5 Amazon Profitability Questions to Ask an AI with Live Account Data

6 min read

Product research ends the day the first unit sells. After that the questions change shape. They stop being about the market and start being about your own account: what this SKU actually earns, where the ad budget went, and why last week looked different from the one before.

A general chatbot cannot answer these questions because it never sees your numbers, and the numbers sit in different places: settlement reports in Seller Central, spend in the ad console, landed costs in a spreadsheet, and VAT somewhere else again. Nobody has joined them up, so nobody can ask a question of them.

An MCP connection to your own account closes that gap. Instead of exporting reports and pasting them into a chat window, you ask the question, and the assistant reads the numbers behind it.

Market data tells you what to sell. Account data tells you what you are making.

Before launch, the signals that matter are public. Prices, demand, review counts, and search terms. They describe a market, and tools like the AMZScout PRO AI Extension and the Product Database supply them with sales estimates, sales history, price history, and competition level for a product before you buy any inventory.

After launch, the signals that matter are private. Your cost per unit, your inbound freight, your advertising by campaign, your refund rate, and your storage bill. None of it appears on a product page, and no amount of general knowledge will reconstruct it.

Both stages benefit from AI, and they need different sources. On the market side, the AMZScout Skill + MCP gives AI agents access to Amazon product data. On the account side, the five questions below only start working once the assistant can read your own account.

1. What does this SKU actually earn?

Revenue minus Amazon fees is not profit. A finished number has to carry the cost of goods, inbound freight, storage, returns, promotions, VAT where it applies, and the advertising attached to that product. Many sellers can produce that figure for the whole account once a month. Producing it per SKU, on demand, means joining several reports.

Example prompt: "Give me contribution margin per SKU for the last 90 days, after cost of goods, Amazon fees, refunds, and ad spend. Sort by total margin and show me the ten products that contribute the most."

The answer is only as good as the cost inputs behind it. Load your real landed costs once, and every answer afterwards inherits them. Skip that step and you get a confident number that happens to be wrong.

2. Which products lose money once ads are counted?

A product can look healthy on gross margin and still be underwater after advertising. Instead of asking what your ACOS is, ask which products run above their break-even ACOS, the level at which a product stops making money. Break-even ACOS equals the product's margin before ad spend, so the threshold is different for every SKU.

Example prompt: "Which of my products had an ACOS above their break-even ACOS last month? Show the gap, the spend involved, and what it cost me in contribution margin."

Some of those products are deliberate: a launch, a defensive bid on your own brand terms, and a push into a new keyword set. Asking lets you separate the deliberate ones from the ones nobody has looked at since they were set up.

3. What changed last week, and why?

Most reporting tells you that a number moved. The hard part is attribution: whether sales fell because a competitor cut price, because you were out of stock for two days, because a campaign budget ran out at noon, or because a listing lost the Buy Box.

Example prompt: "Sales are down 18% week over week. Find the products responsible, and for each one tell me what changed in price, stock, ad spend, conversion rate, and Buy Box share."

An assistant can check a dozen candidate causes in the time it takes you to open the first report, then hand back the two that actually explain the move.

4. Where is ad spend going that it should not?

Search term reports are long, and the rows that matter are the ones nobody scrolls to: terms that spend steadily and convert rarely, terms you already rank for organically, and terms that belong to a different product entirely. To see which terms competitors' products rank for, use AMZScout Reverse ASIN Lookup.

Example prompt: "List search terms from the last 60 days with more than $100 of spend and no orders. Group them by campaign and tell me which ones to add as negatives."

Negative keywords are a low-cost optimization that sellers often postpone because pulling the list is tedious. Asking a question instead of running a report removes that step.

5. What is about to run out, and what will it cost?

A stockout on a low-margin product is annoying. A stockout on the product that carries the account is a bad quarter. Rank by days of cover and contribution margin together, not by units alone.

Example prompt: "Which products will run out of stock in the next 45 days, ranked by the contribution margin at risk rather than by units? Include what is already inbound."

The stockout date is only half the answer. The other half is which dates you can afford to miss, and that decides where the next purchase order goes. The same logic works on the other side of the market: AMZScout Stock Stats shows competitors' inventory, so you can see when their stockouts open a window for your product.

Where the Nova MCP fits

Nova connects Amazon Seller Central accounts to Claude, ChatGPT, and other AI assistants through an MCP server across 21 marketplaces. The assistant reads your sales, fees, advertising, inventory, and search query data with your own cost inputs already applied, so a profit figure comes back finished rather than as an export you still have to reconcile.

It works in the other direction too. You can update cost inputs, log a price change or a new pack size, and have the effect measured against the weeks that follow, without leaving the conversation.

Good answers need your numbers, not general ones

The quality of an AI answer depends on the data behind it. For product research, that means live marketplace signals. For running the business you already have, it means your own account with your own costs applied. With the right data in place, the assistant answers these questions directly instead of sending you back to report exports. Start from the product side with the AMZScout PRO AI Extension, then connect your own account once the first units sell.

FAQs

Can ChatGPT or Claude read my Amazon Seller Central data?

Not on their own. They need a connection to your account, which is what an MCP server provides. Once connected, the assistant queries your sales, fees, advertising, and inventory directly instead of working from files you paste into the chat.

How is this different from asking AI to analyse a spreadsheet export?

In two ways. The data is current rather than whatever you last exported, and it is joined up, so costs, fees, advertising, and stock sit in one model. A profit figure comes back complete instead of needing reconciliation afterwards.

What do I need to set up before the answers are useful?

Your cost inputs. Cost of goods at minimum, then inbound freight, any self-fulfillment costs, and VAT categories if you sell in Europe. Profit answers are only as good as the costs behind them.

Does it work for Vendor Central as well as Seller Central?

Yes. 1P accounts cover shipped revenue, purchase orders, Amazon deductions, and replenishment signals, alongside 3P accounts in the same workspace.

Can the assistant change anything inside my Amazon account?

No. It reads Amazon data and writes only to your own Nova workspace, for example, updating cost inputs or logging an event. Actions inside Amazon stay inside Amazon.