Julius AI: Analyze a Sales Spreadsheet and Check the Results

Julius AI: Analyze a Sales Spreadsheet and Check the Results

A sales summary is useful only when you can explain which transactions it includes. Before accepting a revenue total or a polished chart, establish the reporting period, define the measure, and reconcile the answer to the underlying rows.

Julius AI offers a conversational way to do this work. Its Chat with your data product page describes natural-language analysis, follow-up questions, chart and table outputs, and access to the underlying code. This guide gives you a sequence of prompts and a small sales example with independently calculated answers.

About this walkthrough: The records below are fictional. The expected results are calculated from those records, not presented as outputs from a live Julius session. Use them to check your own run before adapting the workflow to a business file.

1. Define the sales question before uploading

Start with one question: What was our completed-order net sales total for June, and how was it split between East and West? For this exercise, “net sales” means units multiplied by unit price, less the order-level discount and recorded refund. This is a tutorial metric, not a rule for your organization’s accounting.

Write down these assumptions before asking for analysis:

• Each order occupies one row. Order IDs are unique after confirmed duplicate records are removed.
• All amounts use the same currency, USD. Unit price is per item; discount and refund are totals for the order.
• Include completed orders dated from June 1 through June 30, 2026. Exclude cancelled orders and other dates.
• Subtract the refund stored on an included order. The example has no separate refund transaction date.
• A blank discount is unknown, not automatically zero. Keep that order outside the reportable subtotal until clarified.
• There are no tax, shipping, or cost fields. Do not label the result profit or add those amounts.

For a real export, confirm whether rows represent orders, order lines, payments, or refund events. If an order has several legitimate line items, deleting rows solely because their order IDs match would change the dataset incorrectly. Define a suitable row key first.

2. Prepare and upload a small practice file

Copy the following text into a plain-text file named sales-practice.csv. Retain the blank discount and repeated A006 row: they are intentional checks. Keep the raw file unchanged and make cleaning decisions on a working copy.

order_id,order_date,region,status,units,unit_price,discount,refund
A001,2026-06-03,East,Completed,2,100,10,0
A002,2026-06-04,West,Completed,1,200,0,0
A003,2026-06-05,East,Completed,3,50,0,0
A004,2026-06-07,West,Cancelled,2,100,0,0
A005,2026-06-08,East,Completed,1,200,20,0
A006,2026-06-09,West,Completed,2,50,0,50
A007,2026-06-10,East,Completed,1,100,,0
A006,2026-06-09,West,Completed,2,50,0,50
A008,2026-07-01,West,Completed,1,100,0,0

Sign in to Julius, open its chat, and attach the practice file using the available file-upload control. Julius’s official spreadsheet guide lists CSV and Excel support and explains that you can identify individual workbook tabs in your prompts. If you use a workbook instead, name the intended tab explicitly.

Begin with inspection, not a request for recommendations:

Inspect sales-practice.csv without changing it. Report the filename,
row count, columns, inferred data types, date range, status values,
and region values. Identify blank numeric cells and exact duplicate
records, showing their order IDs and original data-row positions.
Do not fill blanks, remove rows, or calculate a sales total yet.

The file should contain nine data rows and eight columns. Expect one blank discount on A007 and two identical A006 records. If the reported structure differs, resolve the import before continuing. For a broader selection of analysis tasks, explore the Data & Analytics Prompts collection.

3. Make every cleaning decision explicit

For this fictional file, the repeated A006 record is a duplicate export row. Keep its first occurrence. A007’s missing discount remains unresolved. Use those decisions in a second prompt:

Use a working copy and preserve the original data. Keep the first
occurrence of the exact duplicate A006 record and remove the second.
Apply these rules in this order:
1. Remove only the confirmed exact duplicate.
2. Exclude orders whose status is not Completed.
3. Keep dates on or after 2026-06-01 and before 2026-07-01.
4. Quarantine any remaining row missing units, unit_price, discount,
   or refund. Do not replace missing values with zero.
Produce a row-count reconciliation and an exclusions table with
original data-row position, order_id, reason, and retained/excluded status.
Do not overwrite the source file.

Applying exclusions in a stated order makes the count reconciliation unambiguous. A row is assigned to the first applicable exclusion, so it is not subtracted twice.

Expected row-count reconciliation for the fictional file

Stage	Rows removed at this stage	Rows remaining
Original file	0	9
Remove second A006 occurrence	1	8
Exclude cancelled A004	1	7
Exclude July order A008	1	6
Quarantine A007: discount unknown	1	5

The result is a reportable subtotal from five orders, with one otherwise eligible June order unresolved. Label it that way; calling it a complete June total would hide the missing information.

4. Calculate the metric and retain the row-level evidence

Use column names and units in the calculation request. Avoid shorthand such as “revenue after adjustments” when the adjustments have not been defined.

For the five retained orders, calculate:
gross_sales = units * unit_price
net_sales = gross_sales - discount - refund
Discount and refund are order-level USD amounts; do not multiply them
by units. Return a row-level table containing all input values and both
calculated columns. Then summarize gross_sales, discount, refund,
net_sales, and distinct order count overall and by region.
Calculate average net sales per included order as total net_sales
divided by the distinct included order count. Show the calculation code.
Label all summaries as excluding A007 pending discount clarification.

Independently calculated results in USD

Order	Gross sales	Discount	Refund	Net sales
A001	200	10	0	190
A002	200	0	0	200
A003	150	0	0	150
A005	200	20	0	180
A006	100	0	50	50
Total	850	30	50	770

The independent arithmetic is 850 - 30 - 50 = 770. East contributes 190 + 150 + 180 = 520; West contributes 200 + 50 = 250. The two regions reconcile to USD 770. Average net sales per included order is 770 / 5 = 154.

These are expected answers for this file and these rules. If your result differs, compare retained order IDs before trying to repair the final total.

5. Check the method as well as the answer

Use the code made available with the analysis to trace the row filter, duplicate handling, date boundaries, missing-value treatment, and formula. Ask for explanations of any operation you cannot interpret. A matching headline total alone is not enough to show that every row was handled correctly.

Diagnostic clues for this practice dataset

Result you see	What to inspect first
USD 820	The second A006 may still be included, adding USD 50.
USD 870	A007 may have been treated as zero-discount, or July order A008 may be included. Either adds USD 100; inspect the row IDs.
USD 970	Cancelled A004 may be included, adding USD 200.
USD 850	The result may be gross sales before discounts and refunds.

These clues assume other steps are correct; combinations of mistakes can produce different totals. Reconcile one stage at a time. Repeating the same question in Julius is useful for clarification, but it is not an independent check. For a larger file, calculate a control total in your spreadsheet and manually verify several selected rows.

Reconcile the output against these control totals: five included
orders, gross sales 850, discounts 30, refunds 50, net sales 770,
East 520, West 250. If anything differs, locate the first processing
stage that diverges. Show the affected source rows and explain the
cause. Do not change the rules just to force the totals to match.

6. Choose a chart that answers the question

For this example, use a two-bar regional comparison. It directly answers how the reportable subtotal is split. A trend chart would distract from the exercise, which contains only a handful of orders and an unresolved value.

Create a bar chart from the verified regional summary: East 520 USD,
West 250 USD. Start the value axis at zero, label each bar, and title
it "June 2026 net sales by region". Add a note: "Five completed orders;
excludes A007 pending discount clarification." Use the calculated
summary table as the chart data. Do not add forecasts or causal claims.

Check that the chart values match the table and that the exclusions note is visible. The supported conclusion is that East contributes more to this five-order subtotal. The file contains no targets, traffic, campaign exposure, or costs, so it cannot establish why that happened or which region is more profitable.

7. Export a reviewable result and a bounded summary

Julius documents CSV and Excel downloads in its spreadsheet export guidance. Request separate outputs for retained rows with calculated columns, excluded rows with reasons, and the regional summary. Keep the raw input alongside them.

Prepare the retained-order table, exclusions table, and regional
summary for download. Include the metric definition and applied rules
in a short method note. Write a three-sentence sales summary using only
the verified results. Mention the unresolved A007 discount. Separate
observations from proposed follow-up questions; do not invent causes.

A suitable summary would be: “The five included completed orders contribute USD 770 in net sales for June 2026. East contributes USD 520 and West USD 250; average net sales per included order is USD 154. A007 remains excluded pending discount clarification, so this is a reportable subtotal rather than the complete June total.”

For the next review, use the Sales Prompts collection to frame questions about targets or customer segments. If the verified findings feed a wider business discussion, the Business Strategy Notes to SWOT Analysis Prompts collection can help organize those notes. Keep every strategic interpretation separate from what the sales file actually demonstrates.

Before you reuse the workflow

Replace the practice assumptions with your actual row key, reporting dates, currency treatment, refund convention, and metric definition. If the source is a multi-line order export, count distinct orders for an order-based average. If the business measures refunds by refund date, obtain those dates instead of adopting this example’s order-based convention.

Finish only when the retained rows, exclusions, subtotal, regional breakdown, and chart agree. Resolve missing inputs explicitly and rerun the same rules when corrected data arrives.