Conference prep

Conference Prep

Phase-gated workflow with two parallel tracks after Phase 2. The agenda document builds in the background while the lead pipeline runs sequentially with user checkpoints.

Phase 1 → Phase 2 → ┬─ Track A: Lead Pipeline (Phases 3→4→5) — sequential with checkpoints
                     └─ Track B: Agenda Document — background agent, no blocking
                     
                     → Phase 6: Final Assembly (both tracks merge)

Core Principle

This skill produces two things: actionable lead intelligence (who to talk to and how to reach them) and a curated agenda document that makes JP Montoya look like the most prepared person in the room. The prospect lists are for internal use. The agenda document is shareable with clients and prospects as a value-add.

Key Operating Rules

1. Apollo-first enrichment: Apollo is always the primary data source for contact-level enrichment. Only fall back to other sources when Apollo has gaps.
2. Processing ledger: Maintain an in-memory ledger tracking every contact through every enrichment and validation step. Structure: {contact_id: {apollo: done|skipped, crustdata: done|skipped|not_needed, hunter_finder: done|skipped|not_needed, hunter_verifier: done|skipped, email, phone, linkedin, email_status, list_assignment}}. Never run the same enrichment or verification call twice for the same contact.
3. No duplicate API calls: Before calling any enrichment or verification endpoint, check the ledger. If a contact already has a result from that source, skip it.
4. Output paths: ALL outputs go to ~/Documents/Claude/Conferences/[ConferenceName]-[Year]/. Create the folder structure on first run:

   ~/Documents/Claude/Conferences/[ConferenceName]-[Year]/
   ├── leads/              ← ICP CSVs + processing summary
   ├── agenda/             ← Curated agenda HTML
   └── logs/               ← Processing ledger, API call log, verification report
   

5. Email validation is mandatory: Every non-null email must pass Hunter.io email-verifier before inclusion in any output list. Only status = "valid" or status = "accept_all" emails are included. Do NOT send null or empty email strings to the verifier.
6. Three output lists: ICP Attendees, ICP Non-Attendees, ICP from Participating Companies. These are distinct populations with no overlap.
7. Phase-gated execution: NEVER proceed to the next phase without user confirmation. Each phase ends with a checkpoint summary and the question: "Ready for Phase [N+1]?" Track B (agenda) runs independently and does NOT block the lead pipeline checkpoints.

API Configuration

All API keys are stored in ~/.claude/.api-keys.json. Read this file at the start of execution. Expected keys:

Service	Key Name in JSON	Auth Method
Apollo	Use Apollo MCP tools directly	MCP connection
Crustdata	crustdata or crustdata_api_key	Authorization: Token {KEY} header
Hunter.io	hunter or hunter_api_key	Query param api_key={KEY}
Instantly	instantly or instantly_api_key	Authorization: Bearer {KEY} header

If a key is missing, ask the user before proceeding. Use the api-key-manager skill to store any newly provided keys.

---

Phase 1: Conference Confirmation & Setup

Goal: Confirm the right conference, verify all tools are connected, create project folder.

Steps

1. Ask for the conference name if not provided
2. Search the web for: official website, dates, location, agenda URL, attendee/sponsor list URL
3. Present what you found and confirm: "Is this the right conference? [Name], [Dates], [Location]"
4. Ask the user for their attendee data source (screenshot, PDF, CSV, URL, or "I don't have one yet")
5. Create the project output folder with subfolders:

   mkdir -p ~/Documents/Claude/Conferences/[ConferenceName]-[Year]/{leads,agenda,logs}
   

6. Read ~/.claude/.api-keys.json and verify all required API keys are present:
   • Apollo MCP tools: test with a simple search
   • Crustdata: verify key exists
   • Hunter.io: verify key exists
   • Instantly: verify key exists
7. Report any missing keys and resolve before proceeding

Checkpoint 1

Present to user:

✅ Phase 1 Complete — Conference Confirmed

Conference: [Name]
Dates: [Dates]
Location: [Venue, City]
Agenda URL: [URL]
Attendee source: [format provided]
Project folder: ~/Documents/Claude/Conferences/[ConferenceName]-[Year]/
  ├── leads/    ├── agenda/    └── logs/

API Status:
  Apollo MCP:  ✅ Connected
  Crustdata:   ✅ Key found
  Hunter.io:   ✅ Key found
  Instantly:   ✅ Key found

Ready for Phase 2? (Extract & classify attendees)

STOP and wait for user confirmation before proceeding.

---

Phase 2: Extract Attendees & Classify ICPs

Goal: Parse the attendee data, classify everyone against ICP criteria, produce a clean list of who matters. After this phase, Track B launches in the background.

Steps

2.1 — Extract Attendees

Handle whatever input format the user provides:

• Screenshot/Image: Read the image directly (Claude has vision). Extract every name, title, and company visible.
• PDF: Read the PDF and extract attendee data.
• CSV/Text list: Parse directly.
• URL: Fetch the page and extract attendee data.
• No list available: Skip extraction, note that Phase 3 will rely on company prospecting only.

Output a structured list: [{first_name, last_name, title, company, source: "attendee_list"}]

Also extract all company names from the conference sponsor list, exhibitor list, and attendee company names. Store these as the participating_companies list.

2.2 — ICP Classification

Read references/icp-criteria.md for the full ICP classification rules.

For each extracted attendee, classify against Solum Health's ICP criteria. Assign:

• ICP Match: Yes / No / Maybe
• ICP Tier: Primary / Secondary / Not ICP
• Role Category: Practice Owner, C-Suite, VP/Director Ops, Clinic Manager, Billing/RCM, Clinical Director, Intake/Access, PE/Investor, Consultant, Other
• Company Type: Behavioral Health, ABA, PT/OT/Speech, Mental Health, SUD Treatment, Multi-Specialty, Virtual/Telehealth, Residential+, Pediatric Therapy, IDD Services, PE-Backed Platform, Other Healthcare
• Reason: Brief explanation of why they are/aren't ICP

Initialize the processing ledger. Add each ICP/Maybe contact with list_assignment: "icp_attendees".

Checkpoint 2

Present to user:

✅ Phase 2 Complete — Attendees Extracted & Classified

Total attendees extracted: [N]
ICP matches (Yes): [N] — [list names + companies]
ICP matches (Maybe): [N] — [list names + companies]
Not ICP: [N] (skipped)

Companies identified for prospecting: [N]
Sponsor/exhibitor companies: [N]

Proceeding to Phase 3 will use Apollo API credits to enrich [X] contacts.
Estimated Apollo calls: [N people searches] + [N org enrichments]

⚡ Track B (Curated Agenda) will launch in the background — it doesn't
   block the lead pipeline. You'll see its results at Phase 6.

Ready for Phase 3? (Apollo enrichment + launch agenda in background)

STOP and wait for user confirmation before proceeding.

---

After Phase 2: Launch Track B in Background

Immediately after the user confirms Phase 2, launch Track B as a background agent. This runs independently while the user continues through Phases 3→4→5 (lead pipeline).

Track B does NOT block the lead pipeline. The user interacts with the lead pipeline checkpoints while Track B works silently. Track B results surface at Phase 6.

Track B: Curated Agenda Document (Background Agent)

Launch as a background Task agent with this full context:

• Conference name, dates, location, agenda URL (from Phase 1)
• Read references/brand-guide.md for voice and brand guidelines
• Read assets/prep-document-template.html for the HTML structure and styling

B.1 — Agenda Research

Search the web for the conference's full agenda:

• Session titles, descriptions, times, tracks, locations
• Speaker names, titles, companies
• Networking events, receptions, breakfasts
• Any pre-conference workshops or special events

Store all raw agenda data in a structured format for verification.

B.2 — Session Curation

Select 6-10 sessions most relevant to behavioral health practice owners/operators. Prioritize:

1. AI and automation in healthcare operations
2. Revenue cycle, billing, denial management
3. Payer strategy, value-based care
4. Workforce retention and burnout
5. Practice growth and scaling
6. Technology adoption and digital transformation
7. M&A and consolidation trends
8. Patient access and intake optimization

For each selected session, write a "Why attend" insight paragraph that:

• Explains why this session matters for a practice owner/operator
• References specific speakers and what makes their perspective valuable
• Connects to real operational impact (revenue, time, margin, growth)
• Sounds like a knowledgeable colleague giving advice, not a summary

B.3 — Data Verification

Run a systematic verification pass on all curated data:

1. Speaker verification: Verify each speaker's name, title, and company against source data. Flag any mismatches or inferred details.
2. Session time/location verification: Cross-reference every time, track, and location against raw agenda data.
3. Completeness check: Every session has title, speaker(s), time, location, "Why attend."
4. Consistency check: Chronological order, correct concurrent session notes, correct dates.
5. Content accuracy: Speaker descriptions match actual roles. No fabricated credentials.

B.4 — Humanization Pass

QA check on all "Why attend" text:

• No AI buzzwords (leverage, streamline, robust, comprehensive, cutting-edge, etc.)
• No dashes as punctuation (use commas, periods, or restructure)
• Vary sentence length
• Sound like a practitioner giving advice, not a content marketer
• No hedging language — be direct

B.5 — Generate Branded HTML Document

Generate an HTML file following the template structure in assets/prep-document-template.html. Include:

1. Header: Solum Health logo (CDN AVIF URL from brand guide) + conference title + "Executive Guide for Healthcare Leaders" subtitle + date/venue badge
2. Intro paragraph: 2-3 sentences on why this conference matters. Mention total sessions and how many were selected.
3. Day sections: Organized by day with colored day-label pills
4. Session cards: Grid layout with time/track column + content column (title, speakers, "Why attend" insight box)
5. Networking callouts: Teal-background bars for receptions, lunches, breaks
6. Concurrent session notes: When sessions overlap, help the reader choose
7. Footer: JP Montoya contact card (name, title, email, phone, website)

Brand colors:

• Navy primary: #011C40
• Accent blue: #468AF7
• Teal: #70D3C6 / #146055
• Accent BG: #E5DFF4
• Background: #F2F2F9
• Surface: #ffffff

Font: DM Sans (Google Fonts).

Save to: ~/Documents/Claude/Conferences/[ConferenceName]-[Year]/agenda/curated-agenda.html

Track B is complete when the HTML file is saved. Results surface at Phase 6.

---

Phase 3: Apollo Enrichment & Company Prospecting

Goal: Enrich all ICP contacts via Apollo, find C-level non-attendees at attendee companies, and prospect sponsor/exhibitor companies.

Track B (agenda) is running in the background while this phase executes.

Steps

3.1 — Apollo Contact Enrichment (ICPs from Phase 2)

For each ICP person (Yes or Maybe):

1. First try apollo_people_match with first_name, last_name, and company domain
2. If no match, try apollo_mixed_people_api_search with name + company name
3. Extract and store: email, phone, LinkedIn URL, current title, company size, revenue

For lists with 10+ people, use apollo_people_bulk_match for efficiency.

Also run apollo_organizations_enrich on each unique company for employee count, revenue, industry classification.

Update ledger: mark apollo: done for each contact.

3.2 — Post-Enrichment Reclassification

Re-evaluate all "Maybe" contacts using enriched company data:

• Solo practice (<5 providers) or outside behavioral health → reclassify to "Not ICP", remove
• Company confirms as a fit (5+ providers, behavioral health) → upgrade to ICP = Yes

3.3 — ICP Non-Attendees (C-levels at attendee companies)

For each unique company from the attendee list:

1. Run apollo_mixed_people_api_search filtered by ICP titles: "CEO", "COO", "CFO", "Owner", "Founder", "President", "VP Operations", "Director of Operations"
2. Cross-reference against the original attendee list. EXCLUDE anyone already on it.
3. Classify remaining people against ICP criteria.
4. ICP matches go into ledger with list_assignment: "icp_non_attendees".

3.4 — ICP from Participating Companies (sponsors/exhibitors)

For each sponsor/exhibitor company in behavioral health / therapy / healthcare ops (not already covered in 3.3):

1. Run apollo_mixed_companies_search to find the company
2. Run apollo_organizations_enrich for company details
3. Run apollo_mixed_people_api_search filtered by ICP titles
4. Classify and add to ledger with list_assignment: "icp_participating_companies"

Checkpoint 3

Present to user:

✅ Phase 3 Complete — Apollo Enrichment Done

ICP Attendees enriched: [N]
  - With email: [N]  |  Missing email: [N]
  - With phone: [N]  |  With LinkedIn: [N]
  - Reclassified from Maybe → Not ICP: [N] (removed)

ICP Non-Attendees found: [N] (C-levels at attendee companies)
  - With email: [N]  |  Missing email: [N]

ICP from Participating Companies: [N]
  - With email: [N]  |  Missing email: [N]

Total contacts needing waterfall enrichment (missing email/phone/LinkedIn): [N]
Estimated Crustdata calls: [N]
Estimated Hunter.io finder calls: [N]

🔄 Track B (Agenda) status: [running / complete]

Ready for Phase 4? (Waterfall enrichment + email validation)

STOP and wait for user confirmation before proceeding.

---

Phase 4: Waterfall Enrichment & Email Validation

Goal: Fill gaps from Apollo using Crustdata and Hunter.io, then validate ALL emails.

Steps

4.1 — Crustdata People Enrichment

For contacts missing email OR phone OR LinkedIn after Apollo:

POST https://api.crustdata.com/screener/person/search
Headers: Authorization: Token {CRUSTDATA_KEY}, Content-Type: application/json
Body: {
  "filters": [
    {"field": "person_name", "operator": "contains", "value": "{FULL_NAME}"},
    {"field": "company_name", "operator": "contains", "value": "{COMPANY}"}
  ],
  "limit": 1
}

Note: If endpoint returns 404, try /screener/people/search as fallback. Log which worked.

Skip if Apollo already found all three fields. Update ledger: mark crustdata: done.

4.2 — Hunter.io Email Finder

For contacts STILL missing email after Apollo AND Crustdata:

GET https://api.hunter.io/v2/email-finder?domain={COMPANY_DOMAIN}&first_name={FIRST}&last_name={LAST}&api_key={HUNTER_KEY}

Extract email from response.data.email if confidence >= 70. Update ledger: mark hunter_finder: done.

4.3 — Email Validation (ALL Emails)

Validate EVERY non-null email through Hunter.io email-verifier, regardless of source.

Rules:

• Do NOT send null, empty, or blank emails. Skip them.
• Check ledger: if already verified, do not verify again.
• For lists with 50+ contacts, add 1-second delay between calls (rate limiting).

GET https://api.hunter.io/v2/email-verifier?email={EMAIL}&api_key={HUNTER_KEY}

Keep only: status = "valid" OR status = "accept_all" Discard: status = "invalid", status = "disposable", status = "unknown" Null email contacts: Include in CSV with empty email field and note "No email found — target via LinkedIn or in-person." Do NOT upload to Instantly.

Checkpoint 4

Present to user:

✅ Phase 4 Complete — Enrichment & Validation Done

Waterfall enrichment results:
  Crustdata found data for: [N] contacts
  Hunter.io finder found emails for: [N] contacts

Email validation results:
  Total emails validated: [N]
  Valid / accept_all: [N] ✅
  Invalid / disposable / unknown: [N] ❌ (removed)
  No email found: [N] (kept for LinkedIn/in-person)

Emails by source:
  Apollo: [N]  |  Crustdata: [N]  |  Hunter.io: [N]

Final counts:
  ICP Attendees (with valid email): [N]
  ICP Non-Attendees (with valid email): [N]
  ICP Participating Companies (with valid email): [N]

🔄 Track B (Agenda) status: [running / complete]

Ready for Phase 5? (Generate CSVs + upload to Instantly)

STOP and wait for user confirmation before proceeding.

---

Phase 5: Output CSVs & Instantly Upload

Goal: Produce the three CSV files and upload validated leads to Instantly campaigns.

Steps

5.1 — Output Three CSVs

Save to ~/Documents/Claude/Conferences/[ConferenceName]-[Year]/leads/:

List 1: icp-attendees.csv Columns: First Name | Last Name | Title | Company | Company Type | ICP Tier | Role Category | Email | Email Status | Phone | LinkedIn | Company Size | Company Revenue | Enrichment Sources | Notes

List 2: icp-non-attendees.csv Same columns plus: Source Company

List 3: icp-participating-companies.csv Same columns plus: Source Company

Also save ~/Documents/Claude/Conferences/[ConferenceName]-[Year]/logs/processing-summary.txt with full breakdown.

5.2 — Create Instantly Campaigns

POST https://api.instantly.ai/api/v2/campaigns
Headers: Authorization: Bearer {INSTANTLY_KEY}, Content-Type: application/json
Body: { "name": "[ConferenceName] [Year] - [List Name]" }

Create one campaign per list (3 total). Record each campaign_id.

5.3 — Upload Leads

POST https://api.instantly.ai/api/v2/leads/add
Headers: Authorization: Bearer {INSTANTLY_KEY}, Content-Type: application/json
Body: {
  "campaign_id": "{CAMPAIGN_ID}",
  "skip_if_in_workspace": true,
  "skip_if_in_campaign": true,
  "leads": [...]
}

Max 1000 leads per request. Only upload contacts with validated emails. Lead fields: email, first_name, last_name, company_name, custom_variables (title, linkedin, phone, icp_tier, role_category, company_type, conference, list_source).

5.4 — Verify Upload via API

For each campaign:

GET https://api.instantly.ai/api/v2/campaigns/{campaign_id}
Headers: Authorization: Bearer {INSTANTLY_KEY}

Verify lead count matches what was uploaded.

Note on sequences: This skill does NOT create email sequences. Remind user: "Leads are loaded. Add your email sequences in Instantly before activating the campaigns."

Checkpoint 5

Present to user:

✅ Phase 5 Complete — CSVs Saved & Leads Uploaded to Instantly

Files saved:
  📄 [ConferenceName]_[Year]_ICP_Attendees.csv ([N] contacts)
  📄 [ConferenceName]_[Year]_ICP_Non_Attendees.csv ([N] contacts)
  📄 [ConferenceName]_[Year]_ICP_Participating_Companies.csv ([N] contacts)
  📄 [ConferenceName]_[Year]_Processing_Summary.txt

Instantly campaigns:
  "[ConferenceName] [Year] - ICP Attendees" → [N] leads uploaded ✅
  "[ConferenceName] [Year] - ICP Non-Attendees" → [N] leads uploaded ✅
  "[ConferenceName] [Year] - ICP Participating Companies" → [N] leads uploaded ✅

⚠️  Add email sequences in Instantly before activating campaigns.

🔄 Track B (Agenda) status: [complete ✅ / still running — will show at Phase 6]

Ready for Phase 6? (Final assembly — review agenda + full summary)

STOP and wait for user confirmation before proceeding.

---

Phase 6: Final Assembly & Review

Goal: Merge both tracks. Present the completed agenda for review and show the full summary of everything produced.

Steps

6.1 — Collect Track B Results

If Track B (background agent) has completed:

• Read the generated HTML file
• Present the curated session list for user review

If Track B is still running:

• Wait for it to complete (it should be done by now — agenda research is typically faster than the full enrichment pipeline)
• Report: "Waiting for agenda document to finish..."

6.2 — Agenda Review

Present the curated sessions to the user for review:

Curated sessions:
  Day 1:
    • [Time] — [Title] — [Speaker(s)]
      Why attend: [first sentence of insight]
    • [Time] — [Title] — [Speaker(s)]
      Why attend: [first sentence of insight]
  Day 2:
    ...

Verification results:
  Speaker data verified: [N]/[N] ✅
  Times/locations verified: [N]/[N] ✅
  Issues found & fixed: [N]

Ask: "Want to add, remove, or edit any sessions before finalizing?"

6.3 — Final Summary

✅ Phase 6 Complete — All Done!

📁 All files in: ~/Documents/Claude/Conferences/[ConferenceName]-[Year]/

Lead Intelligence:
  📄 ICP_Attendees.csv — [N] contacts
  📄 ICP_Non_Attendees.csv — [N] contacts
  📄 ICP_Participating_Companies.csv — [N] contacts
  📄 Processing_Summary.txt

Curated Agenda:
  📄 Curated_Agenda.html — [N] sessions from [Total] total

Instantly Campaigns:
  🚀 [ConferenceName] [Year] - ICP Attendees ([N] leads)
  🚀 [ConferenceName] [Year] - ICP Non-Attendees ([N] leads)
  🚀 [ConferenceName] [Year] - ICP Participating Companies ([N] leads)
  ⚠️  Add email sequences before activating

Pipeline summary:
  Total attendees extracted: [N]
  ICP matches: [N] (Primary: [N], Secondary: [N])
  Emails by source: Apollo [N] / Crustdata [N] / Hunter.io [N]
  Emails validated: [N] / [Total] ([%] pass rate)
  Contacts with no email: [N] (LinkedIn/in-person targets)

Want me to adjust anything? I can re-score ICPs, add/remove sessions,
change the document style, or re-run enrichment for specific contacts.

---

Important Notes

• Track B runs in background: The agenda document (research, curation, verification, HTML generation) launches as a background agent after Phase 2. It does NOT block the lead pipeline. The user continues through Phases 3→4→5 while the agenda builds itself. Results merge at Phase 6.
• Phase-gated execution is critical: Each lead pipeline phase (3, 4, 5) MUST complete and get user approval before the next begins. Track B is exempt from gating — it runs autonomously in the background.
• Apollo is primary: Always try Apollo first. Only use Crustdata and Hunter.io finder as fallbacks for missing data.
• Processing ledger prevents waste: Every contact is tracked through every step. Never call the same API for the same contact twice.
• Hunter.io has two roles: (1) Email FINDER as a waterfall fallback. (2) Email VERIFIER for ALL emails. Different endpoints. Do not confuse them.
• Instantly API: Use POST /api/v2/leads/add (NOT /leads). Max 1000 leads per request.
• API error handling: If an API call fails, retry once after 5 seconds. If it fails again, mark the contact as enrichment_failed in the ledger and continue. Report all failures.
• Rate limiting: Hunter.io — add 1-second delay between calls for 50+ contacts. Report expected credit usage before starting Phase 4 so user can approve.
• Empty-set handling: If a list has zero validated leads, skip Instantly campaign creation for that list. Report it but don't error out.
• Privacy: Attendee data and enriched lists are internal only. The curated agenda is the shareable, client-facing piece.
• Data verification is not optional: Track B verification (B.3) must run before HTML generation (B.5).

File Dependencies

• Logo: assets/solum_logo.png or CDN: https://cdn.prod.website-files.com/66ccf770fcc17846279d79cd/6748cf9b5bac5ff27b9e6bff_solumnewlogo.avif
• HTML Template: assets/prep-document-template.html (reference for layout/styling)
• ICP Criteria: references/icp-criteria.md
• Brand Guide: references/brand-guide.md
• API Keys: ~/.claude/.api-keys.json (Crustdata, Hunter.io, Instantly)
• Apollo: Apollo MCP tools (must be connected)
• Crustdata: HTTP API at https://api.crustdata.com
• Hunter.io: HTTP API at https://api.hunter.io/v2 (email-finder + email-verifier)
• Instantly: HTTP API at https://api.instantly.ai/api/v2
• Web access: WebSearch + WebFetch for agenda research

Output

Default output directory: ~/Documents/Claude/Conferences/[ConferenceName]-[Year]/

[ConferenceName]-[Year]/
├── leads/
│   ├── icp-attendees.csv
│   ├── icp-non-attendees.csv
│   └── icp-participating-companies.csv
├── agenda/
│   └── curated-agenda.html
└── logs/
    └── processing-summary.txt

Instantly campaigns created:

• "[ConferenceName] [Year] - ICP Attendees"
• "[ConferenceName] [Year] - ICP Non-Attendees"
• "[ConferenceName] [Year] - ICP Participating Companies"