questions | What it is worth | From this call |
|---|---|---|---|
| Documentary — a document exists, or was promised by a date | 7 | The only class that survives a change of staff | The current intake form, the six-month complaint log, a contract with a renewal notice dated 31 March 2027 |
| Asserted, with a named source who could confirm | 4 | Worth one follow-up to that person | “The finance director signs anything above $25,000” — name given, not verified |
| Asserted, no source | 2 | A sentence in the record, not a fact | The number of times the process failed last quarter, from memory |
| Unanswered | 5 | Information about the question, not about the deal | Three in the decision process, two in timing |
The arithmetic is short. Seven answers of eighteen could be produced as a document, which is 38.9 per cent of the set on this call. Twenty-three questions at two minutes each is forty-seven minutes. The five unanswered questions are not a failure of the call; they are its most useful output, because they name the five things a proposal would otherwise have had to assume. Three of them sit in the decision process and two in timing, which is the pattern worth noticing: buyers answer readily about how the work runs and reluctantly about who decides and when, because the second pair of questions implicates other people in the room and in the organisation.
That leaves the record in a state a colleague can act on. Thirteen answers were given and five questions were not, and by the end of the following week all seven documentary answers have arrived as documents, because the promise of a document is a dated commitment that either arrives or does not. The five open questions are written into the next contact as questions rather than as gaps to be filled by inference, and the offer is drafted against the situation and pain groups while the decision-process and timing questions remain open — which is the correct order, because a proposal that is right about the work and wrong about the route to a decision is a document that gets read and not decided.
One detail from the example is worth separating, because it is the error the class column is built to catch. The two answers asserted without a source concern the same subject: how often the process fails. On the call, the operations director gives a number from memory and the finance manager, ten minutes later, gives a different one. Neither is dishonest. The class column records both as asserted, and the complaint log — a documentary answer in the same call — becomes the number the record uses. Without the class column the transcript would contain two numbers and no basis for choosing between them.
What a question set cannot establish
Five limits, and each is a thing a seller will assume the call has established when it has not.
An answer is a self-report. Nothing in the method verifies an answer at the moment it is given. The documentary class is a promise about evidence, not the evidence, and the week after the call is when the promise is tested. A set of questions, however well built, produces claims; a file becomes evidence only as the documents arrive.
The person answering may not know. Questions about another person’s authority, another department’s budget or a committee’s criteria are answered from outside that person’s knowledge, and the answer is hearsay in the technical sense: true as a report of what the speaker believes. The set handles this by naming the source in the class column rather than by asking the question twice, because asking the same question twice in one call produces a firmer version of the same belief.
Consequence questions change the room. Asking who has to explain a failure is, from the buyer’s side, an invitation to describe their own exposure to a stranger who is also selling something. There is a real cost to that, and it is the buyer’s to carry rather than the seller’s. The set therefore places the consequence questions after the situation and pain groups and does not ask them at all until the buyer has described the work and one case — a consequence question asked before the process is described is an interrogation.
The set says nothing about whether the deal will happen. It measures the state of a record, not the probability of an order. This publication holds no conversion rate from a discovery call to a proposal, no reply rate and no win rate, and it quotes none here. A seller who reads a thirteen-answer call as a good omen has added a belief to the record, and the belief is the part that turns out to be wrong.
Silence is an answer about the question, not about the deal. Five unanswered questions mean five things the proposal cannot claim. They do not mean the buyer is uninterested, and they do not mean the opportunity is weak — they mean the record is thin in two specific places, which is a more useful and much less dramatic statement.
The largest limit is structural. A question set can only ask about the process the buyer describes. Where the buyer describes a process that does not exist — a requirement that has never been written down, a budget that has never been allocated, a decision that has never been taken before — the answers will be consistent and the record will be about an intention rather than an operation. That is not a defect of the questions. It is the finding, and it is the one a seller is most tempted to overwrite with optimism.
What it costs
The calls here are model calls, priced at the measured Base rate of $0.00700, and the counts are planning assumptions stated where they are used.
One discovery call is five of them: three to read the brief, read the account record and draft the tailored set, one to write the call into the record in its classes, and one to turn the unanswered questions into the next contact. Five calls is $0.03500, or $0.00152 a question across the twenty-three asked. Set against the drafting it governs: a proposal is six calls, $0.04200, so the call that establishes whether a proposal is worth writing costs 83.3 per cent of the drafting it decides — and unlike the proposal, it is spent once and read by everyone who touches the account afterwards.
A month of thirty discovery calls is 150 calls, $1.05000; a year at that pace is 1,800 calls, $12.60000. At the metered ceiling of $0.00955 the same five calls cost $0.04775, which is $0.01275, or 36 per cent, more than $0.03500 — a ceiling is what an agent authorises, not what it spends. Re-running the two calls that write the call into the record on Qwen flash at its measured $0.00875 costs $0.01750, taking the five calls to $0.05250, which is 50 per cent more than $0.03500 for a second read on the one step where a misclassified answer is expensive.
The 0.50 USDC credited to a workspace as working capital is seventy-one calls at this rate: fourteen discovery calls at five calls each, $0.49000, with one call to spare. Fourteen calls is under half a month at the pace above, which is the point worth taking from the figure — discovery is a burst of work at the front of a pipeline, and the balance is spent where the work is front-loaded.
The decision
Choose the three answers that must exist as documents before a proposal is drafted, and ask those three in the first call. In most operations they are the same three: who can commit the money and above what amount, which date traces to a document, and the one requirement that would make the offer unsuitable if it were true. Everything else in the set can be asserted, corrected later or left open. Those three cannot, because each of them decides whether the proposal is worth writing at all.
Then classify every answer in the record as documentary, asserted with a named source, or unanswered, and let the unanswered questions travel into the next contact as questions. This is the discipline that makes the file useful to somebody who was not on the call: a colleague reading thirteen answers with their classes knows exactly which seven to wait for before drafting and which five to ask again.
The agent’s part in this is narrow. It reads what the account already holds — the brief, the thread, the previous note, the price list — drafts the tailored set in the buyer’s vocabulary, writes the call into the record with its classes, and puts the unanswered questions in front of a person. It cannot know what the buyer meant by an answer, cannot decide that an asserted number is good enough, and cannot take the decision the call exists to prepare. A record of what was said is the input to a judgement; it is not a substitute for the person making it.
Where this goes
- The pillar: x402.review/2026/10/31/commercial-operations/ — one record, two readers, and the approvals between them
- The role that asks them: studio.x402.press/roles/sales — the proposal, the follow-up and the mailbox it sends from
- What the answers become: the proposal structure that closes — the brief’s order, the terms beside each answer, the gaps marked
- Where the unanswered questions go next: following up a lead without losing it — a reason per touch and a stop rule written first
- The service it buys: x402.press/services/deepseek/ — the canonical price, seven chains
- The review: x402.review/2026/10/31/deepseek-review/ — the measured handshake and what it does not promise
Method note
The price was read from the live 402 challenge on Base on 31 October 2026, a Saturday and therefore outside DeepSeek’s weekday peak windows: 7000 atomic units under exact, which is $0.00700, and 9550 under upto at the max_tokens tested. Qwen flash’s $0.00875 was read from its own challenge in the same way. Call counts are planning assumptions, stated where they are used: three calls to tailor the set, one to write the call into the record, one to carry the open questions into the next contact.
Larkfield Facilities, the call of 6 October 2026, the answers and the class counts in Table 3 are constructed for this piece; they describe no real buyer and no real conversation. The six groups and the eighteen questions are the discipline as the sales role requires it — every figure in a draft traced to a document the client supplied — not the results of a study of calls. No conversion rate, reply rate, win rate or accuracy rate for classifying an answer is quoted anywhere here, because none has been measured.
x402.review is written and edited by N.P. Vincent. The publication is connected to x402.press, which sells the services reviewed here; that interest is disclosed on every piece that touches a service, and it is stated here rather than left to inference.
