AI for Amazon sellers splits into six jobs: writing listings, setting ad bids, repricing, answering customers, forecasting demand, and analyzing profit on reconciled financial data. Five of those make a seller faster at something. Only the sixth answers "am I making money on this SKU," and it can only answer correctly when the data underneath it carries landed cost, every Amazon fee, ad spend, and refunds at the SKU level. An AI reading a blended P&L will give you a confident number that is wrong.
The distinction is worth an hour of your time, because the vendors selling the first five jobs are loud, and the sixth job is where the money leaks.
Models write titles, bullets, A+ copy, and image prompts. Amazon offers its own version inside Seller Central. The input is your product attributes and, sometimes, competitor listings. The output is text. This job never sees your cost of goods, so it cannot tell you whether the listing it improved is worth selling.
Bid tools read campaign reports (impressions, clicks, ad-attributed sales) and move bids toward a target ACoS. Amazon's own advertising guide defines ACoS as ad spend divided by ad revenue, expressed as a percentage. A bid tool optimizes that ratio. It does not know your margin after fulfillment fees and COGS, so it can drive a campaign to a 25 percent ACoS on a SKU whose contribution margin before ads is 22 percent. That campaign loses money on every attributed order, and the tool reports success.
Repricers watch the Featured Offer and competitor prices and move yours. They see your floor price if you set one. They do not calculate the floor; you do, and if your floor came from a spreadsheet built last quarter, the repricer will hold a price that has since dipped below breakeven when a fulfillment fee band changed.
Drafting replies to buyer messages and review responses. Useful, low risk, and unrelated to profit.
Models extrapolate sell-through from sales history and seasonality. Good ones factor lead time and inbound stock. None of them know that your supplier lost a production line or that you are about to sign a wholesale account. Forecasting AI works on units; it says nothing about whether those units earn anything.
An AI reading books where each Amazon settlement has been split into its transaction lines, where COGS runs on FIFO layers with freight and duty allocated, and where ad spend and refunds sit against the SKU that caused them. This one can answer "which SKUs lost money after fees last month," and it is the only category that can.
Illustrative numbers. Take a seller in Home and Kitchen, where the referral fee is 15 percent of the total sales price per Amazon's published pricing page on sell.amazon.com. The FBA fulfillment fee figures below are illustrative; Amazon sets fulfillment fees by product size tier and shipping weight, and the current schedule is on Seller Central.
The seller asks: which SKUs lost money after fees last month?
SKU A, a silicone utensil set at $24.99
Contribution: $15,993.60 minus $2,399.04 minus $3,040.00 minus $2,150.00 minus $949.62 minus $4,736.00 equals $2,718.94. That is 17.0 percent of revenue. The SKU earned money.
SKU B, a stackable storage bin at $32.99
Contribution: $13,525.90 minus $2,028.89 minus $3,649.00 minus $3,380.00 minus $1,715.48 minus $4,592.00 equals negative $1,839.47. The SKU lost money.
Both SKUs look fine on a Business Report. SKU B had more revenue per unit and a higher average selling price. Its ad-attributed sales were $6,200 against $3,380 of spend, a 54.5 percent ACoS that a bid tool would flag as high but not catastrophic. The bin's problem is that a 12.7 percent return rate and a large-standard fulfillment fee eat the gross margin before ads are even counted. No listing tool, bid tool, or repricer would surface that, because none of them hold the COGS or the refund lines.
The full walkthrough of every fee that sits between gross sales and contribution is in how to calculate true Amazon profitability after every fee.
An AI answers from what it can see. If your Amazon settlement posts to QuickBooks as one deposit, the AI sees one deposit. If COGS is a monthly journal entry from a spreadsheet, the AI sees an average that hides the fact that the last PO landed at a higher unit cost. If ad spend is a single invoice from Amazon Ads, the AI cannot assign it to SKU B.
Reconciliation is the unglamorous prerequisite. ConnectBooks pulls each Amazon settlement into QuickBooks or Xero at the transaction level, so referral fees, fulfillment fees, refunds, reimbursements, and storage charges each land in their own accounts, tied to the SKU and the period. FIFO costing runs per unit as goods sell, with landed cost allocated. The profit reporting built on that data is what makes a question like "which SKUs lost money" answerable in the first place.
Skip that step and every AI in this article, including the sixth category, produces a number that looks precise and is not.
Once the data is reconciled, the analysis job changes shape. Instead of exporting to a spreadsheet and building a pivot, a seller types the question. Crunch, the AI analyst inside ConnectBooks, reads the seller's reconciled Amazon, Shopify, Walmart, TikTok Shop, and eBay data and works each question through the same five steps: what changed, over which periods, on which products, why, and what to do about it.
For the question above, that means Crunch would list SKU B, quantify the loss, attribute it to the return rate and the fulfillment fee band, and recommend a next step such as testing a price change or reviewing the returns for a packaging defect. A seller who wants to go further asks a follow-up: did SKU B lose money in the prior quarter too, or is this new?
This is the section a vendor page tends to skip.
It cannot fix bad data. If landed cost is missing for a SKU, the AI either flags the gap or computes margin on purchase price alone. The second outcome is worse than no answer. Ask any tool you evaluate what it does when COGS is blank.
It does not know your supplier. It cannot tell you that the return rate on SKU B comes from a batch with a cracked lid, or that your factory has a four-week backlog. It can point at the SKU and the return count; you do the phone call.
It takes no actions in Seller Central. Crunch answers and recommends. It does not change a price, pause a campaign, or file a removal order. Those stay with you.
It carries no accountability. A recommendation to liquidate is a recommendation. The decision, the tax treatment of the write-down, and the conversation with your CPA remain yours.
A pasted CSV in a general-purpose chatbot has an additional failure mode: the model can invent totals. Reconciled books with a trace from every figure back to a settlement line are the guard against that.
If listing copy is slow, buy a listing tool. If bids eat your afternoons, buy a bid tool. Neither will tell you whether the business is profitable by product, and both can make a loss faster and more efficiently. Put the reconciled data layer in first, then ask it the questions that decide what you reorder, what you cut, and where the next dollar of ad spend should go.
Running an e-commerce business comes with plenty of challenges, but ConnectBooks is here to make your life easier. With real-time insights, seamless integrations, and detailed tracking of your profitability and inventory, you can stay ahead of the game. Whether you’re selling on Amazon, Shopify, Walmart, TikTok or eBay, ConnectBooks helps you manage your finances with 100% accuracy and confidence, so you can focus on growing your business.
Ready to level up? Start making smarter, data-driven decisions every step of the way. Try ConnectBooks Free Today or Schedule a Demo
