Skip to main content
When a new lead arrives in AgentFlow, it is not simply logged and routed — it is evaluated. The AI qualification layer reads each submission against your configured criteria and returns a structured assessment: a numeric score, a confidence level, and a plain-language summary your operators can act on immediately. This page explains what that process looks like, what data feeds it, and why it is designed to run entirely on the server.

What lead qualification means in AgentFlow

Lead qualification is the step between raw inbound interest and a decision. Instead of relying on operators to manually read every submission and decide its priority, AgentFlow delegates the initial evaluation to an AI model. The result is a consistent, auditable signal that operators can accept, adjust, or override — depending on what your business rules require. In practical terms, qualification turns an unstructured form response into a scored record that travels through the rest of the pipeline.

How AI scoring works

AgentFlow sends each incoming lead to OpenAI for evaluation. The model receives the lead’s submitted data alongside your configured qualification criteria and returns a structured response. That response contains three components:
  • Score — a numeric value representing overall fit
  • Confidence — a rating reflecting how certain the model is in its assessment
  • Summary — a short, readable explanation of why the lead received that score
The model does not make final decisions. It surfaces a signal. Operators retain full authority to approve, reject, or escalate any lead regardless of its AI score.

What data feeds into scoring

AgentFlow draws on two categories of input when evaluating a lead: Form fields — the data a lead explicitly submits through your intake form. This typically includes information such as company name, use case description, team size, and any qualifying questions you have included. Intake metadata — contextual signals captured at the time of submission, such as submission timestamp, entry channel, and referral source where available. The richer your intake form, the more signal the model has to work with. Thin submissions receive lower confidence scores, which surfaces them for closer operator attention.

Qualification criteria you can configure

You define what a qualified lead looks like for your business. AgentFlow evaluates each submission against your configured criteria, which can include:
  • Fit signals — indicators that a lead matches your target customer profile, such as industry, company size, or stated use case
  • Business rules — explicit thresholds or conditions, such as minimum team size or required budget range
  • Disqualifying factors — signals that immediately reduce a lead’s score, such as stated incompatibility with your product
You set these criteria in your AgentFlow configuration. The AI model applies them consistently across every lead, removing the variability that comes from manual review at scale.

What the score output looks like

Every evaluated lead receives a structured qualification record. The table below shows the fields included in that output:
FieldTypeDescription
scoreInteger (0–100)Overall qualification score; higher values indicate stronger fit
confidenceString (low / medium / high)Model certainty in the assigned score
fit_signalsArray of stringsPositive indicators identified in the submission
risk_signalsArray of stringsFlags or concerns identified in the submission
summaryStringPlain-language explanation of the score reasoning
evaluated_atISO 8601 timestampWhen the AI evaluation was completed
criteria_versionStringVersion identifier of the qualification criteria used
Operators see this output alongside the original lead data in the review dashboard, giving them full context for their decision.

Why server-side processing matters

AgentFlow performs all AI qualification calls on the server. Your OpenAI API key and qualification prompts are never sent to the client or exposed in the browser. This keeps your credentials protected and prevents your qualification logic from being inspected or replicated by third parties.
Running qualification server-side also means that every call is logged, traceable, and consistent. You can audit which criteria version evaluated a given lead, when the call was made, and what the model returned — without relying on client-side state that can be tampered with.

Next steps

To configure your qualification criteria and connect your OpenAI provider, see the AI Qualification guide.