Most people searching for an "Amazon database" want a dataset they can download and query at leisure.
No such thing exists, and not because nobody built it. Two structural facts decide it: data splits along authorization boundaries into three blocks that nobody holds all of, and it is a stream rather than a store — every number is only valid for a while.
First: data splits by whose it is
| Block | Who can get it | Typical contents |
|---|---|---|
| Your own store's | Only you, via seller account authorization | Orders, inventory, settlements, ad spend, search term reports |
| The marketplace's public side | Everyone | Products, prices, ranks, reviews, categories, keywords |
| Nobody's to get | No one | A competitor's real order volume, their backend terms, buyer identity |
The third row is not "not yet" — there is no public channel by design. When a plan depends on that row, what changes is the plan, not the data source.
Rows one and two come from completely different authorization models, so no endpoint provides both — which is the direct reason "one database holding everything" does not work. How to choose an Amazon data API breaks down what each road actually returns.
Second: what the public side actually contains
The public block is not one big table. It is a set of endpoints split by business domain — 46 of them currently, and the density varies a lot:
| Domain | Endpoints | What it answers |
|---|---|---|
| Market / category | 14 | How big a category is, who is in it, how it is distributed |
| Traffic / keyword flow | 6 | Which terms give a product visibility, and where |
| ASIN level | 5 | One product's detail, trends and rivals |
| Brand | 4 | Trademark and brand-level information |
| Product filtering | 3 | Shortlisting products by condition |
| ABA search terms | 3 | Published search term ranks and their movement |
| Sales prediction | 2 | Turning a rank into a sales magnitude |
| Keyword mining and conversion | 4 | A term's competition, bids and conversion |
| Reviews | 1 | Review content, stars and source flags |
Market takes 14, by far the densest block. That reflects something real: public data is best at answering "what is this market like", not "what exactly is this one competitor doing".
How to chain these by business scenario: The complete guide to Amazon data APIs.
Third: it is a stream, not a store
This is the deepest problem with "download a dataset". The same field can go stale orders of magnitude apart:
| Field | How fast it expires | What that means |
|---|---|---|
| Price, coupons | Hours | Yesterday's price cannot price today |
| BSR, rank | Days | An overnight rank shows direction only |
| Review count, rating | Accumulates daily | A single point says little; the rate says more |
| Sales estimates | Follows rank | When rank moves, this moves |
| Category structure, seller mix | Months | Worth storing monthly |
| Title, brand, launch date | Effectively fixed | Store once |
So "keeping a copy of the data" always means keeping a snapshot, not the data. Whether a snapshot is useful depends entirely on whether you recorded when it was taken — which is why every template we ship gives collection date its own column.
So can you keep a copy?
You can, and you should. But layer it by the table above: store the slow fields, fetch the fast ones live, refresh the middle on a schedule.
How to make that split and estimate call volume: Five decisions before you integrate. The test is simple: if a field expires before you use it, storing it only manufactures stale data.
Three things no data source provides
A competitor's real order volume. The marketplace publishes no unit sales, so every monthly figure is derived from BSR — the est in the field name is the API's own label. What Amazon sales data can and cannot tell you.
A competitor's backend search term report. Authorized data in their seller account. The reverse lookup returns terms that actually appeared in public results — different source, no correspondence.
Causation. Data shows rank fell. It does not show why. That needs your own change log.
Questions
Is there a ready-made Amazon dataset to download? Public-side data is reachable through endpoints, but it comes back as a snapshot per query rather than a static dataset. When you genuinely need "a copy", it usually means storing endpoint results under your own semantics.
What does "Amazon big data" mean? Usually the second block — aggregated data from the public side. Its value is in relative comparison and trend, not in single-point precision.
How far back does history go? It depends on the field. History endpoints generally accept a past calendar month, but check each endpoint's documentation rather than assuming any arbitrarily old month exists.
What if two sources disagree? Align three things first: marketplace, category level, collection time. Public fields — price, BSR, rating count — should agree, and disagreement there is a semantics problem. Sales figures are estimates, so disagreement is normal. Amazon data analysis covers the full method.