Lead Scorer

Concepts

Agent workflows

How a run works: the phases it goes through, where you are asked to approve, how to replay it, and how the daily drip works.

Last updated

A workflow run is the unit of work in Lead Scorer. One run takes a brief and produces a campaign ready to launch, pausing at the points where a human decision is worth more than a machine one.

The phases of a run

  1. 1

    Brief and interview

    The agent reads your product and ICP, then asks for what is missing rather than guessing.

  2. 2

    Discovery

    Companies pulled from the official registry, or from the web when the brief cannot be mapped to a registry filter.

  3. 3

    Lead approval

    You keep or drop the accounts. Rejections carry a written reason. This gate exists before any credit is spent.

  4. 4

    Enrichment

    Business context for the company, verified contact details for the person.

  5. 5

    Scoring

    Company ICP score and decision-maker score, against your product.

  6. 6

    Drafting and transport-safe cleanup

    Messages are written from real facts and versioned outreach rules, then placeholders and email HTML are sanitized before they reach your review queue.

  7. 7

    Ready to launch

    You approve the drafts, set the cap and the window, and launch.

Replaying a run

Every run is replayable step by step. You can see which companies came back from discovery, which leads were rejected and why, what enrichment produced, what each score was based on, and which quality checks governed a message before your review. If you cannot explain why a lead is in your pipeline, the tool has failed.

The daily drip

A run does not have to be a one-off batch. Set it to drip and a few fresh leads are found, enriched, scored and written every day, arriving as drafts in your queue like work from a teammate — still behind your approval, still inside your daily cap.

Running workflows from your own agent

Runs are drivable over MCP: list the available workflows, start one, advance it, and read its state. The approval gates behave identically — an agent can prepare everything and still cannot send.

list_workflows · start_workflow_run · advance_workflow_run · get_workflow_run · list_workflow_runs