Enterprise Researcher v2.0.

Every B2B seller asks the same three questions before outreach: is this account a fit, is the timing right, and who should I contact? Manual research answers those questions well, but it takes hours. Generic intelligence tools scale, but they give every seller the same firmographic data and the same weak scores.

The missing layer was personalized, evidence-grounded analysis at machine throughput. The output had to be tailored to the seller's products, ICP, timing signals, competitors, and technographic assumptions. It also had to cite public sources and render into a product surface without manual cleanup.

I split the system into specialized stages because the job is not one prompt. Seller profiling, prospect research, account analysis, contact selection, and batch dispatch all have different failure modes, costs, and retry boundaries.

type ProspectAnalysis = {
  engage: boolean
  priority: "critical" | "high" | "medium" | "low"
  fitAssessment: {
    overallFitScore: number
    overallFitConfidence: number
    verdict: string
    checklist: Array<{
      label: string
      met: "Met" | "Partial" | "Not met"
      evidence: string
      source: string
    }>
    competitorPresence: CompetitorSignal[]
  }
  timingAssessment: {
    overallTimingScore: number
    signals: TimingSignal[]
  }
  technographicIntelligence?: {
    overallStackMaturity: string
    buyingWindowSignals: Signal[]
    competitorGapWedges: Wedge[]
  }
  accountClassification: {
    tier: "Tier 1" | "Tier 2" | "Tier 3"
    rationale: string
  }
  evidenceUrls: EvidenceUrl[]
}