Why does your agent's revenue number differ from finance's?

One question about Q3 revenue gets three answers from the same tables. Run the query that gets the first one and see what it counted.

Sample page. Your Maestro, an AI, answers your free questions about this page.

Before you start, one quick question. It doesn't count for anything: it's your starting point, and you'll see the same question again at the end of the course.

Check yourself, Starting point: An agent can query a shop's raw tables (orders, order items, customers, refunds) but has nothing else. Which of these can it NOT find out from the data alone?

Mark where you start: at the end you'll see how far you've come.

Ridgeline Coffee Roasters is a small US roaster (made up, like everything in this course) that sells beans and gear online and to cafes. Its shop app exports six tables to the warehouse every night. Finance asks one question: what was net revenue by region in Q3 2026?

The same tables give three different answers, depending on who asks and what they know:

Who answered	Q3 net revenue
A query that joins everything and sums it	$88,435.40
Claude Code, given the raw tables	$76,137.00
Finance, using the model you build in this course	$75,638.50

The second row is a real run: Claude Code (Sonnet 5.5) answered this question on these tables on October 31, 2026, and you'll read its whole session on the next page. Nobody made a math error. Each answer counted different rows.

Try it

The query below is the first answer: it joins orders to their lines and adds up every line plus the shipping on the order. Run it and you get $88,435.40. Then fix the two problems you can see in the data itself: keep only completed orders, and count only what the customer paid for the items (shipping is not revenue). You're done when the total reads 78,132.00.

Your turn

Run the query to see $88,435.40. Then keep only orders whose status is 'completed', and stop adding shipping_usd. You're done when one row comes back with revenue 78132.0.

Tables: orders (order_id, customer_id, ordered_at, status, shipping_usd), order_items (order_item_id, order_id, product_id, quantity, unit_price_usd)

orders · 640 rows

Column	Type
order_id The order. Unique.	VARCHAR
customer_id	VARCHAR
ordered_at UTC	TIMESTAMP
status completed or cancelled	VARCHAR
shipping_usd Shipping charged on the whole order, in US dollars	DOUBLE

First rows

order_id	customer_id	ordered_at	status	shipping_usd
R1558	C003	2026-06-15T09:49:37	completed	0.00
R1274	C020	2026-06-15T12:02:40	completed	0.00
R1060	C042	2026-06-15T15:07:19	completed	5.95

order_items · 1,202 rows

Column	Type
order_item_id One line of an order	VARCHAR
order_id The order this line belongs to (raw_orders.order_id)	VARCHAR
product_id The product sold (raw_products.product_id). P13 is a holiday blend deleted from the catalog	VARCHAR
quantity	BIGINT
unit_price_usd The price the customer paid for one unit	DOUBLE

First rows

order_item_id	order_id	product_id	quantity	unit_price_usd
L00001	R1001	P10	1	7
L00002	R1002	P07	1	34
L00003	R1003	P09	1	119

Your query (SQL)

At 78,132.00 you've done what any careful analyst does with a new table: looked at the columns and used what they say. Subtract the $1,995.00 in refunds on those orders and you reach $76,137.00, which is exactly where Claude Code landed.

It's still $498.50 too high. The difference comes from something no column tells you, and that's the subject of this course: the rules your company runs on are not in the data, so an agent can't follow them until you put them where it reads. Where does the other $498.50 come from?

Check yourself: Your agent and finance disagree about Q3 revenue by $498.50. Both used the same tables and neither made an arithmetic mistake. What's the most likely cause?

This page is from the course; the rest waits in it.

Buy the course What the course includes