Sales Prospect

Prospect Analysis Orchestrator

Tier 0 Doctrine

• Tier: Hub (orchestrator that launches 5 parallel subagents)
• Authority: Owns the discovery briefing, composite scoring formula, and final report. Each subagent owns one scoring dimension only.
• Determinism first: All scoring uses declared weights, formulas, and grade bands. No LLM interpretation of scores. Company type detection uses signal tables, not judgment.
• No fabrication: Every data point must have a source URL or signal reference. Absence of data is scored (neutral 50), not invented.

IMO (Top-Level)

Constants

Variables

Workflow

BLOCK 1: Discovery & Briefing Compilation

Governed by: C&V

Constants: target_url schema (well-formed URL), interior page cap (5), company type signal table (6 categories), confidence levels (4 tiers).

Variables: target_url, company_name, company_type, industry_vertical, pages_found[], homepage_content, script_output, discovery_briefing, existing_reports{}.

Ingress: Receive target_url from user invocation. Validate: non-empty, well-formed URL. Reject immediately if validation fails.

Middle:

1. Fetch homepage via WebFetch. If unreachable, try alternate formats (www/non-www, http/https). If all fail, HALT with error message -- do not proceed on zero data.
2. Fetch interior pages (up to 5): About, Team, Pricing, Blog/Resources, Careers, Contact. Store URL, title, raw content, key data points for each accessible page.
3. Detect company type using signal table below. Scan homepage, careers, pricing, and investor mentions. Assign best-fit category; note runner-up if ambiguous.

4. Identify industry vertical from site terminology, customer logos, case study industries, job posting requirements, compliance mentions. Assign primary vertical (variable, not hardcoded list).
5. Run structured data extraction via python3 scripts/analyze_prospect.py --url <url> --output json. If unavailable or fails, log error, continue with manual extraction, set script_output = "unavailable".
6. Check for existing reports in cwd: COMPANY-RESEARCH.md, DECISION-MAKERS.md, LEAD-QUALIFICATION.md, COMPETITIVE-INTEL.md, OUTREACH-SEQUENCE.md. If present, flag for subagent consumption.
7. Compile discovery briefing object with all gathered data: URL, company name, company type, industry vertical, pages found, page contents, script output, initial signals.

Egress: discovery_briefing object ready for subagent distribution.

Go/No-Go: Homepage must be accessible (at least 1 page fetched). If zero pages accessible after retries, HALT. If homepage fetched but interior pages limited, proceed with reduced confidence.

BLOCK 2: Parallel Subagent Dispatch

Governed by: IMO

Constants: Subagent count (5), subagent skill file paths, subagent weights, subagent failure neutral score (50).

Variables: discovery_briefing, existing_reports{}, company_fit_score, contact_access_score, opportunity_quality_score, competitive_position_score, outreach_readiness_score, failed_subagents[].

Ingress: Receive compiled discovery_briefing from BLOCK 1. Validate: briefing contains at minimum target_url, company_name, and homepage_content.

Middle:

Launch ALL 5 subagents simultaneously (parallel, not sequential). Each receives the full discovery briefing plus any relevant existing report.

Error handling per subagent:

• If subagent fails or times out: log failure reason, assign neutral score (50), append to failed_subagents[].
• Continue with all available data regardless of individual failures.

Egress: Five dimension scores collected (actual or neutral-50 fallback). failed_subagents[] populated.

Go/No-Go: At least 3 of 5 subagents must return results. If fewer than 3 succeed, set confidence_level = "Very Low" and append prominent data-gap warning to report. Never halt on subagent failure alone -- always produce a report.

BLOCK 3: Score Synthesis & Action Plan

Governed by: CTB

Constants: Score formula (Sum(dimension * weight)), grade band table (5 bands), action plan tiers (3), first email word limit (100), confidence rubric (4 levels).

Variables: company_fit_score, contact_access_score, opportunity_quality_score, competitive_position_score, outreach_readiness_score, prospect_score, grade, confidence_level, failed_subagents[], action_plan, first_email.

Ingress: Receive five dimension scores from BLOCK 2. Validate: each score is integer 0-100.

Middle:

1. Calculate composite Prospect Score:

prospect_score = (
    company_fit_score        * 0.25 +
    contact_access_score     * 0.20 +
    opportunity_quality_score * 0.20 +
    competitive_position_score * 0.15 +
    outreach_readiness_score  * 0.20
)

2. Assign grade from band table:

3. Assess confidence level:

Reduce confidence by one level for each failed subagent.

4. Generate action plan (three tiers):

   • Immediate (24-48h): 3-5 specific outreach actions, LinkedIn connections, content to share, CRM notes.
   • Short-term (1-2 weeks): follow-up sequence, additional research, stakeholder engagement, competitive prep.
   • Long-term (1-3 months): relationship building, content nurture, event opportunities, referral approaches.

5. Craft first email using sales-strategy subagent findings: copy-paste ready, personalized with real data, under 100 words body, 2 subject line options (A/B), specific send target (name, title, company), clear low-friction CTA, recommended send timing and follow-up cadence.

Egress: prospect_score, grade, confidence_level, action_plan, first_email ready for report assembly.

Go/No-Go: Score must be computable (all 5 dimension scores present, even if neutral-50). Grade must map to exactly one band. If any computation fails, log error and report partial results with explanation.

BLOCK 4: Report Assembly & Output

Governed by: Circle

Constants: Output file name (PROSPECT-ANALYSIS.md), bar chart width (10 chars), Unicode fill character, terminal scorecard layout.

Variables: All variables from BLOCKs 1-3 assembled into final artifacts.

Ingress: Receive all computed values: discovery data, dimension scores, composite score, grade, confidence, action plan, first email, subagent detail sections. Validate: all required sections have content (even if partial).

Middle:

1. Write PROSPECT-ANALYSIS.md with these sections (structure defined in output template pointer):

   • Header: company name, URL, date, type, industry, score, grade, confidence.
   • Executive Summary: 3-5 paragraphs for sales leader. Lead with score/grade. Biggest opportunity, biggest risk, recommended approach, top decision maker, timing, go/no-go.
   • Prospect Snapshot table.
   • Score Breakdown table (5 dimensions with scores, weights, weighted values, key findings).
   • Company Profile (from sales-company subagent).
   • Decision Maker Map with buying committee table, text org chart, top 3 priority contacts (from sales-contacts subagent).
   • Opportunity Assessment with BANT scorecard and MEDDIC assessment (from sales-opportunity subagent).
   • Competitive Landscape with current solutions, switching costs, positioning angles (from sales-competitive subagent).
   • Outreach Strategy with framework, channels, personalization, objection prep (from sales-strategy subagent).
   • Prioritized Action Plan (3 tiers from BLOCK 3).
   • Ready-to-Send First Email with subject lines, body, CTA, timing, follow-up.

2. Render terminal scorecard with condensed output:

   • Company name, type, industry, URL.
   • Prospect Score with grade and label.
   • 5-dimension bar chart (10-char wide, Unicode block fill/empty).
   • Key decision maker.
   • Top 3 opportunities, top 3 risks.
   • Single most important next step.
   • File save confirmation.

3. Suggest follow-up commands: /sales outreach for full email sequence, /sales prep for meeting prep, /sales proposal for deal proposal.

Egress: PROSPECT-ANALYSIS.md written to cwd. Terminal scorecard displayed. No further mutations.

Go/No-Go: File must be writable to cwd. Terminal output must render without error. If file write fails, display full report in terminal as fallback.

Rules

1. Never launch subagents sequentially -- all 5 must run in parallel.
2. Never generate a report on zero fetched pages -- HALT if homepage is unreachable after retries.
3. Never invent data points -- absence is scored as neutral (50) or noted as "not found," never fabricated.
4. Never hardcode industry verticals in Constants -- industries are Variables detected at runtime.
5. Never let a single subagent failure halt the entire pipeline -- assign neutral score (50), reduce confidence, continue.
6. Never exceed 100 words in the first email body.
7. Never use LLM judgment to override the composite score formula or grade band assignment -- scores are mechanical.
8. Never modify the subagent weight allocation at runtime -- weights are Constants.
9. Never skip the discovery briefing -- every subagent must receive the full briefing object.
10. Never produce a report without a confidence assessment -- every report declares High, Medium, Low, or Very Low.

Reference Pointers