A reviewed one-page call plan that separates facts, hypotheses and discovery questions without inventing a buyer’s needs.
Before a first sales call, AI can quietly turn a guess into your opening pitch. Use it to organize what you know and identify what you need to learn. Here, a small service business prepares an introductory conversation. AI drafts a plan; you verify it and lead the call. Salesforce supports preparation and an agenda; HubSpot emphasizes open questions and listening. This original fictional scenario applies those principles without claiming improved sales results.
From account evidence to a human-reviewed call plan
A conceptual guide for preparing a sales call. Separate what the supplied account evidence shows from research assumptions, turn gaps into questions, and leave the plan for a person to review. Nothing is sent and no one is contacted.
Play workflow reveal (brief animation; close to stop)
To replay, close and reopen. Reduced-motion settings show a still workflow.
- Sort supplied account evidence
List only the facts present in the materials you were given, and note where each one came from. Separate what the materials explicitly support from assumptions, and keep each claim tied to its source.
- Turn unknowns into questions
Where a fact is missing, write an open question instead of guessing at a need or pain point. These questions become discovery prompts for the call.
- Human reviews the call plan
A person checks the draft agenda, questions and labels before anyone uses them. The plan makes no promises, adds no invented personalization and starts no outreach.
Motion provenance: CSS and initial sequence copy authored by actual Claude through the existing subscription on 14 October 2026; integrated and reviewed by AI & Companies. Conceptual workflow, not measured results. Evidence wording editorially tightened after authoring.
Download the workflow kit
Original prompts, briefs and practice materials. Adapt them to your own task and check the result.
Before you start
- An approved chat assistant and permission to process your material; the fictional pack needs no customer data or connections.
- An approved offer, the prospect’s enquiry and accepted meeting length.
- A human owner to verify the plan and record actual agreements.
Tools: ChatGPT, Claude. Use an assistant approved by your organization. This text-based first version does not require new integrations; account features and data policies still apply.
The workflow, step by step
Define a learning goal before a pitch
Choose one useful outcome for the conversation. Northline Office Support wants to understand how a booking change passes between two coordinators and whether its workflow review is relevant. A clear mismatch can be useful too. State what your business offers and what it cannot promise. In the example, Northline supplies a review and checklist; it does not supply scheduling software. Ask AI to preserve that boundary. Remove imaginary capabilities and unsupported promises.
Build a small, traceable source pack
Collect only the material needed for this call: your approved offer, the enquiry and a meeting note. Label each item with an ID, date and origin so a reviewer can trace the plan’s claims. For real work, remove unnecessary personal details and follow your organization’s data rules. Do not upload a whole CRM export to prepare for one conversation. Use the fictional F1–F4 pack below to practice. It establishes nothing about window-care businesses generally. For public sources, record the URL and date; positioning does not establish private priorities.
Separate known facts from possible explanations
Ask for a fact ledger before asking for questions. A fact should point to a source ID: Robin reports missed handovers, and two coordinators use a spreadsheet. The cause and commercial impact remain unknown. The enquiry records Robin’s account, not an independently investigated failure. Allow a small number of hypotheses only when they are explicitly labelled and useful to investigate. Unclear responsibility is one possibility, but it might be wrong. Ask a question that could reject it. Avoid turning a tool choice into a diagnosis: using a spreadsheet does not itself establish inefficiency.
Draft questions that leave space for another answer
Paste the prompt below and review its questions. Start with an ordinary process walkthrough. Then explore consequences, desired outcomes and how a decision would be made. Ask about constraints without presuming an approved budget or urgent deadline. Rewrite loaded questions. “How much revenue are you losing?” assumes losses. Asking what happens when a handover is missed allows Robin to describe inconvenience, rework, no material effect or something else. Likewise, ask who participates in a decision instead of labelling Robin the buyer. Keep five priority questions available, with optional follow-ups; the prospect’s answers should determine which question comes next.
Make a flexible agenda and check the offer fit
Use the accepted meeting length to propose a realistic sequence: agree the purpose, understand an example, discuss relevance and leave space to recap. The time allocation below is an invented planning aid, not a recommended universal ratio. Ask Robin what to change at the start. Link any offer discussion to something the prospect actually confirms. If Robin mainly wants new scheduling software, Northline should acknowledge that it does not sell it. AI can help rehearse that response without inventing a referral or pressing for a purchase. Do not interpret every hesitation as an objection to overcome. A budget limit or poor fit may be a sufficient reason to stop.
Rehearse the opening, then verify the whole page
Explain why you are meeting and invite correction. Practice following an unexpected answer. An AI role-play can help you try wording, but its fictional buyer responses do not validate demand, pricing or the real prospect’s likely objections. Compare every statement against F1–F4. Remove unsupported lost-booking counts, assumed authority and promises of savings. Keep the plan short enough to consult while listening. During the call, take notes under confirmed information, unresolved questions and agreed actions. Afterwards, distinguish a next step you suggested from one the prospect accepted; record an owner and date only when actually agreed.
Try it with a sample
Use this synthetic input to see the distinctions your output should preserve.
What a good result should preserve
ILLUSTRATIVE PLAN: authored example, not a call or model result.
Facts: Two coordinators share a spreadsheet; Robin reports missed handovers (F2). A 25-minute introduction is accepted; authority, budget, impact and timing are unknown (F3). Northline offers reviews/checklists, not software (F1).
Hypothesis: responsibilities may be unclear. Disconfirming question: “How is responsibility assigned when a booking changes?” A clear process may exist.
Proposed agenda: 3 minutes confirming purpose; 10 on a change walkthrough; 7 on impact, priorities and fit; 5 recapping/next steps. Invite changes.
Priority questions: 1. “Walk me through the most recent booking change, from request to handover.” 2. “When information misses the handover, what happens, if anything?” 3. “What would you want a review to clarify?” 4. “Who would assess this work, and how would that decision happen?” 5. “What timing, spending or operational constraints should we understand?”
Boundary: promise no software or savings. Conditional next step: scope a review if useful and accepted; owner/date remain unagreed. Acknowledge a mismatch if software replacement is the sole need.
A prompt you can adapt
Prepare a one-page discovery-call plan from the fictional source pack below. Use only the pack; do not browse, connect accounts, contact anyone or invent prospect facts. Return: (1) a fact ledger with F1–F4 references; (2) at most two explicitly labelled hypotheses, each with a neutral question that could disprove it; (3) a proposed 25-minute agenda, including room for the prospect to change it; (4) five prioritized, open questions covering the current process, impact, desired outcome, decision process and constraints; (5) an offer boundary and a conditional next step. Keep seller capabilities separate from prospect needs. Budget, purchasing authority, implementation timing and financial impact remain unknown. Do not manufacture urgency, software capabilities, testimonials, ROI or buyer answers. Do not prefill outcomes as if the call happened. If the offer is a poor fit, allow a respectful no. Mark all planned actions as proposed, not agreed. SOURCE PACK: FICTIONAL SOURCES: invented businesses/people; no account access. F1 | Offer, 14 October 2026: Northline Office Support sells workflow reviews and handover checklists, not software. Prices/delivery require scoping. No guaranteed savings. F2 | Enquiry, 13 October: Robin at Meadow Window Care says two coordinators share a booking spreadsheet; changes sometimes miss the handover. Robin asks what a review covers. F3 | Meeting note, 14 October: Robin accepted a 25-minute introduction. Authority, budget, impact and implementation timing are unknown. F4 | Seller goal: understand one booking-change handover and assess review fit together. No purchase/proposal agreed.
Check the result
What to keep in mind
This researched method uses fictional inputs and an illustrative plan, not recorded AI output, a real call or a performance test. Sources support discovery practice, not this prompt or invented offer. Role-play cannot establish customer intent. Use approved handling for real data; the seller remains responsible for claims and follow-up.
Sources & review notes
These sources support the tool context and review principles. The workflow, prompt, and synthetic example are original editorial material.
- Salesforce: What is a discovery call?
Primary vendor guidance on preparing an agenda, understanding a prospect and defining a follow-on action. No conversion or ROI claims are adopted.
- HubSpot: Discovery call questions that reveal real pain points
Primary sales publisher guidance on open-ended questions, active listening, decision process and checking fit. Article examples and performance claims are not reproduced.
Did this help you get the work done?
Tell us what worked, where you got stuck, or which step needs updating.