Skip to main content

Agent Building | Step 1: Connect or Query Your Data

Lay the foundation by effortlessly connecting, filtering, and validating the data sources that power your Agent’s insights.

Chaim Davies avatar
Written by Chaim Davies
Updated this week

Step 1: Connect or Query Your Data

Before any analysis can begin, your Agent needs a reliable foundation of data. In this step, you’ll select one or more “Get Data” actions to pull in the exact datasets your Agent will work with—whether that’s revenue by campaign, customer lists, or even web-scraped intelligence. Every subsequent insight depends on the completeness and accuracy of this initial pull.


1.1 Available “Get Data” Actions

Action

What It Does

Example Use Case

Prompt to SQL

Converts plain-language prompts into SQL queries on your warehouse.

“Show me total revenue, order count, average order value, blended ad spend, blended ROAS, and MER for the last 30 days.” Returns a table with columns like revenue, orders, aov, mer.

Pre-Load Dashboard

Leverages existing Triple Whale dashboards as data sources.

Select your Web Analysis 360 dashboard to grab weekly landing page performance, so you don’t need to rebuild queries you already use.

Search Web

Scrapes external web data via URLs or predefined connectors.

“Pull current competitor pricing for product X from example.com,” enabling you to compare against your own price list.

Google Sheet

Connects to a live Google Sheet in your integrated account.

Link to your ‘Influencer Partnerships’ sheet, which your marketing team updates daily with spend, impressions, and commission rates.

Upload CSV

Imports static CSV files from your computer or cloud storage.

Upload your product catalog CSV (with SKUs, categories, cost, and launch dates) for one-off or scheduled analyses.

Load Previous Agent Run

Re-uses the result set from the Agent’s most recent execution.

Feed last week’s marketing performance report—generated by this Agent—into the next analysis step.


1.2 Prompt to SQL: Specify Data Requirements

If you’ve picked a Prompt to SQL (or Search) action, you’ll configure your query to specify exactly which slice of data you need:

  • Date Ranges: Define start and end dates (e.g., last 7 days, month-to-date, or a custom range keyed to a promotion period).

  • Specific Data Points: Narrow by product category, region, customer cohort, or UTM parameters.

  • Fields & Parameters: Choose the attribution model, window, or any exclusions you'd like to set from the data set.

Prompting Best Practices

For a deep dive into best practices using prompts to generate insights, please refer to our article: Mastering Prompts for Actionable Insights


1.3 Combining Multiple Data Sources

Complex analyses often require enrichment from diverse inputs. You can stack as many “Get Data” steps as needed:

  1. Customer Segments + Sales Data

    • Pull a CSV of high-value customers, then join to dashboard sales metrics to compare purchase frequency.

  2. Web-Scraped Trends + Internal Metrics

    • Use “Search” to grab recent Google Trends interest in your brand, then correlate with daily site visits from a relevant dashboard.

  3. Live Sheets + Historical Backfill

    • Merge a Google Sheet of upcoming promotions with a warehouse backfill CSV to forecast incremental lift.


1.4 Save, Preview, and Validate

After configuring each “Get Data” step:

  1. Click Save (or Next) to lock in your settings.

  2. Preview the Output: Ensure the sample rows match your intent—check date ranges, column names, and data quality.

  3. Fix Errors Early: If a connector can’t authenticate or the SQL throws an error, you’ll be notified immediately.

Validating while you build helps surface any gaps in the prompt, prevents “data not found” failures downstream, and sets your Agent up for a smooth analysis run.


Why It Matters:

A well-constructed Step 1 ensures that every insight, forecast, or recommendation your Agent produces is grounded in the right data. Spend time here up front—defining, filtering, and validating—and you’ll save hours of troubleshooting later while delivering more accurate, actionable results to your team.

Did this answer your question?