Client build

Data integration for an Amazon FBA business · Amazon SP-API · n8n · US + Canada marketplaces

The SP-API Sync Engine.

Forty-five workflows that keep a seller’s back office fed (orders, returns, settlements, inventory, and fees) without anyone pulling a report by hand.

The back office runs on data Amazon doesn’t hand over.

The client’s inventory management system needs a continuous feed of everything Amazon knows (orders, returns, financial events, settlements, inventory, FBA shipments, fee estimates, catalog data) across two marketplaces. Amazon’s Selling Partner API has all of it, but makes you earn it: hourly-expiring tokens, per-endpoint throttling, pagination, and reports that are generated asynchronously and fetched by polling. Pulling any of this by hand doesn’t scale past day one.

One shape, repeated until it covers the business.

Roughly forty-five n8n workflows in a single workspace: scheduled syncs that keep the IMS fed daily, on-demand webhook tools the team calls for product research and fee estimates, gated FBA and listing operations, one-shot backfills, and two shared sub-workflows everything else reuses.

~45

n8n workflows in one workspace

data types synced: orders, returns, financials, settlements, inventory, shipments, fees, catalog

marketplaces live (US + Canada), a third built and staged

Schedule / webhook triggerdaily and hourly syncs · on-demand tools01

Shared LWA token sub-workflowone place to rotate Amazon credentials02

IMS logintenant-scoped bearer, so data lands where it should03

Pull from SP-APIthrottled · paginated · async reports polled04

Reshape to the IMS contractraw TSV reports become clean JSON05

POST /integrations/…into the IMS · failures email immediately06

the shape every sync shares: learn one workflow, understand forty

Six decisions that keep it boring.

Boring is the goal. A sync engine earns its keep by never being interesting at 7am.

One token fetcher, everywhere

Amazon access tokens last an hour. Instead of forty-five workflows each re-implementing OAuth, every one calls a single shared sub-workflow that does the LWA refresh-token exchange. When credentials rotate, they rotate in one place.

Reports that aren’t ready when you ask

Much of Amazon’s data only comes from the asynchronous Reports API: request a report, get an ID, poll until it’s done, then download and parse the raw TSV. That poll loop is factored once and reused, so every report-based sync behaves the same way.

The login decides where the data lands

The client’s IMS is multi-tenant, and the identity a workflow posts with decides which tenant receives the data. Data had historically landed in the wrong tenant through a shared support login. The fix was organizational as much as technical: a dedicated ingester identity, used by every sync, so the question can never come up again.

History is a workflow, not a script

Loading the past is as important as syncing the present. One-shot backfill workflows exist for orders, returns, settlements, financial events, the inventory ledger, and FBA shipments. The daily syncs also have gap-fill twins that recover any missed window.

Failures email, silence doesn’t

Every sync routes failures through a shared error-handler that emails immediately with the workflow name and error. A quiet morning means the pipeline ran, not that nobody was looking.

Destructive operations stay gated

Listing creation and deletion, label printing, and FBA inbound plan tools exist in the same workspace, deliberately gated for manual, on-demand use. A scheduled workflow should never be able to delete a listing on its own.

A system you can hand over is a system you actually own.

The engine ships with a sixteen-section engineering handover: the architecture as a mental model, a per-workflow reference, an endpoint reference, an operations runbook, a known-issues register that states open risks plainly, and a first-week checklist for the next engineer. If it only runs while I’m holding it, it isn’t finished.

The IMS stays fed. Nobody pulls reports.

Orders, returns, financial events, settlements, inventory, shipments, and fees land in the IMS daily across both marketplaces, in the right tenant, with history backfilled. When something fails, the team knows by email before anyone notices a gap. The next marketplace, Walmart, is already built and staged against the same pattern.

Still exporting Seller Central reports by hand?

Tell me which reports your team pulls every week. I’ll tell you what a sync engine for your operation would look like, and what it wouldn’t touch.

Book a call See more builds