Skip to main content
All CollectionsAgents
Building and Editing Agents
Building and Editing Agents

Follow the instructions below to set up powerful agentic automations

Chaim Davies avatar
Written by Chaim Davies
Updated this week

Ready to put Triple Whale’s Agents to work for your business? This step-by-step guide walks you through creating and customizing an Agent in Triple Whale. Whether you start from scratch or tweak a pre-built template, follow the instructions below to set up a powerful, automated workflow.

1. Access the Agent Builder

  1. Navigate to the Agents Section

  2. Create or Edit an Agent

    • Create New Agent: Click New Agent to start building a custom Agent.

    • Edit Existing Agent: Locate an existing Agent from your list, click and click Edit Agent to modify its steps.

      • Note: If you are not the creator of this agent, you will not be able to edit it. Instead, you can select Duplicate Agent to make a complete copy.

You’ll see the Agent Builder interface, which includes options for Get Data, Analyzing Data, Destinations, and Agent Ops—all neatly organized to help you craft an Agent in logical stages.


2. Add and Configure “Get Data” Steps

Your Agent starts by gathering the data it needs. This is crucial because every subsequent step relies on this initial dataset.

  1. Choose a Get Data Step

    • From the Builder’s sidebar, select a Get Data option:

      • Query Builder (convert natural language to SQL)

      • Search (pull in web data)

      • Dashboard Data (use existing Triple Whale dashboards)

      • Previous Run (pull data from the last Agent execution)

      • CSV (upload a CSV file)

      • Google Sheet (connect a sheet from your integrated Google account)

  2. Specify Your Data Requirements

    • Provide details like date ranges, filters, or the exact dataset you want to query.

    • If using Query Builder, enter your natural language prompt or refine the auto-generated SQL.

  3. Add More Data Sources (Optional)

    • You can include multiple Get Data steps if your Agent needs data from various places (e.g., combining a CSV file with data from your dashboard).

  4. Save and Continue

    • Click Save or Next to confirm your data settings and move on to analysis.


3. Add “Analyzing Data” Steps to Unlock Insights

Now that you’ve collected your data, it’s time to do something with it. The Analyzing Data steps let you transform raw numbers into actionable insights.

  1. Select an Analyzing Data Step

    • From the Builder’s sidebar, pick one or more of the following:

      • Insights & Recommendations: Quickly generate insights, offer recommendations, and summarize data to guide decisions.

      • Analyze Data: Transform natural language requests into Python code for advanced analytics.

      • Forecasting: Use historical data to predict future trends and inform strategic planning)

      • MMM (Marketing Mix Modeling): Analyze marketing performance to optimize future spending

      • Vision: Leverage image and video analysis to extract meaningful information from creative content.

  2. Customize Your Analysis

    • Enter prompts or instructions for the type of insights you want

  3. Review the Output

    • Click Test to simulate the output of the agent. Make any updates to the prompt as needed.

  4. Chain Multiple Analyses (Optional)

    • You can combine different analysis steps for a more comprehensive result. For example, run Insights & Recommendations after you’ve generated a Forecasting insight.

Pro Tip: Make sure your prompts and queries only reference data sources included in the Get Data step. If the data isn’t included in the query, the Agent can’t analyze it!


4. Decide on Your “Destinations”

After the analysis is done, you need to choose where to send the results. This step ensures the right people or tools receive your data.

  1. Pick a Destination: Send the Agent to any of the following options:

    • Dashboard: Send outputs directly to one of your Triple Whale dashboards.

    • Email: Receive insights straight to your inbox.

    • Slack: Integrate with Slack for real-time team updates.

    • Google Sheet: Export data for collaborative analysis.

    • Warehouse: Sync with your data warehouse for large-scale analytics.

    • Webhook: Push data to external systems for further processing.

    • Push Notification: Get instant alerts when important data is available.

    • Feed: Share outputs on your Triple Whale feed for continuous updates.


5. Leverage “Agent Ops” for Advanced Control

For more complex workflows, use Agent Ops to fine-tune the logic and sequence of your Agent.

  1. Sub-Agents

    • Run other Agents within the current Agent to gather multiple datasets or perform sequential tasks.

    • Example: Use one Agent’s output as the input for another Agent’s forecasting step.

  2. Loop

    • Repeat a specific step multiple times. This is useful if you need to iterate over a list of items (e.g., analyzing each product category in turn).

  3. GPT Rule

    • Add AI-driven if-else logic that relies on GPT-based reasoning. This allows for more nuanced, context-aware decisions based on the data the Agent has already gathered.

  4. Rule

    • Set straightforward if-else conditions. For instance, “If conversion rate < 2%, send an alert email; otherwise, continue to next step.”


6. Review and Save Your Agent

Once you are ready to launch your agent, click Finish Up to finalize your Agent settings.

  • Set any recurring schedules or triggers if needed (e.g., “Send daily at 9 AM”).

  1. Name and Describe

    • A title and description will be auto-generated based on the content of the agent, and you may edit those at this time.

  2. Schedule the Runs

    • Select the cadence for this Agent to run on, as frequent or infrequent as you wish.

  3. Save and Run

    • Click Save and Run to finalize your Agent. It will run once right away and then follow the schedule you’ve set. You can return at any time to make changes.


Tips for Building Effective Agents

  • Keep Data Relevant: Only reference datasets you’ve included in the Get Data step.

  • Use Clear Prompts: The more specific your prompts, the more accurate the Insights & Recommendations will be.

  • Test and Iterate: Test Run the Agent to ensure the output meets your expectations, and make adjustments accordingly.

  • Combine Analyses: Consider chaining multiple steps (e.g., MMM followed by Forecasting) for deeper insights.

  • Stay Organized: Label each step with descriptive titles so you can easily track what’s happening at a glance.


Conclusion

Building and editing an Agent in Triple Whale is a straightforward process once you understand the four key stages: Get Data, Analyzing Data, Destinations, and Agent Ops. By following these steps, you’ll create powerful workflows that automate repetitive tasks, deliver meaningful insights, and help you make data-driven decisions. With just a bit of practice, you’ll have a library of Agents running on autopilot—freeing you to focus on strategic growth for your e-commerce business.

Did this answer your question?