You looked at the numbers on Monday.
On Tuesday you did the same things anyway.
Not because you ignored them. Because the chart said conversion is down and that is not an instruction. The instruction is one layer further in and it needs two tables at once: this SKU converts fine but comes back at twice the category rate, so the third photo is setting the wrong expectation and the price was never the problem. That one is profitable only because a single search term is subsidising it. Reading across five exports is the part that never gets done, and it is the part that produces something you can act on.
What only shows up when the tables sit together
The product that is not a pricing problem
A thin margin reads like a pricing question right up until the returns data is next to it. Returns at twice the category rate on a product that converts normally usually means the listing or the photos are setting an expectation the box does not meet, and the fix is a rewrite rather than a discount. Discounting it would have made the margin worse and the returns identical.
The stock decision that is really a demand decision
Stock turn on its own says restock. Stock turn read against the ad spend that produced those sales sometimes says the demand was bought rather than found, which is a different decision with a different risk and a different amount of your cash tied up in it. Both numbers are already in your exports; putting them in the same sentence is the work.
An action list, not another chart
What comes out is short and specific with the evidence attached to each line: restock these two, pause this term, rewrite this bullet, replace this photo, leave the rest alone. The lines you accept become to-do items handed to whichever skill can carry them out, so the analysis and the change happen in the same place instead of a week apart.
It recommends. You decide.
The skill definition is explicit that final sourcing, stocking, ad spend and pricing decisions are not made for you, which is the right boundary for something reading your books.
Percentages without denominators do not get printed
A 40% improvement on a base of five orders is noise wearing a suit. Every rate comes with what it was computed over, and where the window is too short or the sample too small to support the comparison, that is said instead of the number being shown anyway.
Platform figures stay labelled as estimates
Displayed sales counts, best-seller ranks, heat scores and third-party estimates are indicators, not ground truth, and quietly folding them into a margin calculation is how a confident wrong answer gets built. Sources are stated per figure so you can discount the ones you do not trust.
No promise about the outcome
Sales, profit, ranking and ad return are not committed to, here or anywhere else in these skills. What is on offer is that you can see the reasoning and disagree with it on a specific ground, which is more than a dashboard gives you and less than a promise.
The same pass, every week
Scheduled runs matter here more than anywhere else, because two weeks are only comparable when both reports were produced the same way. A skill on a timer gives you that. A fresh prompt every Monday gives you two documents that quietly measure slightly different things and let you conclude whatever you were already inclined to conclude.
Before you point it at your books
What data does it need to be useful?
Less than you think. A business report and an advertising report get you most of the way. Adding returns reasons and a review export is what turns a margin observation into a cause you can act on. Excel, CSV, JSON and DOCX are all readable, so whatever your back office hands you is usually enough to start.
Is this a BI tool?
No, and the difference is worth being clear about. A BI tool gives you a chart you interpret. This gives you a written argument you can argue with, ending in a list of actions. If you already have dashboards, this sits after them rather than replacing them, and the connectors read the same sources your dashboards do.
Can it compare several stores at once?
Yes, and that is where the alignment work pays off, because four back offices produce four column layouts for the same idea. Connectors exist for Shopify Admin, Amazon Seller Central, eBay, Etsy, Walmart Marketplace, WooCommerce and others, each set up with an application you register, and CSV exports work with no setup at all.
Will it decide what to restock?
It makes a recommendation with the reasoning and the assumptions visible, and it does not place the order or commit you to a quantity. Final sourcing, stocking, ad spend and pricing decisions are explicitly outside what these skills do, because they depend on your cash position, your supplier terms and your risk appetite, none of which appears in any export.
Put this week's exports in one folder and read the list
Free, macOS and Windows. It writes the list; you decide what to ship.