Message Generator

Message Generator

Generates personalized outreach messages across all channels. Supports 3 tiers of personalization, from basic merge fields to fully individualized messages referencing specific signals and research. All tone, rules, and constraints are read from agency.config.json.

Prerequisites

• agency.config.json at repo root with outreach, services, case_studies, and agency sections
• Lead data with at minimum: name, company
• For Tier 2+: ICP segment data (industry, company size)
• For Tier 3: Output from company-researcher, person-researcher, cro-auditor, or signal-scanner

Phase 0: Intake

1. Read agency.config.json from the project root.
2. Extract:
   • agency.name -- the agency name for sign-offs and references
   • agency.domain -- for email domain
   • agency.founder -- for personal sign-offs
   • outreach.tone -- voice/style guide for all messages
   • outreach.banned_phrases -- phrases that must NEVER appear in any message
   • outreach.sign_off -- default sign-off line
   • outreach.cadence -- which message types to send on which day
   • outreach.daily_caps -- per-channel daily limits
   • services[] -- agency's service offerings for value prop
   • case_studies[] -- proof points to reference
3. Accept parameters:
   • message_type -- one of: cold-email, gig-proposal, community-reply, linkedin-note, instagram-dm, call-script, follow-up-email, breakup-email
   • personalization_tier -- 1, 2, or 3
   • lead -- the lead data object
   • research -- (Tier 3 only) output from researcher skills
   • framework -- (optional) override: PAS, BAB, AIDA, Signal-Proof-Ask
   • cadence_day -- (optional) which day in the cadence sequence

Phase 1: Framework Selection

If no framework is specified, auto-select based on message type and tier:

Framework definitions:

PAS (Problem-Agitate-Solve):

1. Name the problem the lead is facing
2. Amplify the cost of not solving it
3. Present the agency's service as the solution

BAB (Before-After-Bridge):

1. Describe their current state
2. Paint the better future state
3. Bridge: how the agency gets them there

AIDA (Attention-Interest-Desire-Action):

1. Hook with a bold statement or stat
2. Build interest with relevance
3. Create desire with proof/results
4. Clear CTA

Signal-Proof-Ask:

1. Reference the specific signal (their post, job listing, funding, etc.)
2. Prove credibility with a relevant case study result
3. Ask one clear question or propose one action

Phase 2: Personalization

Tier 1: Merge Fields Only

For bulk outreach. Replace placeholders with lead data:

• {{first_name}} -- lead's first name
• {{company}} -- company name
• {{agency.name}} -- agency name
• {{agency.founder}} -- founder name
• {{sign_off}} -- from config
• {{case_study.name}} -- case study client name
• {{case_study.result}} -- case study result line

The message body uses generic industry angles. No lead-specific research.

Tier 2: Segment-Based

Group leads by ICP segment and customize messaging:

• Industry angle: Reference industry-specific pain points
  • D2C fashion: "Product pages with size guides and lifestyle imagery convert 2x better"
  • Sports/fitness: "Sports brands need trust signals, athlete endorsements front and center"
  • Beauty/skincare: "Skincare brands live or die by before/after proof and ingredient transparency"
• Company size angle:
  • Early stage (<50): "You're growing fast, your store needs to keep up"
  • Mid-market (50-200): "Your team is stretched thin, outsourcing Shopify work frees your ecomm lead to focus on strategy"
• Market angle:
  • India: Reference local market dynamics, pricing in INR
  • US/UK/AU: Reference remote team benefits, timezone overlap, cost advantage

Tier 3: Per-Lead (Hyper-Personalized)

Requires research data from company-researcher, person-researcher, cro-auditor, or signal-scanner.

Personalization sources (use all that are available):

• Signal reference: "I saw your post on Reddit looking for a Shopify developer..."
• CRO finding: "I checked {{company}}'s product pages, noticed {{specific_issue}}..."
• Personal interest: "Your recent LinkedIn post about {{topic}} resonated..."
• Company news: "Congrats on the Series A, perfect time to invest in your storefront..."
• Tech stack note: "I see you're on {{theme}}, we've migrated several stores from that to..."

Build the opening line from the strongest personalization hook available.

Phase 3: Message Generation

Channel-specific rules:

cold-email:

• Subject line: max 50 characters, no spam trigger words, personalized when possible
• Body: max 150 words for Tier 1-2, max 200 words for Tier 3
• One CTA only (book a call, reply, or view case study)
• No attachments or links in first email (deliverability)
• P.S. line optional for Tier 3 (add a personal touch)

gig-proposal:

• Open with their specific requirement (mirror their language)
• Show relevant experience in 2-3 bullet points
• Include timeline and availability
• Max 250 words
• End with a question that invites dialogue

community-reply:

• Help first: answer their question genuinely
• Subtly mention experience: "We did something similar for a sports brand..."
• No hard pitch, no links in first reply
• Match the community's tone (casual for Reddit, professional for Shopify Community)
• Max 150 words

linkedin-note:

• Max 300 characters (LinkedIn connection note limit)
• One personalization point
• One reason to connect
• No pitch in the connect note

instagram-dm:

• Max 100 words
• Reference something specific from their feed (a product, a post, their aesthetic)
• Casual, conversational tone
• No links in first DM

call-script:

• Opening hook (15 seconds)
• Qualification questions (3 max)
• Value prop with case study reference
• Objection handling notes
• Close with meeting ask
• Total talk track: 2-3 minutes

follow-up-email:

• Reference the previous touchpoint (email, call, LinkedIn)
• Add new value (a relevant insight, article, or finding)
• Max 100 words
• Different CTA than the first email

breakup-email:

• Acknowledge they're busy
• Summarize value in one line
• "If timing isn't right, no worries"
• Leave the door open
• Max 75 words

Phase 4: Quality Check

Before returning, verify every generated message:

1. Banned phrase scan: Check against outreach.banned_phrases. If any match, rewrite the offending sentence.
2. Word count: Verify within channel limits.
3. CTA count: Exactly one CTA per message (except call scripts).
4. Tone check: Re-read against outreach.tone description. Rewrite if it sounds generic, desperate, or salesy.
5. Spam word check: No "free", "guaranteed", "act now", "limited time", "click here" in emails.

Phase 5: Output

Return structured JSON:

{
  "message_type": "cold-email",
  "personalization_tier": 3,
  "framework": "Signal-Proof-Ask",
  "subject_line": "Your product pages are leaving money on the table",
  "body": "The full message text...",
  "personalization_points": [
    "Referenced their Reddit post about needing Shopify help",
    "Mentioned their missing product reviews (from CRO audit)",
    "Used Kibi Sports case study (same industry)"
  ],
  "cta": "Reply to grab 15 min this week?",
  "word_count": 142,
  "channel_rules_met": true,
  "banned_phrases_clear": true,
  "sign_off": "Ekata | Plasho (plasho.com)",
  "cadence_day": 1,
  "cadence_next": "Day 2: LinkedIn connect"
}

For batch generation, return an array of message objects.

Phase 6: Log

If writing to CRM, use crm-writer to log: lead name, company, channel, message type, tier, subject line, cadence day, generated_at timestamp.

Example Usage

Trigger phrases:

• "Draft a cold email for this lead"
• "Write a LinkedIn note for [name]"
• "Generate outreach messages for these leads"
• "Write a gig proposal for this Freelancer post"
• "Create a follow-up email for [company]"
• "Generate Tier 3 messages using this research"

User: Write a cold email for the founder of BrandX, they just raised a seed round
Assistant: [reads config, selects Signal-Proof-Ask framework, generates Tier 3 email referencing funding news, includes case study, returns structured JSON]

User: Generate Tier 1 cold emails for these 10 leads
Assistant: [reads config, uses BAB framework, applies merge fields only, returns 10 message objects]