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

# When to Use Pipelines

> Understand the scenarios where Pipelines are the right choice, and when to use Tools instead.

## Overview

Pipelines shine when you need data to be **ready before an agent runs**. If your agent requires context — historical records, aggregated totals, filtered lists — a Pipeline ensures that data is pre-processed and available instantly.

This page covers the most common use cases for Pipelines and helps you decide when to use a Pipeline versus a Tool.

***

## Core Use Cases

### 1. Aggregating Financial Data for Analysis

**Scenario:** A finance agent calculates R\&D tax credits each month. To do this, it needs a clean, structured view of all relevant expenses from your accounting system.

**Why a Pipeline:** Pulling thousands of expense records from QuickBooks at agent runtime would be slow and unreliable. Instead, a Pipeline extracts and aggregates expenses daily — so when the agent runs, it simply reads from the pre-built `aggregated_expenses` outcome.

> Use Pipelines when you need to aggregate expense data from QuickBooks for monthly R\&D tax credit calculations.

***

### 2. Pre-Processing CRM Data for Sales Agents

**Scenario:** A sales support agent needs to show a rep the full history of a customer — deals, contacts, notes — before a call.

**Why a Pipeline:** CRM data is large and relational. A Pipeline extracts and joins the relevant records on a schedule, producing a clean `customer_records` table. The agent reads from this table instead of making multiple live API calls during the session.

> Use Pipelines to pre-process customer records so your support agent can show relevant deal history and contact details without latency.

***

### 3. Monitoring Risk and Compliance

**Scenario:** A compliance agent flags any deals or transactions above a risk threshold for human review.

**Why a Pipeline:** Risk data needs to be continuously refreshed. A Pipeline set to run hourly pulls deals from your CRM, filters those above the threshold, and writes them to a `deals_over_10k` outcome. A HITL Experience then displays this table to the reviewing team.

***

### 4. Populating HITL Dashboards

**Scenario:** Your operations team reviews an AI-generated summary each morning. The dashboard needs to show live data from multiple systems.

**Why a Pipeline:** HITL Experiences bind to Pipeline Outcomes to display structured data in tables and charts. The Pipeline runs overnight to prepare the data; when the reviewer opens the Experience, the dashboard loads instantly.

***

### 5. Normalizing Data Across Systems

**Scenario:** Your company uses both Salesforce and HubSpot. An agent needs a unified view of accounts.

**Why a Pipeline:** A Pipeline can extract data from both systems (using two Connectors), transform and normalize the schemas, and write a unified `accounts` table. The agent works from a single, clean dataset.

***

## Pipelines vs. Tools: Decision Guide

Use this table to decide whether to use a Pipeline or a Tool for a given need.

| Need                                          | Use          | Example                                         |
| --------------------------------------------- | ------------ | ----------------------------------------------- |
| Pre-fetch and store data before an agent runs | **Pipeline** | Pull all deals from CRM nightly                 |
| Display data in a HITL dashboard              | **Pipeline** | Show expense totals in a review table           |
| Aggregate or transform large datasets         | **Pipeline** | Group expenses by category and project          |
| Refresh data on a schedule                    | **Pipeline** | Update customer records every hour              |
| Send an email during agent execution          | **Tool**     | `Send Email` tool called when agent triggers    |
| Create a record in real time                  | **Tool**     | `Create Salesforce Opportunity` called by agent |
| Fetch a single live record on demand          | **Tool**     | `Get Contact by ID` called during agent run     |
| Trigger a webhook or external action          | **Tool**     | `Notify Slack Channel` called by agent          |

<Info>
  A good rule of thumb: if the data exists before the agent runs and doesn't change per-request, use a **Pipeline**. If the action or data fetch happens *during* agent execution in response to something specific, use a **Tool**.
</Info>

***

## What Data Sources Work With Pipelines?

Pipelines connect to external systems via **Connectors**. Any system with a Connector configured in Adopt AI can serve as a Pipeline data source. Common examples include:

* **CRM systems** — HubSpot, Salesforce
* **Accounting & finance** — QuickBooks, Xero
* **Cloud storage** — AWS S3, Google Drive
* **Databases** — internal databases via JDBC or REST
* **Productivity tools** — Google Sheets, Airtable
* **Support platforms** — Zendesk, Zoho Desk

<Tip>
  If you don't yet have a Connector for your data source, you can create one during the Pipeline setup flow. See the [Connectors documentation](/essentials/connectors) for details.
</Tip>

***

## When NOT to Use Pipelines

Pipelines are not the right choice when:

* The data is **user-specific and requested at runtime** (e.g., "fetch the details of *this specific* ticket the user just mentioned")
* The operation **writes data to an external system** (e.g., creating a record, sending a notification)
* The data changes so rapidly that **scheduled pre-computation would always be stale**
* The operation requires **real-time context from the conversation** (e.g., looking up a specific customer by name the user just said)

For these cases, use Tools instead.

***

## Next Steps

* [Building Your First Pipeline](/essentials/pipelines-create) — step-by-step creation guide
* [Pipeline Features](/essentials/pipelines-features) — explore scheduling, transformations, and monitoring