architecture

Sales AI Agents

AI agents for sales efficiency: prospect research, email drafting, meeting prep, CRM hygiene, and proposal generation. Includes off-the-shelf options, custom build approaches, and training examples.

Overview

Sales roles have high-frequency, repetitive tasks with clear efficiency capture potential.

Efficiency Potential: 8-12 hrs/week at Adoptive tier

Focus Areas:

  • Prospect research
  • Email drafting
  • Meeting preparation
  • CRM hygiene
  • Proposal generation

Agent Summary

Agent Efficiency Target Complexity Priority
45 min β†’ 5 min (89%) Low-Medium High
10 min β†’ 2 min (80%) Low High
25 min β†’ 5 min (80%) Low-Medium Medium
30 min β†’ 5 min/day (83%) Medium Medium
3 hrs β†’ 45 min (75%) Medium-High Medium

Prospect Research Assistant

Problem Statement

Sales reps spend 20-45 minutes researching each prospect before outreach. Information is scattered across LinkedIn, company websites, news, CRM history, and industry reports. This research is repetitive and doesn't require senior judgmentβ€”just aggregation.

Efficiency Target: 45 min β†’ 5 min per prospect (89% reduction)

Off-the-Shelf Options

Tool Capability Pricing Limitations
Clay Waterfall enrichment, AI research $149-800/mo Credit-based
Apollo.io Contact + company data $49-119/user/mo Less synthesis
Clearbit Real-time enrichment Enterprise API-focused
ChatGPT + Browse Web research with synthesis $20/mo Manual

Custom Build Approach

Architecture: See Pattern: Agent Chaining & Orchestration

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”     β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”     β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚  Trigger        β”‚     β”‚  Data Collection β”‚     β”‚  Synthesis      β”‚
β”‚  (CRM webhook)  │────►│  (Parallel APIs) │────►│  (LLM)          β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜     β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜     β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

Data Sources:

  • LinkedIn (Proxycurl, PhantomBuster)
  • Company website (Diffbot, scraping)
  • News (Google News API, NewsAPI)
  • CRM history (Salesforce, HubSpot)
  • Technographics (BuiltWith, Wappalyzer)
  • Funding (Crunchbase, PitchBook)

Build Options:

Approach Stack Effort See Guide
No-Code Clay + Zapier 2-4 hours -
Low-Code n8n + OpenAI 1-2 days Implementation Guide: Building Agents with n8n
Custom Python + LangChain 1-2 weeks Implementation Guide: Building Agents with LangChain

Training Example: Research Prompt

You are a sales research assistant preparing a briefing for a B2B sales rep.

Research target:
- Company: {{company_name}}
- Contact: {{contact_name}}, {{contact_title}}
- LinkedIn: {{linkedin_url}}
- Website: {{company_website}}

Create a 200-word briefing covering:

1. COMPANY SNAPSHOT
   - What they do (1 sentence)
   - Size and stage (employees, funding if available)
   - Recent news or changes (last 6 months)

2. CONTACT CONTEXT
   - Role and likely priorities
   - Career trajectory (previous roles relevant to current)
   - Shared connections or commonalities

3. CONVERSATION STARTERS
   - 2-3 specific talking points based on research
   - Avoid generic observations

4. POTENTIAL PAIN POINTS
   - Based on role + company stage, what problems might they face?

Format as scannable bullet points. Be specificβ€”generic insights are useless.

Email Draft Generator

Problem Statement

Reps write 30-50 emails daily. Each email takes 5-15 minutes when personalized properly. Most reps either spend too much time or send generic templates that don't convert.

Efficiency Target: 10 min β†’ 2 min per email (80% reduction)

Off-the-Shelf Options

Tool Capability Pricing
Lavender Real-time coaching $29-69/user/mo
Regie.ai Sequence generation Enterprise
Copy.ai Template-based $49/mo

Training Example: Email Prompt

You are writing a sales email for {{rep_name}} at {{company}}.

CONTEXT:
- Prospect: {{prospect_name}}, {{prospect_title}} at {{prospect_company}}
- Stage: {{sales_stage}} (cold outreach, follow-up, re-engagement)
- Previous interactions: {{interaction_history}}
- Research insights: {{prospect_research}}

OUR VALUE PROP:
{{value_proposition}}

TASK:
Write a {{email_type}} email that:
1. Opens with something specific to them (NOT "I hope this finds you well")
2. Connects their situation to our value prop in 1-2 sentences
3. Includes a clear, low-friction CTA
4. Stays under 150 words

TONE: {{tone}} (professional, casual, direct)

Do not use:
- "I hope this email finds you well"
- "I wanted to reach out"
- Exclamation points
- More than one question

Meeting Prep Generator

Problem Statement

Reps spend 15-30 minutes before each meeting reviewing CRM notes, email history, LinkedIn, and trying to remember context.

Efficiency Target: 25 min β†’ 5 min per meeting (80% reduction)

Off-the-Shelf Options

Tool Capability Pricing
Gong Call intelligence, meeting prep Enterprise
Dooly CRM-connected prep $35/user/mo
Fireflies Transcripts + summaries $19/user/mo

Training Example: Meeting Brief Prompt

Generate a 1-page meeting prep brief.

MEETING DETAILS:
- With: {{contact_name}}, {{contact_title}} at {{account_name}}
- Time: {{meeting_time}}
- Type: {{meeting_type}} (discovery, demo, negotiation, QBR)

CRM DATA: {{crm_deal_data}}
RECENT EMAILS: {{email_thread_summary}}
PREVIOUS CALL NOTES: {{call_notes}}

OUTPUT FORMAT:

## Meeting Brief: {{account_name}}

### Quick Context (30 sec read)
- [Deal status and key numbers]
- [What happened last interaction]
- [What needs to happen this meeting]

### Key Stakeholders
| Name | Role | Stance | Notes |

### Suggested Agenda
1. [Based on stage and context]

### Landmines to Avoid
- [Based on history]

CRM Hygiene Automator

Problem Statement

CRM data decays 30% annually. Reps hate data entry. Managers can't trust reports.

Efficiency Target: 30 min/day β†’ 5 min/day (83% reduction)

Off-the-Shelf Options

Tool Capability Pricing
Scratchpad Pipeline management $29/user/mo
Dooly Notes β†’ CRM fields $35/user/mo
People.ai Activity capture Enterprise

Training Example: Email-to-CRM Extraction

Analyze this sales email thread and extract structured CRM data.

EMAIL THREAD:
{{email_content}}

CURRENT CRM DATA:
- Deal Stage: {{current_stage}}
- Next Step: {{current_next_step}}

EXTRACT:

1. SENTIMENT: Positive / Neutral / Negative / At Risk
2. NEXT STEP: [specific action, not generic]
3. STAGE CHANGE: Yes / No (with reasoning)
4. KEY DATES MENTIONED: [date]: [context]
5. STAKEHOLDERS MENTIONED: [name]: [role if stated]
6. SUMMARY FOR CRM NOTES: [2-3 sentences]

Output as JSON.

Architecture Pattern: Consider ENT-FINE-TUNING for consistent extraction.


Proposal Generator

Problem Statement

Custom proposals take 2-4 hours. Most content is boilerplate with light customization.

Efficiency Target: 3 hrs β†’ 45 min per proposal (75% reduction)

Off-the-Shelf Options

Tool Capability Pricing
PandaDoc Template-based, e-sign $35-65/user/mo
Proposify Proposal builder $49/user/mo
Qwilr Interactive proposals $35/user/mo

Training Example: Custom Section Generator

Generate a custom "Why {{our_company}}" section for this proposal.

PROSPECT CONTEXT:
- Company: {{prospect_company}}
- Industry: {{industry}}
- Key Challenges: {{challenges}}
- Decision Criteria: {{criteria}}

OUR DIFFERENTIATORS:
{{differentiator_list}}

RELEVANT CASE STUDIES:
{{relevant_cases}}

TASK:
Write a 200-300 word section that:
1. Opens with their specific situation (not generic value prop)
2. Maps 2-3 differentiators to their stated challenges
3. Includes 1 specific proof point
4. Avoids superlatives and fluff

TONE: Confident but not boastful. Specific, not generic.

Chains

Research β†’ Email Chain

Combines and .

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”     β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”     β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚ New Lead        │────►│ Prospect        │────►│ Email Draft     β”‚
β”‚ (CRM trigger)   β”‚     β”‚ Researcher      β”‚     β”‚ Generator       β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜     β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜     β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
                               β”‚                        β”‚
                               β–Ό                        β–Ό
                        [Structured Brief]       [Ready-to-send Email]

Why chain vs single agent:

  • Research quality validated before email
  • Research cached for multiple email variants
  • Different models per stage (fast vs creative)
  • Failure isolation

See Pattern: Agent Chaining & Orchestration for implementation details.

Full Sales Cycle Chain

Research β†’ Outreach β†’ Meeting Prep β†’ Follow-up β†’ Proposal
    β”‚           β”‚           β”‚            β”‚           β”‚
    β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
                            β”‚
                    [Shared Prospect Memory]

Uses Pattern: Memory & State Management for persistent context.