Amazon Ads reporting data does not stick around. Sponsored Products data is retained for roughly 95 days. Sponsored Brands and Sponsored Display sit at roughly 60. After that, it is not archived somewhere inconvenient. It is gone from what you can pull.
Most sellers discover this at exactly the wrong moment: planning for Prime Day, wanting to look at last year's Prime Day, and finding there is nothing there.
The actual windows
Treat these as planning figures rather than contractual guarantees. Amazon adjusts reporting behavior periodically, and the practical availability of very granular data can be shorter than the headline window. The safe assumption is that anything older than two months may not be there when you go looking.
What this actually costs you
Year-over-year is impossible natively. This is the big one. Ninety-five days does not reach back a year. It does not reach back two quarters. If you want to know whether your Prime Day performance improved against last year, the only way to answer that is if you exported it last year. There is no button.
Seasonal baselines have to be built, not queried. Categories with real seasonality, and that is most categories, need multi-year context to separate a trend from a season. A 95-day window shows you one season with no comparison point.
Granular data ages out fastest in practice. Hourly and day-part level detail is the most valuable data you have for scheduling decisions and the least likely to be sitting in anyone's spreadsheet. Campaign-level monthly totals can be roughly reconstructed from invoices and sales data. An hour-by-hour conversion curve cannot be reconstructed from anything.
What to export, in priority order
If you are starting an export routine from nothing, the order matters, because the first month you do it is the month you are most likely to do it incompletely.
1. Hourly and day-part data. Highest value, zero reconstructability. If you make any scheduling decisions at all, this is the data those decisions rest on. It is also the data you need 30 to 60 days of before you can trust an hourly pattern, which means you are always close to the edge of the retention window when you finally have enough of it.
This is the practical trap. You need roughly two months of hourly data to see a reliable pattern. Sponsored Brands and Display retain roughly two months. There is no margin. Miss one export cycle and you restart the clock.
2. Search term reports. Search term data drives negative keyword decisions and match type strategy. Historical search term data also tells you how demand language shifts season to season, which is genuinely hard to get any other way.
3. Placement reports. Needed to evaluate whether top-of-search premiums are earning their cost. Useful in aggregate over long windows.
4. Campaign and ad group daily performance. Important, but the most reconstructable of the four, so it goes last if you are triaging.
Setting up a routine that survives contact with a busy quarter
Monthly is the minimum viable cadence. It keeps you inside even the 60-day window with room for one missed cycle. Weekly is meaningfully better if you run event-heavy campaigns, because a single skipped month during Q4 costs you the only data anyone will want to look at next Q4.
Two practical notes. Store raw exports, not summaries, because you will want to ask a question in eighteen months that you have not thought of yet. And store them somewhere that is not one person's laptop.
Northlane Goods spent a full planning cycle rebuilding a seasonal baseline by hand from invoices and order data after discovering their hourly detail had aged out. It took most of two weeks and produced something noticeably worse than an export would have. The export takes about ten minutes a month.
The scheduling angle
The retention window has a specific consequence for dayparting that is worth naming. Good scheduling decisions need 30 to 60 days of hourly data, and they need to be revisited as seasonality shifts. If your only copy of that data lives inside Amazon's retention window, then every scheduling decision you make is built on evidence that is actively expiring.
Keeping your own hourly history means you can compare this October's conversion curve to last October's, rather than to whatever the last 60 days happened to contain. That comparison is where the actual insight lives, and it is only available to people who wrote it down.
Frequently asked questions
How long does Amazon keep Sponsored Products data? Sponsored Products reporting data is retained for roughly 95 days. Sponsored Brands and Sponsored Display data is retained for a shorter window, roughly 60 days. Once data ages past the retention window it is no longer available to pull from reporting, so anything you want for long-term analysis has to be exported before it expires.
Can I do year-over-year analysis in Amazon Ads? Not natively. With a retention window of roughly 95 days at most, last year's Prime Day or Black Friday data is long gone by the time you want to compare against it. Year-over-year analysis is only possible if you exported and stored the data yourself at the time.
What Amazon Ads data should I export first? Hourly and day-part level data first, because it is the most granular and the hardest to reconstruct. Then search term reports, then placement reports, then campaign-level daily performance. Campaign-level summaries are the easiest to approximate later; granular data is not.
How often should I export Amazon Ads reports? Monthly at minimum, which keeps you comfortably inside even the 60-day window. Weekly is better if you run seasonal or event-heavy campaigns, because it means a missed month never costs you a peak period.
Off Hours is dayparting and budget automation for Amazon Ads. Set your rules once in plain English, and they run every day across every account you manage. Start a free 14-day trial.