Find verified prospects
with AI-powered research

Click the 🔑 Config button in the top-right nav bar to open the API Configuration panel. The app supports multiple provider modes — pick one with the radio buttons at the top. Your choice is saved in this browser and shown as a badge (GPT / ANT / DBX) in the top nav.

Uses the Anthropic API directly with Claude Sonnet 4.6's native web search and web fetch tools. The AI runs its own live searches and reads actual 10-K filings, press releases, and leadership pages — no SerpAPI needed. Full page content (not just snippets) grounds every card, which enables Verified contact upgrades and fetched-page trigger proof. This is the default and recommended mode.

Uses the OpenAI Responses API with gpt-5.4 and its native web_search tool. Runs the same shared pipeline as Anthropic mode — same prompts, validators, budget rules, and drop logic. One structural difference: OpenAI exposes search snippets and citation URLs but not full page text, so cards are grounded on snippets — contacts top out at Needs Review (never auto-Verified) and the Research Log shows model-reported URLs instead of fetched pages. Select in API Configuration — requires CHATGPT_API_KEY.

No segment is selected by default. Choose one segment for a focused run — this produces the highest-quality results. Turn on Advanced Multi-Segment only when you intentionally want to search across 2–3 segments. All 8 available: E-Commerce, FinServ, Healthcare, Education, Travel, Auto, Real Estate, SaaS/B2B.

Budget estimates use verifiable data sources in this priority order: SEC filings/10-K disclosures, IPEDS (education), NCUA (credit unions), press-reported revenue with industry ratios, location count × benchmarks, and Crunchbase funding data. Each prospect shows a Budget Source with full methodology, data year, and a confidence badge (High / Medium / Low) so you know how reliable the estimate is.

Start with 5 for fast runs. Max 30. Each prospect goes through a 4-step pipeline: discovery → deep research → contact verification → personalized email. Typical run: ~3–5 min for 5 prospects.

Auto lets Claude pick the best angle per company based on their segment and pain points. Or force a specific angle across all prospects: Attribution Pain, Performance Guarantee, or Competitive Intel.

Optional. AiMG's full product brief (Atrilyx, proof points, pricing tiers) is built into every run. Use this to emphasize a specific case study, vertical, or angle — e.g. "Focus on call attribution" or "Lead with the senior living CPL case study."

Companies come from AiMG's curated database pool (~60K US mid-market companies built from SEC EDGAR, FDIC/NCUA, IPEDS, and industry sources) — not from a live web guess, so a fabricated company can never appear. The draw is ranked: confirmed advertisers lead — companies whose own SEC filings disclose an annual advertising expense in the shippable band are pulled first. Companies you've already seen in the last 90 days and your excluded companies are skipped automatically.

Each company first passes a cheap revenue-resolution gate (one focused web search resolves the revenue figure and drops prohibited industries or pure-B2B non-advertisers early). Survivors get full enrichment: targeted searches for SEC filings, leadership, funding, and industry-specific data (IPEDS for education, NCUA for credit unions). When a company's own SEC filing discloses its advertising spend, that disclosed figure becomes the budget basis directly — stronger than any revenue-ratio model. If no credible revenue exists, the card ships honestly as a Low-confidence firmographic estimate instead of inventing a number.

A dedicated verification step searches for each company's current leadership, fetches their actual website pages (about/team/leadership), and confirms who currently holds the role. Former executives are filtered out. If a contact can't be verified, the field is left empty rather than showing a wrong name.

The AI writes a tight 60–90 word cold email per prospect: a sub-6-word lowercase subject, three short lines (trigger reference → what Atrilyx does → 15-minute ask), and a first-name-only sign-off. Each email references that company's specific trigger, pain point, and Atrilyx hook — never templated.

Each card shows the company name, verified contact (name + title), website, segment badge, priority (A or B), and estimated monthly ad budget. Click to expand and see the full research details and generated email.

Full methodology behind the ad budget estimate — data year, calculation, and source. Header shows source type (SEC Filing, IPEDS, NCUA Call Report, Press Release, Location Benchmarks, Industry Analysis) and a confidence badge (High / Medium / Low). Quality flags to watch for: (1) A BUDGET: NEEDS REVIEW chip or orange internal warning means the budget source needs manual verification — for example, no primary-source pages were fetched or the formula/source figure is incomplete. Verify manually before citing. (2) High confidence automatically downgrades to Medium if the source is not primary (10-K, SEC, press release, earnings, NCUA, IPEDS, or major business press). (3) Aggregator-only sources (ZoomInfo, RocketReach, Growjo, LeadIQ, Datanyze, Owler) automatically force Low confidence — the figures are modeled estimates, not disclosed. Prioritize High and Medium confidence prospects; treat Low cards as leads to verify, not facts to cite.

A dated business event from the last 12 months that gives the sender a "why now" reason to reach out — funding round, exec change, product launch, retail expansion, partnership, acquisition. Surfaced via live AI research from Tier-1 sources (PRNewswire, BusinessWire, major business press, trade outlets). Trigger chips to watch: (1) TRIGGER: VERIFIED — source URL confirmed + direct event verb + fresh date. (2) TRIGGER: NEEDS REVIEW — source exists but event is a leadership hire, LinkedIn-only, or date is stale (2+ years). Verify before sending. (3) TRIGGER: WEAK — evergreen operating context or no source. Find a real trigger before outreach. Internal warnings (orange panel) flag stale dates, sensitive events, or unverified location openings — always read these before sending.

One specific attribution problem THIS company faces right now, derived from their actual channel mix, retail partners, business model, and the trigger above. Not a generic segment-level complaint — prompt rules explicitly reject pain points that could be copy-pasted to any other brand in the same vertical. Should read as a logical consequence of the trigger (e.g. "launched in Whole Foods" trigger → pain point about retail-DTC attribution overlap, platform ROAS double-counting customers). If a pain point feels generic, flag it — the prompt missed, re-run or edit manually.

One sentence naming the specific Atrilyx capability that solves the specific pain point above. Pain-to-capability mapping is enforced: retail/DTC channel split → offline conversion tracking + Full-Funnel Visibility Dashboard; Meta/Google overlap → Universal Pixel + cross-channel de-duplication; B2B long cycles → multi-touch attribution + Identity Resolution Engine; iOS signal loss → AiTRK + Privacy-First Architecture; creative fatigue → Automated A/B Testing; multi-location → Full-Funnel + offline conversions. Quality flag to watch for: An orange internal warning may flag an unsupported Atrilyx capability reference. If flagged, replace the hook with a documented capability before sending.

A 60–90 word personalized cold email: lowercase sub-6-word subject, three short lines (trigger reference, what Atrilyx does, a 15-minute ask), first-name-only sign-off. The body carries the selected angle (Attribution Pain, Performance Guarantee, or Competitive Intel) and is bound by a no-new-facts rule — it can only reference details present in the card's trigger / painPoint / budget / hook / persona fields; it can't invent article titles, quotes, statistics, partner names, or specifics beyond the research. If you see anything oddly specific in the email that isn't in the card fields above, that's a sign the no-new-facts guardrail missed — flag it. Replace [Your Name] before sending.

Copy Email / Copy Subject — one-click copy of the drafted email. 🔍 Find email — when a card has a named contact but no email address, this looks up the contact's work email via RocketReach (uses the LinkedIn profile when available, otherwise name + company). 🔍 Find contact — appears only on cards where research found no contact at all ("Contact: Not Found"): it searches RocketReach for the company's marketing decision-maker (CMO → VP Marketing → Head of Growth priority order) and fills in the name, title, LinkedIn, and email when found. A discovered contact is marked Needs Review, not Verified — sanity-check before sending. Build Kit / Queue Sequence — Atrilyx MCP actions that assemble the full sales kit or queue the outreach sequence for the card. Don't show again — permanently excludes the company from your future runs (per-user; manage the list via the 🚫 Excluded modal). Also check the 🔍 Research Log button above the results for every search and source URL the AI used this run.

Computed deterministically. A-Priority requires all three: (1) budget confidence is High or Medium, (2) trigger is substantive and company-specific (not a generic placeholder), (3) CONTACT: VERIFIED from an official company source. B-Priority = anything missing one of the three — including contacts that are named but only verified via LinkedIn (CONTACT: NEEDS REVIEW). User-requested companies (via Custom Pitch Notes) stay A regardless. Focus outreach on A-priority prospects first; B-prospects need more verification before outreach.

High = direct primary-source financial disclosure (SEC filing, 10-K, earnings, press release with disclosed spend, NCUA, IPEDS, major business press with explicit revenue figure). Medium = revenue × industry ratio math, location-count × per-location benchmark, or Crunchbase funding inference. Low = aggregator-only source (ZoomInfo / RocketReach / Growjo / LeadIQ / Datanyze / Owler), no primary-source fetch, or general estimate with limited data. Auto-demotion rules apply: High without a primary-source keyword → Medium; aggregator-only → Low; zero successful fetches on company domain or sources → Low + BUDGET: NEEDS REVIEW / internal warning. Treat Low as a lead to verify manually, not a fact to send.

Click ⬇ Export CSV for a spreadsheet with all prospect data including contacts, budgets, triggers, pain points, and email copy. Works with Excel, HubSpot, Salesforce, Apollo, and Outreach.