You are here:
Excel Agent (Beta) in Agentforce Operations
Use the Excel agent (beta) as an automation engine in Agentforce Operations. Instead of translating complex spreadsheet logic into multi-step instructions, the spreadsheet does the work.
Required Editions
| License Required |
|---|
This feature is a workspace that’s external to your Salesforce org and requires an Agentforce Operations license. To purchase an Agentforce Operations license, contact your Salesforce account executive. |
What the Excel Agent Does
Assign a task to the Excel agent, configure shared and requested fields as you would for any other agent task, and provide an existing Excel file or describe the calculations you want. The agent inputs the shared field values, runs the spreadsheet's formulas, and extracts the results into the requested fields.
When to Use the Excel Agent
The Excel agent is best for tasks that rely heavily on calculations and logic, especially when you already have a spreadsheet that does that work. Rather than writing complicated instructions to explain weighted scoring, thresholds, or multi-variable formulas, you upload the spreadsheet and Excel's formulas handle the math.
For example, instead of writing instructions to calculate a supplier risk score based on multiple variables, you upload your existing scoring spreadsheet. The agent plugs in the shared fields, uses Excel to calculate the weighted score, and extracts the final result.
The Excel agent overlaps with the Regrello agent, which can also perform multi-step math calculations. The key advantage of the Excel agent is usability. Business users already working in Excel can build and maintain automations without writing complex instructions or learning new tools.
Using an Existing Excel File
Upload an existing .xlsx file to the Documents field of the task. In the task Description field, write instructions telling the agent where to write the shared field values into the spreadsheet and where to read or extract the requested information values. This path works best for simple to medium-complexity spreadsheets.
Generate a New Excel File
If you don't have an existing spreadsheet, in the task Description field, describe the calculations and logic you want to perform on the fields in the task's Shared information section. The agent generates a structured Excel file from scratch to compute the requested fields.
Reviewing Agent Output
After configuring the fields, run a test from the Test tab. The test generates a plan and, if needed, creates the Excel file. Review the output before publishing the blueprint.
After a task runs, you can download the completed Excel file from the task's audit history to review how the agent arrived at its outputs. The file is organized into three sections:
| Section | Contents |
|---|---|
| Inputs | The shared field values passed to the task. These are the starting values the agent wrote into the spreadsheet. |
| Intermediate | Values used during calculation. These show the agent's work between inputs and outputs. |
| Outputs | The requested field values calculated by the spreadsheet. These are the values the agent fills in on the task. |
Updating an Excel Agent Task
To change how the agent calculates results, update the Excel file rather than rewriting the task prompt. In the blueprint, open the task and download the Excel file from the Documents field. You can edit:
- Formulas for intermediate values. For example, increase the points for a scoring category from 5 to 10.
- Hard-coded intermediate values. For example, update a risk threshold from 20 to 25 points.
- Formulas for outputs. For example, exclude a specific input from the final score calculation.
After editing, re-upload the file to the task's Documents field.
Limitations
- Only .xlsx files are supported. The .xls, .numbers, and .csv formats are not supported.
- Upload the Excel file directly to the task's Documents field. You can't pass a dynamic Excel file through a shared field.
- The agent can't insert documents, signatures, or data sources into the Excel file.
- Tabular data isn't supported as input or output.
