Case Study E-commerce · Amazon FBA 2026

Clarity for an Amazon seller

An operational-intelligence platform (Orbit), built end to end — from raw Amazon reports to a plain-language answer for every single product.

By Javier Aragón Navarro
HorizonLabs · Data & AI · Windsor, ON

The problem: Amazon shows you the costs, just not where they hurt

For an Amazon seller, the real number is the deposit that lands in the bank — the settlement. Amazon itemizes plenty of fees, but only part of the way. The costs that decide whether a product is worth keeping get lumped into bulk totals you can't trace to a single item. That's exactly where margin quietly disappears.

Broken down per product
Referral & selling fees
Fulfillment (FBA) fees
Returns & refunds
Core logistics charges
Hidden in one bulk total
Advertising spend, not split by product
Storage, aging & low-stock penalties
Inbound shipping into Amazon
So you can't see which product is bleeding
Phase 1 · True cost allocation

Every cost, assigned to the product that caused it

Orbit consolidates the settlement report, the fulfillment fee reports, the live inventory ledger, and the advertising data from Seller Central — then pushes those bulk totals down to the product level. For the first time the owner sees the real monthly profit and loss of each individual product, not a category-wide estimate.

Consolidates Settlement · Fulfillment fees · Inventory ledger · Seller Central Ads
Phase 2 · Discovery

Why a product that looks healthy still loses money

Amazon's calculator gives a best-case estimate, so a product can look profitable while the real deposit says otherwise. Orbit groups products into archetypes by commercial behaviour, price, and physical size, learns the normal spending pattern for each, and draws health bands around it — so the owner sees which cost is eating the margin (advertising, storage, returns), not just that something went wrong. With enough history, it also projects short forecast curves to time restocks before stock runs out.

The extra step · Just ask

With the data and AI wired together, you simply talk to it

Because the full data foundation sits behind an AI layer exposed over a custom MCP server, the owner can skip the dashboard and just ask — and get a fresh, up-to-the-moment answer:

"Which products do I need to restock this week?"
"How much would I lose if I don't restock these in 15 days?"
Built with  Apache Airflow · Python · PostgreSQL · Prophet · LangChain + Claude · FastAPI MCP server · Next.js

Have a business buried in its own data?

That's the kind of problem I build for. Let's talk about yours.

Get in touch