Skip to main content
This walkthrough takes you from a file to a checked analysis. Use the sample CSV below, or a small CSV or Excel file you are authorised to upload. The sample contains fictional order data for June–August 2026. Each row is one order; amount is the order total in AUD excluding tax, not a unit price. Three orders are cancelled. There are no refunds. Substitute your own columns and business definitions if using another file.
Copy the CSV below into a plain-text file named sales-orders.csv, then upload it in step 2. You can also paste it into a spreadsheet and save that as CSV.
sales-orders.csv

​
1. Choose where to work

Sign in to Dawn. Check that you are in the intended organisation and workspace before uploading data. Open a suitable folder in Workspace, or create one using Create → Folder.

​
2. Upload your data

Choose Create → Upload dataset, select your CSV or .xlsx file, and complete the upload form. Give the dataset a recognisable name such as sales_orders. Wait for the import to finish, then open the dataset. Check that the headers, dates, amounts, and sample rows look right. See loading data if they do not. You should now have: a dataset you can open and inspect in the workspace.

​
3. Start a focused conversation

Choose New conversation. Use the + control beside the message box, labelled Add context or image, to attach the dataset. Ask:
Review the response before asking for analysis. If a column has a special meaning, explain it. For example, say whether the amount includes tax or refunds.
You should now have: an answer tied to a particular dataset and an explicit definition of sales.

​
4. Check the result

Expand the relevant tool results in the conversation. Check the source, filters, date range, units, and totals against a few known records. Ask follow-up questions when an assumption is unclear.
If you used the sample file, you should have 15 uploaded rows and 12 completed orders. Excluding cancelled orders, the monthly totals are: Do not multiply amount by quantity: the amount already represents the whole order. If Dawn reports different totals, ask it to show the filter and calculation. An answer can be well written and still use the wrong business definition. Correct the definition before building on the result.

​
5. Save something useful

Ask Dawn to create a dashboard or report from the checked analysis:
Open the resulting asset and verify its contents. A dashboard is useful for exploration; a report is useful for a written explanation. For work that needs to run on refreshed data, use a pipeline.

​
6. Save the knowledge worth keeping

A definition that applies beyond this conversation belongs in Context. Record the agreed sales definition, source dataset, exclusions, and effective date in the appropriate context area. You can add recurring agent work later with Jobs. First make sure the one-off workflow produces the result you want.