Upload, ask, follow up. A step-by-step walkthrough using an example e-commerce orders file, from a raw export to a canvas of answers you can share.
Key takeaways
- Spend two minutes cleaning the CSV first: clear headers, one row per record, real dates.
- Start with one headline question, then use follow-ups to narrow it down.
- Ask for a chart type or breakdown when you want one; otherwise the tool picks a sensible default.
- Check one number you already know before you share anything.
What you need
- A CSV file. Any export works: orders, subscriptions, ad spend, support tickets. The free Explore plan includes 10 MB of uploads a month and Team has no limit, and Excel files work too.
- A question you care about. "Why were sales lower last month?" is a better starting point than "show me the data".
- Five minutes.
This walkthrough uses an example orders export from an online store, with one row per order and these columns:
| Column | Example value | What it means |
|---|---|---|
order_id |
100482 | Unique order number |
order_date |
2026-08-14 | Date the order was placed |
customer_id |
C-20931 | Who placed it |
region |
West | Shipping region |
category |
Home & Kitchen | Product category |
channel |
Paid search | How the customer arrived |
order_value_usd |
84.50 | Order total in US dollars |
discount_usd |
10.00 | Discount applied |
status |
delivered | delivered, returned or cancelled |
Step 1: Prepare the file (two minutes)
A few small fixes make every answer more reliable:
-
Use clear column names.
order_value_usdis better thanvalorAmount (incl. tax) - USD. Keep them short, lowercase and without special characters. - Keep one row per record. One row per order here. Remove subtotal rows, blank rows and notes at the bottom of the sheet.
-
Put dates in one format.
YYYY-MM-DDavoids the 03/04 problem (March 4 or April 3?). -
Keep numbers as numbers. Remove currency symbols and thousands separators inside cells:
1250.00, not$1,250.00. - Drop columns you don't need, especially personal fields such as email addresses and phone numbers.
Step 2: Upload and let the columns be profiled
Create a free account, then upload the CSV. DashboardGPT profiles each column as it loads: it detects that order_date is a date, order_value_usd is a currency amount, region and category are categories, and order_id is an identifier that should be counted, not summed.
A sample orders file with 3,089 rows, ready a few seconds after uploading. Build a canvas is offered straight away.
Take ten seconds to check the Profile tab. It says what the import couldn't do cleanly: rows it couldn't read and left out, with their line numbers, columns it had to name because a header was empty or repeated, and columns that may have the wrong type, such as a zip code read as a number or amounts like $1,200 read as text.
The Profile tab for a clean sample file. Nothing needed fixing, so it shows the counts and how each column will be used.
If a type is wrong, fix it now: accept the suggestion on the Profile tab, or choose the pencil beside the type on the Schema tab. The file is imported again with the column read your way, and DashboardGPT warns you first about any chart the change would affect. Every later answer depends on it.
The Schema tab. customer_zip was read as text, so codes such as 02139 keep their leading zero. The pencil beside each type changes it.
Want the big picture before you ask anything? Choose Build a canvas when the upload is ready, and DashboardGPT lays out the charts that fit your columns. Turn a spreadsheet into a canvas of charts explains what you get.
Step 3: Ask one headline question
Start with the number that matters most. For an orders file, that is usually revenue over time:
Try it on the demo data: “What was total order value by month for the last 12 months?” Try a question like this on demo data
You get a line chart of monthly revenue on the canvas, with a short written insight underneath, for example that August was the lowest month since March and down about 12% on July.
That is a chart, not a paragraph. The written insight explains it and cites the figures, but the chart is the answer.
A headline question on the playground's retail sales demo data. The chart goes on the canvas, and the written answer appears in the chat.
Step 4: Follow up on the chart
Select the chart and ask the next question. Because the follow-up knows which chart you are pointing at, you do not need to repeat the dataset, the metric or the time window:
- Break August down by region. The drop is almost all in the West region.
- Now by channel, West only. Paid search orders fell by half; other channels were flat.
- Compare paid search orders per week in July and August. The fall starts in the second week of August.
Three follow-ups turn "revenue is down" into "paid search in the West dropped from the second week of August". That is something you can act on: check whether a campaign was paused or a budget ran out.
With the chart selected, the follow-up only needs the new part: break this down by region.
Step 5: Get the chart you want
The tool picks a sensible chart for each result, but you can ask for a specific one:
| You want | Ask like this |
|---|---|
| A trend | "... by week as a line chart" |
| A ranking | "Top 10 categories by revenue, as a bar chart, largest first" |
| A share of a total | "Share of revenue by channel this quarter" |
| A single number | "What was the average order value in August?" |
| A comparison | "Compare July and August revenue by region side by side" |
| A table | "List the 20 largest orders in August with region and channel" |
You can also refine a chart after it appears: "Show this as a percentage change", "Exclude cancelled orders" or "Only include orders over $50".
Step 6: Check the numbers
Before you share anything, confirm one figure you already know. If your store dashboard says July revenue was $412,000, ask "What was total order value in July?" and compare.
Every chart can switch to a table, so you can check its numbers against figures you know.
If it does not match, the cause is usually one of these:
- Returned or cancelled orders are included. Ask again with "only delivered orders".
-
Discounts are counted differently. Decide whether revenue is before or after
discount_usd. - Dates cut off differently. The export may use UTC while your dashboard uses local time.
- The file is not complete. Exports are sometimes capped at a row limit.
You can also open the chart's menu and choose How it's calculated. It shows, step by step, which columns the chart uses and the kind of value each holds, which rows it keeps, how it groups them and how each figure is worked out. Natural language to chart explains what to look for.
Step 7: Organise and share the canvas
Every answer stays on the canvas, so the analysis builds up as you go. Before sharing:
- Group related answers: the headline chart, the regional breakdown and the channel breakdown together.
- Delete dead ends that did not lead anywhere.
- Put the conclusion first: the chart that explains the drop goes at the top.
On the Team plan you can share the canvas with colleagues, so they see the answers and the questions that produced them.
Ten questions to try on an orders CSV
- What was total order value by month for the last 12 months?
- What is the average order value by channel?
- Which 5 categories grew fastest this quarter compared with last quarter?
- What share of orders were returned, by category?
- How many new vs returning customers ordered each month?
- Which region has the highest average discount?
- What day of the week has the most orders?
- Which customers placed more than 5 orders this year?
- How does order value differ between discounted and full-price orders?
- Which channel brings customers with the highest repeat rate?
Frequently asked questions
How big can the CSV be?
The free Explore plan includes 10 MB of CSV and Excel uploads a month, and Team has no upload limit. Warehouse connections for live data are coming soon on Team.
Do I need to clean the file first?
Not always, but two minutes of cleanup (clear headers, one row per record, consistent dates, numbers without symbols) makes answers noticeably more reliable.
Can I chat with an Excel file instead of a CSV?
Yes. Excel files are supported on every plan, the same way as CSV files.
Why did I get a different number from my dashboard?
The usual causes are returned or cancelled orders being included, discounts counted differently, time zone differences or an incomplete export. Ask how the number was calculated to see the query.
Can I ask for a specific chart type?
Yes. Add it to the question, for example "as a bar chart, largest first", or ask to change a chart after it appears.
Originally published on DashboardGPT Learn. DashboardGPT lets you ask questions of your data in plain English and get charts back. You can try it in the playground without an account.






Top comments (0)