Cold Call to CRM: Evidence-Led Prep, Practice & Logging

What this chain helps you accomplish

Prepare for a specific outbound call without inventing account facts, practice a realistic conversation, then turn what actually happened into a clean CRM record.

You will finish with an evidence-labeled account brief, a concise call plan with objection responses, role-play coaching, a prospect-confirmed fact ledger, and a CRM-ready call summary with accurately assigned outcomes and follow-up actions.

The key safeguard is a hard separation between research, practice, and real customer statements. A good role-play never becomes customer intelligence, and a possible pain point never becomes a CRM fact without confirmation. The prompts run in your own AI assistant; you review and manually save the result in your CRM.

When to use it

• Before calling a named prospect or account for the first time, particularly when you have limited but potentially useful research.
• When an SDR or AE needs a relevant opener and a low-pressure meeting ask rather than a generic pitch.
• When practicing common responses such as “I’m busy,” “Send information,” or “We already have a solution.”
• After a connected call, voicemail, no answer, or wrong-number outcome when the CRM record and next step need to reflect what actually occurred.
• When preparing for a calling block and a consistent way to review call outcomes matters.

Do not use this as a substitute for a full discovery process, as evidence that an opportunity is qualified, or as permission to auto-update customer records. For an unanswered call, skip prospect-discovery claims and log only the actual contact attempt. Use the prospect's explicit opt-out or do-not-contact direction according to company policy, without further persuasion.

Before you begin

Tools and permissions

Required: one text-chat AI assistant (ChatGPT, Claude, Gemini, or Copilot) that can accept pasted text and maintain a conversation; a way to place the call; and access to your organization's CRM for manual entry. Web browsing, CRM connectors, call recordings, and paid AI features are optional—not assumed. Use only AI services and sharing methods approved by your employer.

Do not paste confidential customer records, regulated data, personal phone numbers, or a recording/transcript into an AI tool unless your organization's policy and applicable recording/consent rules allow it. Sanitized notes are sufficient. Keep customer information separate from fictional practice content.

Input checklist: use these exact placeholders

• {{ACCOUNT_NAME}} — Required. The organization being called; from your prospect list or CRM. Example: Northbank Logistics (fictional). If unknown, obtain the company before researching.
• {{PROSPECT_NAME}} — Optional. Name from an approved contact record; example: Morgan Hale (fictional). If unavailable, use “not supplied,” not a guessed name.
• {{PROSPECT_ROLE}} — Required. Buyer job title or function; example: VP of IT. If unknown, use the best known function and flag the title as unconfirmed.
• {{INDUSTRY}} — Optional. Known segment; example: logistics. If uncertain, leave blank.
• {{OFFER}} — Required. What you sell, target use case, and a truthful one-sentence value statement. Example: software for tracking IT contract renewals; avoid invented outcomes.
• {{APPROVED_PROOF}} — Optional. Seller-approved product facts, case studies, and claims with sources. If missing, omit numbers and product-result promises.
• {{APPROVED_SOURCES}} — Optional. Dated links, permitted excerpts, first-party CRM notes, or other verifiable research. Include each source title, URL or document identifier, date and the specific claim it supports; “none” is valid.
• {{CALL_OBJECTIVE}} — Required. One realistic desired outcome, such as permission to book a 15-minute exploratory conversation; not a presumed commitment.
• {{SELLER_CONTEXT}} — Optional. Legitimate prior relationship or referral with source; if unknown, state “cold outreach, no prior contact confirmed.”
• {{DATA_HANDLING_RULES}} — Required. What may be pasted and whether web browsing/transcript processing is permitted. Example: sanitized notes only; no external browsing; no recording.
• {{CRM_SYSTEM}} — Optional until CRM preparation. Salesforce, HubSpot, or another CRM. If unknown, use “generic CRM draft.”
• {{CRM_OUTCOME_OPTIONS}} — Required for a final mapped disposition. Paste the exact allowed outcomes from your CRM; if unavailable, the AI must output “unmapped—seller selection required.”
• {{CALL_DATE_TIMEZONE}} — Required after the actual call. Exact date (YYYY-MM-DD) and time zone, e.g., 2026-11-05 / America/Chicago; if unknown, leave date unresolved.
• {{REAL_CALL_NOTES}} — Required after the actual call. Sanitized observations of what was actually said/done, with speaker tags and any precise commitments. If no connection, provide the actual no-answer/voicemail/wrong-number result instead.
• {{REACHED_STATUS}} — Required after the actual call. “Connected,” “No answer,” “Voicemail left,” “Wrong number,” or other observed result, from the caller's real activity record; do not assume a connection.
• {{ACTUAL_PARTICIPANTS}} — Optional after the call. Names/roles you actually reached and verified; use “identity not confirmed” for uncertainty, and never infer identity from caller ID alone.
• {{CRM_RECORD_IDS}} — Optional for Prompt 5. Exact related CRM record IDs/links that the seller already verified and is allowed to share; otherwise select the record manually in the CRM.
• {{CURRENT_CRM_STAGE}} — Optional for Prompt 5. Actual recorded opportunity stage from a verified CRM record; if none exists, say “not provided,” and never invent or promote one.

Prepare inputs and evidence

1. Gather only approved account information; label each claim with its source and date. Do not assume the assistant can open a URL or read CRM records.
2. Write a concrete call objective and accurate offer statement. Distinguish externally verified facts from seller-provided but unchecked details.
3. For the post-call phase, collect your company's permitted disposition choices and a simple notes template: call date, reached/not reached, participant(s), prospect statements, objection, commitments, owner, due date, and unresolved questions.
4. If no browsing or approved research exists, the chain still works with a role-based opener—but outputs must clearly state that account-specific facts are unavailable.

How to run the chain

1. Open a new chat in an approved text-based AI tool. Paste the completed Starting Context from Chain Steps, then run Prompt 1 without merging it with later prompts.
2. Review the PRE-CALL_EVIDENCE_BRIEF_v1. Correct sources, remove unsupported facts, and approve it by replying: APPROVED — PRE-CALL_EVIDENCE_BRIEF — v1.
3. Run Prompt 2 in the same chat, referencing that approved artifact. Review relevance and claims; approve CALL_PLAN_v1 using the same approval pattern.
4. Run Prompt 3 for a short interactive buyer simulation. Respond as yourself; type END ROLEPLAY to receive coaching. Retry the weakest moment once, then save PRACTICE_NOTES_v1. This is practice only; no simulated statement may enter CRM.
5. Conduct the actual call yourself. Use the approved plan as a guide, not a script to read mechanically. Immediately capture genuine notes or the actual no-connect outcome. Do not record or upload audio without necessary permission.
6. For stronger separation, open a fresh chat for Prompt 4 and paste only the actual call notes, call date/time zone, allowed CRM outcomes, and any necessary safe context. Never include the role-play transcript. If you remain in the first chat, explicitly tell the model to disregard all practice and pre-call hypotheses as evidence.
7. Review FACT_LEDGER_v1 against your raw notes. Fix missing, ambiguous, or misstated commitments. Approve it, then run Prompt 5 with the approved ledger and exact permitted CRM outcomes.
8. Check the two-sentence summary, disposition, owner, due dates, related contact, open questions, and stage treatment. Copy the CRM_LOG_v1 to the appropriate CRM activity manually; create the follow-up task only after verifying it. No prompt authorizes an AI to modify your CRM.

Approval, correction and handoff

For any generated artifact, use: APPROVED — [ARTIFACT_NAME] — v1. If edited, say REPLACE — [ARTIFACT_NAME] — v2, followed by the corrected text; only the latest explicitly approved version may be used. Reject unsupported content rather than asking the model to fill it in.

If starting a new chat, copy the Starting Context and only the approved artifacts required by the next step. For the post-call extraction, exclude simulated dialogue and avoid carrying unnecessary prospect research. If the source notes are incomplete, use UNKNOWN or REVIEW REQUIRED rather than creating a fact.

Completion means one manually reviewed CRM-ready log and a clearly owned next action—or an explicit “none/TBD” where the call did not establish one.