Skip to main content
The AgentFlow pipeline is the central organizing concept of the platform. Every lead that enters the system travels through the same five stages, in the same order, with a clear output at each step that feeds directly into the next. Understanding the pipeline helps you configure AgentFlow purposefully, reason about where human judgment applies, and identify how your existing tools connect to the system at the handoff layer.

​
How the pipeline is structured

The pipeline is not a loose collection of features — it is a deliberate sequence. Each stage has a defined input, a transformation, and an output. Stages do not run in parallel; a lead moves forward only when the current stage produces a result. This sequential structure is what makes the pipeline auditable and what gives operators a clear, consistent view of every lead’s status at any moment. The diagram below shows the end-to-end flow: 
┌─────────────────┐
│   Lead Intake   │  ← Form submission, webhook signal, or channel entry
└────────┬────────┘
         │  Raw lead record
         ▼
┌─────────────────┐
│ AI Qualification│  ← OpenAI evaluates against your criteria
└────────┬────────┘
         │  Score + confidence + summary
         ▼
┌─────────────────┐
│ Operator Review │  ← Human decision: approve / reject / escalate
└────────┬────────┘
         │  Approved lead record + decision log
         ▼
┌─────────────────┐
│Billing Readiness│  ← Checkout trigger via Stripe or PayPal
└────────┬────────┘
         │  Payment intent or checkout session
         ▼
┌─────────────────┐
│  CRM / Handoff  │  ← Downstream system receives enriched lead record
└─────────────────┘

​
The five stages

1

Lead Intake

Lead Intake is where inbound interest enters the system. AgentFlow captures leads through form submissions, direct webhook signals, or other configured channel entry points. At this stage, the system records the raw data exactly as submitted — no transformation or evaluation occurs yet.The intake layer is intentionally permissive: its job is to capture, not filter. Every submission becomes a lead record, regardless of apparent quality. Filtering and scoring happen in the next stage.Key outputs: a timestamped lead record containing submitted form fields and intake metadata (channel, entry point, referral source where available).
2

AI Qualification

AI Qualification is where AgentFlow evaluates fit. The system sends each lead record to OpenAI, which assesses the submission against your configured criteria — fit signals, business rules, and disqualifying factors — and returns a structured score.The output of this stage is not a decision; it is a structured signal. The score, confidence level, and plain-language summary give operators the context they need to make a fast, informed decision in the next stage.Key outputs: qualification score (0–100), confidence rating (low / medium / high), fit and risk signals, and a plain-language evaluation summary.For a detailed explanation of how scoring works, see AI Lead Qualification.
3

Operator Review

Operator Review is the human checkpoint in the pipeline. After the AI returns its assessment, an authenticated operator opens the review dashboard and sees the lead’s original data alongside the qualification output. The operator then makes one of four decisions: approve, reject, escalate, or request more information.This stage exists because AI scoring produces a signal, not a verdict. Business context, edge cases, relationship history, and strategic priorities belong to human operators — not models. The review stage is where that judgment is applied and recorded.Every decision is written to an audit trail with a timestamp and optional reasoning note, creating a durable log of how each lead was handled.Key outputs: an operator decision record, an updated lead status, and an immutable audit log entry.For a full explanation of the review interface and decision options, see Operator Review.
4

Billing Readiness

Billing Readiness activates when a lead is approved. AgentFlow triggers a checkout flow through your configured payment provider — Stripe or PayPal — presenting the approved lead with a customer-facing payment step.This stage connects qualification to revenue. Rather than manually following up with a payment link, your team can configure AgentFlow to trigger checkout automatically on approval, reducing time-to-payment and removing a common manual handoff point.AgentFlow’s billing layer is designed with EU VAT compliance posture in mind, and supports both test mode and live mode operation.Key outputs: a checkout session or payment intent routed to the approved lead.For configuration details, see Billing Readiness.
5

CRM / Workflow Handoff

CRM and Workflow Handoff is the final stage. Once a lead has completed the billing step — or once your configuration specifies the handoff trigger — AgentFlow passes the enriched lead record downstream to your CRM or workflow system.At this point the record carries everything accumulated through the pipeline: original submission data, AI qualification output, operator decision, audit log entries, and payment status. Downstream systems receive a fully enriched record rather than a raw form submission.The handoff layer is designed as an integration surface. The exact systems you connect — whether a CRM, a project management tool, a notification channel, or a custom webhook — depend on your RevOps configuration.Key outputs: an enriched lead record delivered to one or more downstream systems via the configured integration layer.

​
How stages connect

Each stage’s output is the next stage’s input. A lead does not skip stages, and stages do not run until their predecessor completes. This strict sequencing means:
  • The AI qualification stage always receives a complete intake record
  • Operators always review a scored, structured record — never a raw, unassessed submission
  • Billing triggers only on an explicit operator approval — never automatically from scoring alone
  • Handoff carries a complete, enriched record — not a partial or intermediate state

​
Where human oversight sits

The pipeline has one explicit human checkpoint: Operator Review. Every other stage is automated. This is a deliberate design choice — automation handles volume and consistency; humans handle judgment and accountability. Operators can intervene beyond the review stage too. Your configuration can require re-approval at specific transitions, flag certain lead types for escalation regardless of score, or hold leads pending additional information before billing triggers.

​
Next steps