> ## Documentation Index
> Fetch the complete documentation index at: https://www.landbase.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Qualify Leads

> Define custom yes/no criteria in plain language; every record is evaluated with cited justification and optional web research.

<div id="breadcrumb-wrapper" />

<div
  style={{display: 'none'}}
  data-related-resources={JSON.stringify([
{ title: 'Build Audience', href: '/reference/tools/build-audience' },
{ title: 'Search the Web', href: '/reference/tools/search-the-web' },
{ title: 'Tools overview', href: '/reference/tools' }
])}
  id="page-related-resources"
/>

# Qualify Leads

<div className="blog-post-content">
  Most lead qualification is rules-based — score contacts by title seniority, company size, industry match. If the field exists and matches a threshold, it's "qualified." This works for simple cases but falls apart when the criteria require judgment.

  "Does this company sell to enterprise customers?" There's no field for that. "Is this person likely the budget owner?" No checkbox in your CRM. "Has this company raised a Series B or later?" Maybe, but only if the funding data is complete and current.

  Qualify Leads evaluates every record using an AI model that reads the actual data and makes a judgment call — the same way a human SDR would, but across hundreds or thousands of records in minutes instead of days.

  ## How it works

  You define your criteria in plain language. No formulas, no field mapping, no scoring rubrics.

  * "Is this company likely a fintech?"
  * "Does this person's job description mention managing a team?"
  * "Is this company actively hiring for data roles?"

  The system evaluates every record against every criterion and returns three things per record:

  1. **A verdict** — Yes, No, or Unknown
  2. **Evidence** — the specific reasoning behind the verdict, citing what it found
  3. **Confidence context** — which data it used (description, keywords, web research) so you can audit the decision

  <Expandable title="What's happening under the hood">
    Each record is evaluated by a language model optimized for the task. The model receives the qualification criteria as numbered yes/no questions and the record's data formatted as context — company description, keywords, employee data, funding info, or any other field in the schema. It returns a structured response with a verdict and evidence string per criterion. Records are processed in parallel batches, backed by a distributed workflow engine for reliability. Responses are cached so re-runs are near-instant.
  </Expandable>

  ## It reads context, not just fields

  Traditional qualification checks whether `industry == "fintech"`. Qualify Leads reads the company description and decides whether it's a fintech — even if the industry field says "Software" or is blank. It evaluates semantic information (descriptions, keywords, job postings, skills) that rule-based systems can't touch.

  ## It can go to the web

  When the answer isn't in the structured data, the tool can perform live web research per record — searching Google, reading company websites, checking job boards. A criterion like "Does this company have a public pricing page?" triggers a targeted web search for that specific company. The AI model decides per record whether research is needed.

  <Expandable title="What's happening under the hood">
    When web tools are enabled, the qualification model upgrades to a more capable variant with tool-calling support. It gets access to web search and webpage scraping. The model decides autonomously whether to use these tools — it won't waste time researching a record where the offline data already answers the question. This adds a few seconds per record that needs research, but only for those records.
  </Expandable>

  ## Results stream in real time

  You don't wait for the entire table to finish. Verdicts appear row by row as each record completes. You can start reviewing qualified records while the rest of the table is still being evaluated. An early-stop parameter lets you halt after finding N qualified records — useful for quick prospecting passes.

  ## Every verdict is auditable

  No black-box scores. Every Yes, No, or Unknown comes with a written explanation citing what the model found. "Yes — company description mentions 'serving enterprise clients including Fortune 500 companies'" is a verdict you can trust or override.

  ## Performance

  * **1,000 records without web research** — approximately 8–12 minutes
  * **1,000 records with web research** — approximately 15–25 minutes
  * **Early stopping** — configurable threshold to halt after N qualified records found
  * **Incremental runs** — "run on remaining" mode re-qualifies only rows without verdicts
</div>