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

# AI Properties

> AI Properties are custom research signals that the AI evaluates for each company in your pipeline. You define the question — "Has this company recently raise...

<Snippet file="early-access.mdx" />

## Overview

AI Properties are custom research signals that the AI evaluates for each company in your pipeline. You define the question — "Has this company recently raised funding?", "Is this company hiring for roles related to our product?" — and the system researches every target account to produce a structured, evidence-backed answer.

AI Properties power multiple parts of the platform: they appear as signals on company pages, feed into prospecting task generation, and inform the AI agents when they reason about accounts.

## Property output types

Each AI Property has a configured output type that determines the format of the answer.

| Output type  | What it produces                 | Example                                                                                                  |
| ------------ | -------------------------------- | -------------------------------------------------------------------------------------------------------- |
| **Yes / No** | A binary answer                  | "Does this company use Kubernetes?" → Yes                                                                |
| **Dropdown** | One value from a set you define  | "Primary industry vertical" → FinTech, HealthTech, DevTools, Other                                       |
| **Score**    | A numeric rating on a 0–10 scale | "How strong is this company's product-market fit signal?" → 7                                            |
| **Freeform** | An open-ended text answer        | "What is this company's primary competitive differentiator?" → "AI-powered supply chain optimization..." |

All output types include a 1–3 sentence description explaining the answer and up to 3 citations linking to the evidence sources.

## How to create an AI Property

1. Navigate to the **AI Agent Builder** page.
2. Click **Add Property**.
3. Fill in:
   * **Name** — A clear label (e.g., "Recent Funding Round").
   * **Instructions** — Tell the AI what to research and how to evaluate the answer. Be specific — the more detail you provide, the more accurate the output.
   * **Output type** — Select Yes/No, Dropdown, Score, or Freeform.
   * For **Dropdown**: Add at least 2 options for the AI to choose from.
   * **Refresh Every (days)** — Enter the number of days before the answer is re-evaluated (e.g., 7). Leave blank for no expiration.
   * For **Score**: Optionally set the scale description to guide how the AI assigns ratings.
4. Click **Save**.

## Writing effective instructions

The instructions field drives the quality of research output. Include:

* **What to look for** — "Check the company's careers page, recent press releases, and LinkedIn job postings for roles mentioning Kubernetes, Docker, or container orchestration."
* **How to evaluate** — "Answer Yes only if there is direct evidence of production usage, not just job postings mentioning the technology."
* **Edge cases** — "If the company is a consultancy that implements the technology for clients but does not use it internally, answer No."

Vague instructions produce vague answers. "Is this company a good fit?" gives the AI nothing to work with. "Does this company have an engineering team of 50+ that builds cloud-native applications on AWS or GCP?" gives it a concrete research target.

For detailed guidance, see [Writing Effective AI Property Instructions](/sales-agents/guides/write-effective-ai-properties). For ready-to-use configurations, see the [AI Property Examples Library](/sales-agents/guides/ai-property-examples).

## How research works

When the AI researches a property for a company, it:

1. Gathers available data about the company (CRM records, website, public sources).
2. Searches the web for relevant evidence.
3. Evaluates the evidence against your instructions.
4. Produces a structured answer with the output value, a description explaining the reasoning, and citations linking to sources.

Research results are stored per company and refreshed based on the property's refresh policy.

## Refresh interval

Each property has a **Refresh Every (days)** setting that controls how often answers are re-evaluated. Enter the number of days between re-research cycles (e.g., 1 for daily, 7 for weekly, 30 for monthly). Leave blank for no automatic expiration.

When a property's answer is still within its refresh interval, re-research is skipped to avoid unnecessary work. Use shorter intervals for fast-changing signals (e.g., hiring activity) and longer intervals for stable attributes (e.g., technology stack).

## Auto AI Research

Auto AI Research runs your property research automatically across companies in your target sales views on a recurring schedule.

### How to enable

1. Open the **AI Agent Builder** and select a property.
2. In the property editor, find the **Auto Research** section.
3. Toggle **Enable Auto Research**.
4. Set the **Refresh Every (days)** interval to control how often answers are re-evaluated.
5. Select one or more **sales views** to target. The AI researches companies that appear in these views.
6. Click **Save**.

### How it runs

When auto research is enabled:

* The system evaluates all companies in the selected sales views.
* Companies whose existing answer has exceeded its refresh interval are re-researched.
* Companies with fresh answers are skipped.
* Results are stored with updated evidence and citations.

### Monitoring progress

After triggering a research run (or when the scheduled run starts):

* The property editor shows a **progress indicator** with the number of companies processed out of the total.
* You can check the status at any time — it updates automatically.
* If a run is taking too long or you need to stop it, click **Cancel Research** to halt the current run.

### Manual trigger

You can trigger a research run on demand without waiting for the next scheduled cycle:

1. Open the property in the AI Agent Builder.
2. Click **Run Research Now**.
3. Monitor progress in the status indicator.

## Value distribution

After research has run across companies, you can view the distribution of answers to understand how your target accounts break down.

The distribution view shows:

* **Yes/No properties** — Count and percentage of Yes vs. No answers.
* **Dropdown properties** — Breakdown by each option.
* **Score properties** — Distribution across the 0–10 scale.
* **Freeform properties** — Summary of the most common themes.

Use distributions to validate that your property is producing useful signal. If 95% of companies answer "Yes," the property may not be discriminating enough to be useful for prioritization.

## Where AI Properties appear

| Location                      | How they're used                                                                            |
| ----------------------------- | ------------------------------------------------------------------------------------------- |
| **Company Signals tab**       | Each property appears as a signal card with the answer, description, and citations          |
| **Signal Summary**            | Properties feed into the AI-generated Fit, Timing, and Context scores                       |
| **Prospecting configuration** | Properties can be selected as research signals in the Signals tab                           |
| **Task generation**           | The AI references property answers when deciding what tasks to create and how to prioritize |
| **Company Agent chat**        | The agent has access to property answers when responding to questions about a company       |

## Constraints

* AI Properties are scoped to companies — they research company-level signals, not deal-level or contact-level.
* Auto research processes companies in batches. Large sales views may take time to complete.
* Each research operation consumes agent credits.

## FAQ

<AccordionGroup>
  <Accordion title={"How long does research take per company?"}>
    A few seconds per company. For large sales views with hundreds of companies, a full research run may take several minutes.
  </Accordion>

  <Accordion title={"Can I edit a property after it has been researched?"}>
    Yes. Editing instructions or output type creates a new version of the property. Existing answers are preserved until the next research cycle refreshes them.
  </Accordion>

  <Accordion title={"What happens if I change the output type?"}>
    Previous answers are kept as-is. New research produces answers in the updated format. The next auto research cycle or manual trigger re-evaluates all companies with the new output type.
  </Accordion>

  <Accordion title={"Why are some companies skipped during auto research?"}>
    Companies are skipped when their existing answer is still within its refresh interval. If you need to force a full re-evaluation, use the manual trigger with the resync option.
  </Accordion>

  <Accordion title={"Can I use AI Properties without auto research?"}>
    Yes. Properties are researched individually when company context is generated (either manually or by the overnight worker). Auto research is optional — it pre-computes answers across your sales views so they are ready before anyone visits the company page.
  </Accordion>

  <Accordion title={"How do I know if my property instructions are good?"}>
    Check the value distribution after a research run. If answers cluster heavily on one value (e.g., 90%+ "Yes"), the question may be too broad. If descriptions are vague or citations are irrelevant, the instructions need more specificity.
  </Accordion>
</AccordionGroup>