account-research-v2

Account Research v2

Generate a full-page, dark-mode HTML account intelligence dashboard. All reo data comes from a single composite call — reo_get_account_research — and web searches supplement it with external context. The skill's job is to compose everything together and render the dashboard.

Claude's job:

1. Call reo_get_account_research once to get all reo data (firmographic, scores, activities, developers, prospects, hiring)
2. Run 2–3 web searches for news, competitive context, and buyer KPIs
3. Write inference/analysis blocks based on the data
4. Render the HTML dashboard

Step 1 — Resolve Domain

If the user gives a company name but not a domain, infer the most likely domain (e.g. "Vercel" → "vercel.com"). If ambiguous, ask before proceeding.

Step 2 — Fetch Reo Data (single call)

Call reo_get_account_research. Use a 90-day signal window, max 5 contacts per section, and standard high-intent page labels:

reo_get_account_research(
  domain = "<company domain>",
  signal_lookback_days = 90,
  max_contacts = 5,
  high_intent_pages = ["Website: Pricing", "Form: Cloud PAYG", "Form: Self-hosted PAYG", "Form: Contact Sales", "Website: Case studies"],
  prospect_seniority_filter = ["VP", "Director", "Head", "C-Level"]
)

This is the only reo call. Everything in the dashboard is driven by this response:

Response Section	What It Contains
meta.org_id	UUID for the account
firmographic	Company name, industry, HQ, tech headcount, tech stack, customer_fit_score (ICP tier), developer_stage, developer_activity_score, reo_account_link
engagement	Intent level, event counts, high-intent hits, activity type breakdown
top_activities	Recent activity feed — use to render the Activity Summary section
developers	Up to 5 deanonymized developers with email, activity summary, notable pages
prospects	Up to 5 LinkedIn prospects with title, seniority, LinkedIn URL
hiring	Total open roles, breakdown by tech function, date range

If any section is empty, display "— Not available" in the dashboard rather than skipping the section.

Step 3 — Web Search Supplement (parallel)

Run these three searches in parallel — they supplement the reo data with external context:

3a. Competitive Context

Search for tools/vendors the company uses in your category. Look for job descriptions, LinkedIn posts, tech blog posts. For each competitor found: name, status (Active/Primary, Partial, Churned, Complementary), 2-sentence description. Include objection prep if a primary competitor is found.

3b. News & Trigger Events

Search for news from the last 30 days: funding, exec hires, product launches, outages. For each event: icon, title, 1-sentence sales implication, date. Omit section if nothing found.

3c. Buyer KPIs by Role

Use the contacts found in the reo response to identify buyer personas. For each, state their likely KPIs based on role + industry. Brief web search if needed for industry-specific KPIs.

Step 4 — ICP Fit (from composite)

Use firmographic.customer_fit_score (e.g. "STRONG", "MODERATE", "WEAK") directly. No custom scoring — the composite endpoint is the source of truth. Display the tier prominently and list the supporting signals from the firmographic/hiring data (industry, tech headcount, hiring shape, dev stage) alongside it as evidence.

Step 5 — Write Inference Blocks

These are where Claude adds real value beyond raw data. Be specific and opinionated — tie observations to actual people, actions, and dates from the data. No generic filler.

Activity Summary — "What this suggests"

2–4 sentences on what the activity pattern means. Consider: velocity (accelerating/stalling?), breadth (one person or many?), depth (deep in docs or just browsing?), intent signals (pricing/demo hits?). Connect to a likely sales scenario.

Developer Contacts — "What this tells us"

Synthesise what the collective developer behaviour signals. Examples of the kind of specificity expected:

• Multiple engineers with COPY_TEXT on CLI pages → active eval, likely POC
• Senior engineer + procurement visitor on pricing → evaluation committee forming
• Activity spread over 3+ months with no high-intent signals → slow eval, needs nudge

Suggested Next Steps

3–5 concrete, prioritised actions. Each includes:

• Action: specific to actual people and signals (e.g. "Send CLI quickstart to Tyler Schultz — copied 5 commands last week")
• Why: 1-sentence rationale tied to a signal
• Urgency: 🔴 This week / 🟡 This month / ⚪ When ready

Order by urgency. Highest-signal, lowest-effort first. If the account is cold, suggest prospecting actions based on hiring signals and contacts.

Step 6 — Render the HTML Dashboard

Generate a full dark-mode HTML dashboard. Self-contained (all CSS inline, no external deps).

Design Specifications

Color palette:

• Background: #1a1a1a (page), #242424 (card), #2d2d2d (nested card)
• Text primary: #e8e8e8; secondary: #9a9a9a; label: #6b6b6b (uppercase, 11px, 0.08em tracking)
• Accent green: #4ade80; blue: #60a5fa; orange: #f59e0b; red: #f87171; purple: #a78bfa
• Score bar track: #3a3a3a; ICP fill: #4ade80; Activity fill: #60a5fa
• Competitor status: Active=#f87171, Partial=#f59e0b, Churned=#4ade80, Complementary=#9a9a9a
• Hiring badge: dark green bg #14532d, green text #4ade80
• High-intent banner: #14532d bg, #4ade80 text
• Objection banner: #422006 bg, #f59e0b text
• Activity dots: blue #60a5fa (product), teal #2dd4bf (email), orange #f59e0b (event), grey #6b6b6b (outbound)

Typography: system-ui / sans-serif. Section labels: 11px uppercase 0.08em tracking. Body: 13–14px.

Layout: Single-column scroll. Cards with border-radius: 12px. Max-width 820px centered.

Dashboard Sections (in order)

1. Header — Company initials avatar, company name, tagline (industry · HQ · headcount), status tags (ICP tier from customer_fit_score, Intent level from engagement.intent_level), last updated timestamp. In the top-right of the header, render a "View account on reo →" link (anchor with target="_blank") pointing to firmographic.reo_account_link. Style it as a small pill-shaped link: #2d2d2d background, #60a5fa text, border-radius: 999px, 11px uppercase label with 0.08em tracking, subtle border 1px solid #3a3a3a, and hover state that lifts to #333. If reo_account_link is missing or empty, omit the link entirely (do not render a broken/placeholder button).

2. Overview — Row 1: Industry, Tech headcount (firmographic.tech_headcount), Founded Year, HQ. Row 2: Tech stack badges from firmographic.technology_names. Missing fields show "— Not available" in muted grey.

3. Scores — Two-panel row:

   • Left: ICP Fit — tier badge from customer_fit_score with supporting signals listed (industry match, tech headcount, hiring shape, dev stage)
   • Right: Activity Score — use firmographic.developer_activity_score (0–100) with a filled bar, plus Intent level badge from engagement.intent_level and Dev Stage from firmographic.developer_stage

4. Activity Summary — Chronological feed from top_activities[]: colored dot, actor name, activity type, page label, source, date. Below: "What this suggests" inference block. Cold badge if empty.

5. Developer Champions — From developers[]: avatar initials, name, email, title, activity summary (counts by type), notable pages, first/last seen. Below all cards: "What this tells us" inference block. Cold badge if none.

6. More Prospects — From prospects[]: avatar initials, name, title, location, LinkedIn URL (linked), outreach note.

7. Hiring Signals — hiring.total_open_roles as a big number, tech function badges with counts from hiring.by_tech_function[], insight banner inferring what the hiring shape means.

8. News & Trigger Events — From web search. Icon, title, implication, date. Omit if empty.

9. Competitive Context — From web search. Competitor cards with status. Objection prep if applicable.

10. Buyer KPIs by Role — 2×2 grid: role label, KPI list.

11. Suggested Next Steps — 3–5 action cards with urgency badges.

Error Handling

• Empty reo section → show "— Not available" in muted grey, never skip a section
• Thin web search results → show what's available, note it's limited
• Never hallucinate contacts, scores, or activities — only display what was returned
• Unresolvable domain → ask user to confirm before proceeding

Technical Notes

Response field reference — reo_get_account_research uses these field paths:

• meta.org_id — UUID for the account
• firmographic.company_name, .industry, .hq.city/.state/.country
• firmographic.tech_headcount — integer, use for "Tech Headcount"
• firmographic.technology_names — array of strings, render as badges
• firmographic.customer_fit_score — string like "STRONG", "MODERATE" — drives the ICP section
• firmographic.developer_stage — string like "HIGH", "LOW"
• firmographic.developer_activity_score — integer 0–100 — drives the Activity Score panel
• firmographic.reo_account_link — URL for the "View account on reo →" header link (omit if missing)
• engagement.intent_level — "High", "Medium", "Low" — surfaced as a tag in the header
• engagement.high_intent_hits[] — array of {page, date, actor}
• engagement.activity_type_breakdown — object with type→count
• top_activities[] — the activity feed for the Activity Summary section (actor, type, page, source, timestamp)
• developers[] — {name, email, title, linkedin, github, location, first_seen, last_seen, signal_strength, activity_score, activity_summary, notable_pages, reo_link}
• prospects[] — {name, title, seniority, tech_function[], location, linkedin, email}
• hiring.total_open_roles, .by_tech_function[].function/.count, .date_range

Activity signal hierarchy (for prioritising and narrating): FORM_CAPTURE > DEMO_LOGIN > COPY_TEXT/COPY_COMMAND > PAGE_VISIT > Identity

COPY_TEXT and COPY_COMMAND both indicate someone copied something — these are implementation signals stronger than PAGE_VISIT. Treat them as active evaluation indicators.