> ## Documentation Index
> Fetch the complete documentation index at: https://hc.starbridge.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Custom AI Analysis

> Analyze the data already in your Bridge with a natural-language prompt, and run it across every row at once.

## Example use cases

* **Scoring.** Score accounts or signals against your own criteria, e.g. a 1-5 readiness score.
* **Summaries.** Summarize a meeting, RFP, or set of items into a few lines.
* **Drafting.** Draft a personalized email or call script from each row's data.

## Set up a Custom AI Analysis column

From any Bridge, click **"Add enrichment"**, choose **Custom AI Analysis**, name the column, then configure it.

<Steps>
  <Step title="Write your prompt">
    Describe what you want in plain language. Type **@** to reference another column and pull its value into the prompt, e.g. *Summarize @Meeting in two sentences for a sales rep.*

    <Tip>
      The more specific you are, the better the output. Add validation criteria like word limits, tone, and the exact format you want back.
    </Tip>
  </Step>

  <Step title="Set your output fields">
    Choose how the result is shaped:

    * **Single Output** returns one value per row.
    * **Multiple Output** breaks the result into several fields, each in its own column.

    Then pick a **Field Format** (e.g. text, number, or date) so the column is typed correctly for sorting, filtering, and CRM sync.
  </Step>

  <Step title="Set Run Conditions and Rerun Settings (optional)">
    * **Run Conditions** control when the enrichment runs. Without one, it always runs.
    * **Rerun Settings** control how often the column refreshes. Without a rule, it won't re-run automatically.
  </Step>

  <Step title="Add & run the enrichment">
    Click **"Add Enrichment"** to run it. Test on the first few rows first: prompting is iterative, so check the output, refine, and rerun before running every row.
  </Step>
</Steps>

<Note>
  Custom AI Analysis only uses data already in Starbridge (the signal and your other columns). To pull in information from outside the platform, use a [Web Agent column](/builders/example-prompts-for-web-agents) instead, which searches the web.
</Note>

## Example Prompts

<AccordionGroup>
  <Accordion title="Meeting opportunity analysis" icon="magnifying-glass">
    ```text theme={null}
    Analyze the {Meeting} for signals related to:
    - 911 dispatch upgrades or system modernization
    - Emergency communication or alerting infrastructure
    - Interagency coordination (e.g., police, fire, EMS)
    - Gaps in response time or incident command systems
    - Any awarded grants or planned funding for emergency or public safety technology

    Output instructions:
    Return the following:
    - Opportunity Type (e.g., "911 System Modernization", "Alerting System Upgrade", "Public Safety Interoperability Grant")
    - Positioning: 1–2 sentence pitch for a sales rep to use, based on the insight

    Example output:
    Opportunity Type: Interagency Emergency Communication Upgrade
    Positioning: The board noted delays in cross-department communication between police, fire, and EMS. EXAMPLE_ACCOUNT’s secure, multi-channel messaging system can reduce response time and improve coordination during critical events.
    ```
  </Accordion>

  <Accordion title="Customer reference picker (closest 1–3)" icon="users">
    ```text theme={null}
    Your task is to identify up to three of our organization's existing customers that are most similar to the buyer described. These will be used as sales references.

    Carefully analyze the provided buyer profile and customer list, then select the customers that most closely resemble the buyer based on the following hierarchy of criteria:
    1) State or geographic proximity — prioritize customers in the same state or nearby regions
    2) Organization's size and type — match similar organization types (e.g., university, hospital, enterprise, school district)
    3) Industry or operational similarity — match on characteristics like: public vs. private, K-12 vs. higher education, district size, urban/suburban/rural setting, and special focus areas (e.g., STEM programs, special education services)

    Available context columns:
    - {buyer:name} — the name of the buyer organization
    - {buyer:stateCode} — the buyer's state code

    Our current customer list:
    - Active Floor
    - Allen Roberts
    - Beekmantown Central School District
    - Benhurst Primary School
    - Chicago Public Schools
    - Cradle of Aviation Museum
    - Eastern Suffolk BOCES
    - Edward Tracy Elementary
    - Greenwich Central School
    - Henry Viscardi School
    - Kings Park Central Schools
    - Milltown Public Schools
    - Oak Grove Elementary School
    - Potter-Thomas School
    - Putnam County School District
    - Vernon-Verona-Sherrill Central School District
    - West Islip School District
    - Wildwood School

    Output format (comma-separated list of short names):
    Return ONLY a comma-separated list of up to 3 short, colloquial names for the organizations (e.g., "UCLA" instead of "University of California, Los Angeles").

    Example output:
    UCLA, UArizona, UAlaska

    Instructions:
    - Use short, colloquial names for organizations
    - Always return the most similar 1–3 customers from the list, even if the matches are weak
    - Prioritize the hierarchy: geographic proximity, then organization type, then industry similarity
    - Select the closest matches available, even if none are perfect
    - Return ONLY the comma-separated list with no explanations or additional text
    ```
  </Accordion>
</AccordionGroup>