# Example 4: Agent V2 — Intelligent Data Analysis Agent

> Analyze a reproducible sales sample, calculate product units and channel advertising ROI, and inspect downloadable artifacts.

Source: https://doc.fastgpt.io/en/guide/getting-started/examples/data-analysis

## Example 4: Agent V2 — Intelligent Data Analysis Agent

### 4.1 Use Cases

Agent V2 chooses processing steps from the task and data. This example runs Python in the virtual machine to produce a sales summary and channel revenue chart.

### 4.2 Preparation

Download [sales.csv](/examples/sales.csv). The four fictional sales records use one consistent currency unit:

```csv
product,channel,units,revenue,ad_cost
A,Search,10,1000,200
B,Search,5,750,150
A,Social,8,800,400
B,Social,12,1800,600
```

You need application creation permission, a model supporting tool calls, an available virtual-machine service, and model quota. The captured run uses qwen3.8-flash. The execution environment must provide or install Python plotting dependencies. Network access, resource limits, and isolation follow the execution-service configuration.

### 4.3 Configuration Steps

1. Go to **Studio → Agents**, click the first **New** card in the list (or **Create your first Agent** in the center of the page when the list is empty), select **Chat Agent V2**, enter `Intelligent Data Analysis Agent` in the name box, and click **Create** to the right of the name box.

   ![Select Chat Agent V2 on the Create Agent page](https://cdn.jsdelivr.net/gh/labring/fastgpt-img@main/optimized/current/app-types.png.webp)

2. Choose a model supporting tool calls and enable **Computer**. This example uses built-in file and command tools; **Skills** can remain empty.

3. Enter this prompt:

   ```text
   Use Python in the virtual machine to analyze the uploaded sales CSV. Calculate units by product and revenue, ad_cost, and ROI by channel. ROI = (revenue - ad_cost) / ad_cost. Save channel_summary.csv and a revenue chart as revenue.png. Return download links and concise results.
   ```

4. Upload `sales.csv` in debug preview and wait for completion. Ask: “Analyze the attached file: aggregate units by product and revenue, advertising cost, and advertising ROI by channel. Generate a summary CSV and revenue chart.”

5. Expand Processed to inspect file reads, commands, and their results. For missing dependencies, check installation output or configure a [startup script](/en/guide/build/agentv2/vm#virtual-machine-lifecycle).

6. Click the **View Sandbox files** icon button at the upper right of debug preview. Check `channel_summary.csv` and `revenue.png`. Preview and download the image in the file panel; use the file link in the answer to download the CSV.

![Sales analysis configuration and results: product units and channel advertising ROI](https://cdn.jsdelivr.net/gh/labring/fastgpt-img@main/optimized/current/data-analysis-result.png.webp)

![Generated files and revenue chart in the virtual-machine file panel](https://cdn.jsdelivr.net/gh/labring/fastgpt-img@main/optimized/current/vm-files.png.webp)

### 4.4 Verification of Results

| Metric                          | Expected value            |
| ------------------------------- | ------------------------- |
| Product A units                 | 18                        |
| Product B units                 | 17                        |
| Search: revenue / ad cost / ROI | 1750 / 350 / 4.00 (400%)  |
| Social: revenue / ad cost / ROI | 2600 / 1000 / 1.60 (160%) |

ROI is `(revenue - ad_cost) / ad_cost` and includes advertising cost only. Search has higher advertising efficiency; Social has higher revenue. These four rows validate calculations; business decisions require a larger sample and complete costs. Report zero-cost groups separately because this ROI formula would divide by zero.

The captured run was checked against the answer, downloaded CSV, and chart. The CSV has two channel summaries; the chart bars are 1750 and 2600. Planning steps and wording may vary, so verify the calculations and artifacts. Save and publish the configuration, then repeat the test in a new conversation.

### 4.5 Business Value

Keep the input, metric definitions, and independently calculated expected values as a regression sample. Repeat it after model, prompt, or execution-image changes to check file handling, calculations, and usable artifacts.

[Return to getting started](/en/guide/getting-started)