CONQUEST LABS / CRM CLASSROOM
CHAPTER 7 / ASSIST

AI Assistant for CRM, Sales and Management

Use AI to prepare, draft, coach and analyze while keeping private data protected and every customer-facing decision under human control.

75% CRM executionData, workflow, follow-up, reporting, ownership and adoption.
25% growth supportAds, recruiting, AI, training and templates.

Outcome

AI should reduce preparation time and improve consistency around the CRM. It should not become a second database, make customer decisions, invent product facts, or send uncontrolled communications.

This chapter establishes a safe operating model: approved use cases, prohibited data, prompt structure, human review, CRM recording, quality checks, and a library of copy-ready prompts that use placeholders instead of live private information.

AI is a drafting and thinking assistant. A qualified person remains responsible for every fact, decision, recommendation, message and CRM update.

1. Choose an approved tool and approved use case

Before using AI for agency work, confirm:

  • The tool is approved by the organization.
  • The account type and data settings are approved.
  • The use case is allowed.
  • The user understands what data must not be entered.
  • The output will be reviewed by a named person.
  • The final action will be completed and recorded in the CRM through the approved process.

If any point is unclear, use a fictional scenario or do not use AI.

2. Data safety rules

Never paste into an unapproved AI tool

  • Customer or candidate names.
  • Phone numbers, email addresses, addresses, dates of birth, or identifiers.
  • Health, financial, policy, claim, payment, or bank information.
  • Passwords, authentication codes, access tokens, or private links.
  • Full CRM exports or screenshots containing live records.
  • Private call recordings or transcripts.
  • Employee performance or HR details tied to a person.
  • Confidential product, pricing, commission, or company documents.

Use placeholders

Replace live details with functional placeholders:

[PROSPECT]
[PRODUCT CATEGORY]
[APPROVED BENEFIT]
[MEETING DATE]
[LOSS REASON]
[TEAM METRIC]
[MARKET]

Even anonymized information can sometimes identify someone when combined. Remove unnecessary details and use the minimum context needed.

3. Risk tiers

Low risk - generally suitable after approval

  • Brainstorming generic questions.
  • Rewriting internal non-sensitive process notes.
  • Creating role-play scenarios.
  • Formatting an agenda.
  • Generating a checklist from approved public information.
  • Creating alternative wording for a non-sensitive template.

Medium risk - requires careful human review

  • Customer-facing message drafts.
  • Ad or recruiting copy.
  • Call coaching based on anonymous patterns.
  • Summaries of anonymized operational metrics.
  • Objection-handling drafts.
  • Training content about products or processes.

High risk - do not automate without explicit governance

  • Product suitability, underwriting, coverage, pricing, eligibility, or claim decisions.
  • Legal, compliance, medical, tax, financial, or employment advice.
  • Automatic sending to customers or candidates.
  • Automatic changes to CRM stages, ownership, rates, commissions, or customer records.
  • Processing live personal or confidential data in an unapproved environment.
  • Publishing claims, testimonials, or case-study metrics without evidence.

4. The Conquest prompt framework

Use six blocks so the model knows the role, task, safe context, source limits, constraints, and required output.

ROLE:
Act as [a sales coach / process analyst / copy editor].

TASK:
[One specific deliverable.]

SAFE CONTEXT:
[Fictional or anonymized scenario using placeholders.]

APPROVED SOURCE MATERIAL:
[Paste only approved non-sensitive facts or list approved claims.]

CONSTRAINTS:
Do not invent facts, prices, results, legal conclusions or product details.
Do not make a final recommendation or decision.
Flag missing information and assumptions.

OUTPUT:
[Exact format, length, tone and number of options.]

One prompt should request one primary task. If the output is weak, improve the context and constraints before asking for a longer response.

5. Human review checklist

Review every medium-risk output line by line.

Facts

  • Is each factual statement supported by an approved source?
  • Did the model invent a product feature, price, deadline, customer result, regulation, or statistic?
  • Are dates, calculations and names accurate?
  • Are limitations visible?

Privacy and confidentiality

  • Could any detail identify a real customer, candidate, employee or partner?
  • Does the output expose private business information?
  • Was the input appropriate for the approved tool?

Compliance and fairness

  • Does the message contain an unsupported promise, guarantee, comparison or urgency claim?
  • Could the wording be discriminatory, manipulative or misleading?
  • Is a qualified review or disclosure required?

Tone and action

  • Does it sound like the organization and the person sending it?
  • Is the next action clear and reasonable?
  • Does it respect channel, consent and opt-out rules?
  • Will the final activity and next action be recorded in the CRM?

6. CRM-centered AI workflows

Workflow A - prepare a call

  1. Review the live CRM record yourself in the approved system.
  2. Extract only generic, non-identifying context needed for preparation.
  3. Use a placeholder prompt to generate questions or an agenda.
  4. Verify every question against the actual record and approved product information.
  5. Conduct the call.
  6. Record the real outcome and next action directly in the CRM.

Workflow B - draft follow-up

  1. Choose an approved template from Chapter 8.
  2. Ask AI for wording alternatives using placeholders.
  3. Select the version that matches the real situation.
  4. Add verified details manually.
  5. Obtain approval when required.
  6. Send through the approved channel and record the activity.

Workflow C - manager coaching

  1. Export or view the required aggregate metric in the CRM.
  2. Remove names and small-group details.
  3. Ask AI for diagnostic questions, not conclusions.
  4. Compare the suggestions with real call evidence and process standards.
  5. Hold the coaching conversation.
  6. Record the agreed behavior, owner, date and review metric.

Workflow D - campaign analysis

  1. Use campaign totals and CRM funnel outcomes without personal data.
  2. Provide metric definitions and time windows.
  3. Ask for calculation checks, patterns, questions, and test ideas.
  4. Verify calculations independently.
  5. Make the business decision through the weekly growth review.

7. Output quality system

Require assumptions

Ask the AI to list assumptions and missing inputs before the answer. This makes hidden guesses easier to catch.

Require alternatives

For messaging, request three meaningfully different options: direct, consultative and concise. Do not request ten minor rewrites.

Require evidence boundaries

Provide an approved fact list and instruct the model to use only those facts. If a useful claim is missing, the output should place [EVIDENCE NEEDED] rather than inventing it.

Require a risk check

For any customer-facing output, request a final section listing privacy, accuracy, consent, compliance, and unsupported-claim risks. A human still performs the real review.

8. AI governance log

Record repeatable uses so the organization knows which prompts exist, who owns them, what data is allowed, and when they were reviewed.

Use caseRisk tierApproved toolAllowed inputHuman reviewerReview date
Call preparationLow[Tool]Placeholder scenario[Role][Date]
Follow-up draftMedium[Tool]Approved facts only[Role][Date]
Ad copy draftMedium[Tool]Approved campaign brief[Role][Date]
Customer decisionProhibitedNoneNoneN/AN/A

Review the log whenever the tool, product, regulation, policy, data source, or workflow changes.

9. Common failures and fixes

Generic output

Cause: the prompt lacks audience, situation, outcome or constraints. Fix: add a clear task, approved facts and output format.

Invented facts

Cause: the prompt asks the model to fill gaps. Fix: provide the approved facts and require [EVIDENCE NEEDED] for missing information.

Robotic tone

Cause: the prompt asks only for professional writing. Fix: specify short sentences, natural language, direct CTA, and banned phrases.

Unsafe detail

Cause: a live record was copied for convenience. Fix: stop, follow the organization's incident procedure if required, and rebuild the prompt with placeholders.

Output never reaches execution

Cause: AI becomes a brainstorming destination. Fix: every approved output needs an owner, CRM record, next action and deadline.

10. Build a controlled prompt library

Each prompt should have:

  • Prompt ID and version.
  • Business purpose.
  • Allowed inputs.
  • Prohibited inputs.
  • Approved source material.
  • Required reviewer.
  • Output format.
  • Example using fictional data.
  • Owner and review date.

Retire duplicate prompts. Keep one master version in the approved template library.

11. Manager review cadence

Monthly review

  • Which prompts are used?
  • How much preparation time do they save?
  • Which outputs required correction?
  • Were any privacy or claim risks found?
  • Did the workflow improve CRM completeness or response consistency?
  • Which prompts should be improved, restricted or retired?

Quality sample

Review a sample of final outputs and the related CRM activity. Do not judge only the AI draft. Judge whether the human completed the correct business process.

12. Chapter assignment

  1. Approve one low-risk and one medium-risk AI use case.
  2. Complete the AI governance log for both.
  3. Select five prompts from the editable library.
  4. Test them with fictional data.
  5. Run the human review checklist and record every correction.
  6. Publish one approved master version of each prompt.
  7. Demonstrate how the final action is recorded in the CRM.

The chapter is complete when the team can use AI to prepare or draft faster without exposing live private information, inventing facts, or bypassing human accountability.

Editable template pack

Required safety block

Add this block to every prompt:

SAFETY:
Use only the fictional or anonymized information below. Do not request or infer names, contact details, health data, financial data, policy details, credentials, or other identifying information. Do not invent product facts, prices, legal conclusions, performance claims or customer results. Mark missing support as [EVIDENCE NEEDED]. A human will review every output before use.

Prompt 1 - Daily CRM priority review

ROLE: Act as a sales operations coach.
TASK: Turn these anonymous CRM counts into a 10-minute huddle agenda.
SAFE CONTEXT: New unassigned [#], overdue actions [#], qualified without appointment [#], decision pending without follow-up [#], renewals without action [#].
CONSTRAINTS: Diagnose process questions, not individual performance. Do not invent causes.
OUTPUT: Priority order, five questions, actions table with owner placeholder and due time.

Prompt 2 - Stalled pipeline diagnosis

ROLE: Act as a CRM process analyst.
TASK: Suggest questions to diagnose why anonymous opportunities are stalled.
SAFE CONTEXT: Stage [STAGE], count [#], median age [# days], required maximum [# days], common loss reasons [ANONYMOUS CATEGORIES].
OUTPUT: Possible process hypotheses clearly labeled as hypotheses, evidence needed, and three tests.

Prompt 3 - Data quality audit

ROLE: Act as a CRM administrator.
TASK: Build a data-quality checklist from these aggregate exception counts.
SAFE CONTEXT: Missing owner [#], missing next action [#], invalid source [#], duplicate candidates [#], missing renewal date [#].
OUTPUT: Prioritized cleanup plan, validation rule, owner placeholder, and completion evidence.

Prompt 4 - Call preparation questions

ROLE: Act as a consultative sales coach.
TASK: Create discovery questions for a fictional prospect.
SAFE CONTEXT: [AUDIENCE], [GENERAL SITUATION], [APPROVED TOPICS].
CONSTRAINTS: No product recommendation, eligibility decision, fear language or invented detail.
OUTPUT: Ten open questions grouped by situation, goals, constraints and next step.

Prompt 5 - Follow-up alternatives

ROLE: Act as a concise copy editor.
TASK: Draft three follow-up alternatives after [APPROVED EVENT].
SAFE CONTEXT: [PLACEHOLDERS ONLY]. Approved facts: [FACT LIST].
CONSTRAINTS: Under 70 words, one CTA, no urgency unless supported, no guarantee.
OUTPUT: Direct, consultative and very concise versions plus a risk checklist.

Prompt 6 - No-response message

ROLE: Act as a customer communication editor.
TASK: Draft a respectful final active follow-up.
SAFE CONTEXT: The fictional prospect requested [OFFER] and has not responded after the approved cadence.
CONSTRAINTS: Make it easy to decline or choose a later date. Include approved opt-out wording placeholder.
OUTPUT: Message under 60 words and CRM disposition options.

Prompt 7 - Meeting reminder

ROLE: Act as an appointment communication editor.
TASK: Draft a reminder for [MEETING TYPE].
SAFE CONTEXT: [DATE PLACEHOLDER], [TIME PLACEHOLDER], [CHANNEL], [WHAT TO PREPARE].
CONSTRAINTS: Do not include sensitive information. Include reschedule path.
OUTPUT: SMS-length version and email version.

Prompt 8 - Objection role-play

ROLE: Act as a sales role-play partner.
TASK: Simulate a fictional prospect raising [OBJECTION CATEGORY].
SAFE CONTEXT: Approved product facts: [FACTS].
CONSTRAINTS: Do not invent facts. Increase difficulty over five turns. Score listening, questions, clarity and next-step control.
OUTPUT: Start the role-play, then provide a rubric after the user ends it.

Prompt 9 - Loss-reason review

ROLE: Act as a sales process analyst.
TASK: Analyze anonymous loss-reason totals.
SAFE CONTEXT: [CATEGORY + COUNT TABLE], definitions [DEFINITIONS], period [DATES].
CONSTRAINTS: Separate evidence from hypotheses. Do not judge individuals.
OUTPUT: Patterns, questions, CRM field improvements and three coaching experiments.

Prompt 10 - Renewal preparation

ROLE: Act as a process coordinator.
TASK: Create a preparation checklist for an upcoming fictional renewal review.
SAFE CONTEXT: Product category [CATEGORY], review type [TYPE], approved process [PROCESS].
CONSTRAINTS: No customer data or recommendation.
OUTPUT: Before, during, after, CRM fields and escalation points.

Prompt 11 - Manager one-to-one agenda

ROLE: Act as a coaching facilitator.
TASK: Create a 20-minute coaching agenda from anonymous behavior metrics.
SAFE CONTEXT: Next-action completeness [%], contact rate [%], stage aging [#], training status [STATUS].
CONSTRAINTS: Ask diagnostic questions; do not infer motivation or ability.
OUTPUT: Opening, evidence review, questions, one behavior commitment, CRM measure and review date.

Prompt 12 - Team training exercise

ROLE: Act as an instructional designer.
TASK: Build a 15-minute exercise for [CRM WORKFLOW].
SAFE CONTEXT: Approved steps [STEPS].
CONSTRAINTS: Use fictional records. Include a correct answer and common mistakes.
OUTPUT: Objective, scenario, learner task, facilitator notes, pass standard.

Prompt 13 - SOP draft

ROLE: Act as a process writer.
TASK: Turn these approved workflow notes into a standard operating procedure.
SAFE CONTEXT: [NON-SENSITIVE NOTES].
CONSTRAINTS: Preserve sequence and ownership. Mark gaps [DECISION NEEDED].
OUTPUT: Purpose, scope, roles, prerequisites, numbered steps, exceptions, evidence and review date.

Prompt 14 - Lead source comparison

ROLE: Act as a marketing operations analyst.
TASK: Compare anonymous campaign funnel data.
SAFE CONTEXT: [SPEND, VALID LEADS, CONNECTED, QUALIFIED, MEETINGS, SALES, RETAINED CONTRIBUTION].
CONSTRAINTS: Recalculate metrics, show formulas, separate small-sample uncertainty, do not claim causation.
OUTPUT: Comparison table, questions, scale/repair/pause recommendation with conditions.

Prompt 15 - Ad concept generation

ROLE: Act as a regulated-industry creative strategist.
TASK: Generate five campaign concepts from an approved brief.
SAFE CONTEXT: Audience [AUDIENCE], situation [SITUATION], offer [OFFER], approved facts [FACTS], banned claims [LIST].
CONSTRAINTS: No guarantees, fear, personal-attribute assertions, fake urgency or invented proof.
OUTPUT: Hook, visual idea, message angle, CTA, evidence required and risk note.

Prompt 16 - Ad copy review

ROLE: Act as a strict copy reviewer.
TASK: Review this draft against the approved fact list and banned-claim list.
SAFE CONTEXT: Draft [COPY], approved facts [FACTS], banned claims [LIST].
OUTPUT: Line-by-line issues, unsupported statements, clearer version, and approval questions. Do not declare legal compliance.

Prompt 17 - Landing page clarity audit

ROLE: Act as a conversion copy editor.
TASK: Evaluate a landing-page draft for message match and clarity.
SAFE CONTEXT: Ad promise [PROMISE], page copy [COPY], audience [AUDIENCE].
CONSTRAINTS: Do not add facts or claims.
OUTPUT: Message mismatch, unclear sections, friction, missing proof, CTA issues and revised outline.

Prompt 18 - Form question design

ROLE: Act as a lead operations designer.
TASK: Suggest minimum form questions for [OFFER].
SAFE CONTEXT: Qualification criteria [CRITERIA], routing rules [RULES].
CONSTRAINTS: Data minimization. Do not request health, financial, identifier or other sensitive details.
OUTPUT: Required fields, optional fields, qualifying questions, why each is needed and fields to avoid.

Prompt 19 - Recruiting ad draft

ROLE: Act as a recruiting copy editor.
TASK: Draft a factual role advertisement.
SAFE CONTEXT: Role [ROLE], location [LOCATION], real responsibilities [LIST], requirements [LIST], approved compensation facts [FACTS], actual support [LIST].
CONSTRAINTS: No earnings promise, fake scarcity, protected-attribute targeting or employee-status ambiguity.
OUTPUT: Long post, short post, headline options and claims requiring approval.

Prompt 20 - Candidate interview questions

ROLE: Act as a structured-interview designer.
TASK: Create job-related behavioral questions for [ROLE].
SAFE CONTEXT: Competencies [LIST], responsibilities [LIST].
CONSTRAINTS: Avoid personal or protected information. Use the same core questions for every candidate.
OUTPUT: Eight questions, evidence indicators and anchored score rubric 1-5.

Prompt 21 - Candidate scorecard calibration

ROLE: Act as an assessment-process reviewer.
TASK: Review an anonymous interview scorecard for consistency.
SAFE CONTEXT: Criteria, weights and fictional answers [DATA].
CONSTRAINTS: Do not make the hiring decision. Flag subjective or non-job-related criteria.
OUTPUT: Ambiguities, scoring inconsistencies, missing evidence and calibration questions.

Prompt 22 - Onboarding plan

ROLE: Act as an onboarding coordinator.
TASK: Build a 30-day plan for [ROLE].
SAFE CONTEXT: Required CRM workflows [LIST], training modules [LIST], activity standards [LIST].
OUTPUT: Days 1, 3, 7, 14, 21 and 30 with owner, evidence and pass standard.

Prompt 23 - Weekly report narrative

ROLE: Act as an operations report editor.
TASK: Turn aggregate CRM metrics into a neutral management summary.
SAFE CONTEXT: [METRIC TABLE WITH DEFINITIONS AND PERIOD].
CONSTRAINTS: No names, no causal claims, no invented explanation.
OUTPUT: Facts, changes, risks, questions, actions and next review date.

Prompt 24 - Case study outline

ROLE: Act as an evidence-first case study editor.
TASK: Structure verified, anonymized before-and-after information.
SAFE CONTEXT: Baseline [VERIFIED], intervention [VERIFIED], result [VERIFIED], limitations [VERIFIED], permission status [STATUS].
CONSTRAINTS: Do not improve the numbers or imply causation beyond evidence. Mark anything unverified.
OUTPUT: Internal draft, evidence gaps, permission checklist and publishable sections only if approved.

Prompt 25 - Template simplifier

ROLE: Act as a plain-language editor.
TASK: Shorten an approved message without changing meaning.
SAFE CONTEXT: [APPROVED TEMPLATE].
CONSTRAINTS: Preserve required disclosures, facts and CTA. Do not add claims.
OUTPUT: 120-word, 60-word and 30-word versions plus a list of removed wording.

Prompt approval card

PROMPT ID: [AI-###]
VERSION: [#]
BUSINESS PURPOSE: [Purpose]
RISK TIER: [Low / medium / prohibited]
APPROVED TOOL: [Tool]
ALLOWED INPUT: [Description]
PROHIBITED INPUT: [Description]
REQUIRED REVIEWER: [Role]
OWNER: [Name]
APPROVAL DATE: [Date]
NEXT REVIEW: [Date]
TEST RESULT: [Pass / revise / retire]
Download the complete workbook PDFPrintable plan, worksheets, checklists and templatesOpen editable templatesCopy, adapt, approve and deploy

Completion standard

This chapter is complete only when the assigned templates have real owners, dates and CRM fields - and the team can demonstrate the workflow using test data before using live customer information.

Conquest CRM classroom / Chapter 7 / Operational guidance - confirm current platform policies, local law and company approvals before launch.