On 31 October Microsoft 365 Backup began supporting recovery windows of up to ten years for SharePoint and OneDrive. Getting a single file back still only works within the most recent fourteen days, and no option you can buy extends that. The 31 October entry in the product's what's-new log says so: "Granular file and folder restore is available only for recovery points within the most recent 14 days, regardless of the configured recovery window." And on a different page, the restore one, the restore point frequency table finishes the job: the file-and-folder cell beyond two weeks holds two words, "Not available".
The day-fifteen consequence is not in the announcement, it sits in the considerations section of the restore page, and it is the least-quoted sentence in this whole business: when you take a site back to a prior point, the site is not locked into a read-only state, so "users might not realize their current edits will be imminently rolled back and lost". Tuesday, eleven in the morning. The bids team has been working in that site all morning. Somebody asks for a document from a month ago and the only route left to give it to them wipes out whatever those people have done today, with no warning, no lock, nothing.
In the board meeting both things go by the same name. Somebody asks "how long do we keep things?", somebody answers "a year", and both leave reassured while thinking about different things: one is thinking about retention, the other about pulling one specific document from one specific date. They are two columns of the same table and they do not move together.
A fortnight ago we wrote about this same window when it only had four values and the point was the asymmetry between raising and lowering it: increasing applies immediately, decreasing is a destructive action with a thirty-day grace period. The list has stretched since. For SharePoint and OneDrive it is now three months, six months, or any whole number of years from 1 through 10. Exchange Online stays where it was: three months, six months, one year or two. And whatever you already have keeps its one-year default until somebody goes in and changes it. A correction of our own while we are here, since this is the place for it: in August, going over what this product covers and what it does not, we wrote that retention was one year across all three services and that we had found no Microsoft announcement extending it. On 31 October there was one.
The table worth reading twice
Microsoft publishes its restore point frequency — its RPO — in three rows. Translated, they say this:
| What you restore | Within 14 days | Beyond 14 days |
|---|---|---|
| Whole SharePoint site or OneDrive account | every 10 minutes | once a week, through the end of the window (up to 10 years) |
| Exchange Online | every 10 minutes | every 10 minutes |
| Individual files and folders in SharePoint and OneDrive | roughly daily | Not available |
The third row is the one that changes plans, and the sentence beside it shuts the door on reading it any other way: "Configuring a longer recovery window doesn't extend granular restore availability." There is a second one worth reading on signing day, because it is the one that catches out whoever has just moved to ten years: "Increasing the recovery window doesn't create older recovery points or recover points that have already expired." Extending today does not hand you ten years of history. It gives you permission to start accumulating it.
What "restoring" means on day 30
Past the fourteen days, the unit of recovery stops being the file and becomes the site. The option is called "Replace sites with backups" and the documentation describes it without decoration: "The entire original site is replaced by the backed-up version chosen based on the restore point. File and folder permissions and all metadata states are also reverted to the selected date and time." It takes the whole site back to that date, permissions included.
Which is why the opening matters: that rewind does not warn whoever is inside. The alternative is the wizard's other option, restoring to a new URL: the site reappears with an R and a number stuck on the end — …/sites/projectR0, the number incrementing to avoid name collisions up to a thousand times, at which point you have to delete old copies to free up names. That destination does come with a read-only lock, which is what you want. And here is the part that made us change an internal procedure: that lock can only be removed by "the Global Administrator", per the documentation, who is also the one who can download the documents. With the SharePoint Backup Admin role — the one granular restore requires — you cannot finish your own rescue.
So the actual procedure for "get me back the March quote" consists of standing up a full copy of the site at another address, asking a Global Administrator to release the lock and rescuing one file by hand. It gets done and it works. But saying "restore" in a meeting and doing this in the console set very different expectations about how long it takes and how many people you have to bother.
The arithmetic, with the numbers in
Ten years is 3,653 days. Fourteen of them allow a file-level recovery: 0.38 % of the window you just bought. Across the other 3,639 there will be, after those ten years, some 520 weekly points — today you only have whatever you have accumulated since you switched the policy on, because extending the window does not manufacture a past — and there your RPO is not ten minutes but one week: rewinding a site to five months ago can cost you, by design, up to seven days of that site's work. With the default one-year window the ratio is no better: fourteen days out of 365, 3.8 %.
On the bill, the published rate is "$0.15 per GiB per month for all data protected by Backup", with restores free. We already did the sum of which part of that base depends on the window, using Microsoft's own example, when full workload backup was announced: of the 3.5 GB billed for a 1 GB site, most was live size and only half a gigabyte depended on what was retained. Where you do need to look carefully is the other direction, because the documentation says it three times: when you remove a site or a mailbox from a policy, "existing backups aren't deleted and will be charged", and removed items keep the last recovery window that applied to them. Choosing ten years commits you to ten years of billing for the things you eventually take out, too.
In Exchange the arithmetic comes out the other way
On email the product does what people expect. The ten-minute points cover the whole window and not just the first two weeks, and item-level restore works with search by date range, sender, recipient, attachment, subject and content type. Recovering a message from eight months ago without touching the rest of the mailbox is a routine operation. With one caveat worth flagging because the source contradicts itself: the restore page says those ten minutes cover the configured window, and the feature table on the overview page says "10 minutes for prior 52 weeks", which is not the same thing if you picked two years. If one of those two numbers is going to end up in a recovery plan, make it the conservative one.
Exchange's fine print sits elsewhere, on what gets copied and where it goes back to. Each granular restore is "limited to restoring 1000 items at a time". Drafts "aren't backed up or restorable". Whatever the user moved to Deleted Items is not restored by the product — the user pulls it out of their own bin: only modified items, items deleted into Recoverable Items and purged items qualify. And one that shapes procedures during investigations and leaver processes: "Mailbox items can only be restored to the current mailbox. They cannot be restored to another mailbox." No room or group mailboxes either, and no hybrid Exchange: only mailboxes fully hosted in Exchange Online.
What this does to your RTO
For the scenario it is built for — mass encryption, mass deletion — the published numbers are good, and they are the honest argument for switching it on. A single SharePoint or OneDrive unit with an express restore point, about thirty minutes; a mailbox, about two hours. Beyond a thousand units, "up to 250 protection units per hour", and the what's-new log adds a cumulative throughput of "roughly 6,000 sites and accounts per day". Three thousand sites works out at twelve hours. It is worth saying where that twelve comes from before taking it to a board: it is a ceiling ("up to"), measured with sites averaging 12 GiB, in-place restores and express restore points, and the same page warns that large recoveries can take "a few hours to warm up". It is the discipline we already asked for when we pointed out that the vendor publishes three different performance figures for the same operation: the plan gets the footnote number, not the headline one. With all those caveats in place, it is still fast, and considerably faster than dragging the same terabytes back from a remote vault.
Where the argument falls down is in using the same product as if it were a searchable historical archive. Behind the same button there are two tools: one recovers the state of your organisation at a date, the other recovers one specific thing. If your RTO and RPO are written without separating those two cases, they are written for the one that will almost never happen to you.
Two drills a month per unit, says the vendor
Under the heading "Abusive restore actions aren't permitted", the restore page slips in a line that changes how a test plan gets written: "You should limit restores for testing purposes to no more than twice a month per protection unit. Restores for real recovery purposes aren't limited." Read it precisely: that is a recommendation from the vendor, not a counter the product enforces and cuts you off with. But it is the sentence somebody will meet the day they ask why you restored the same site forty times in March, and the quota it proposes is per site, per account and per mailbox.
What that means in practice is that the drill gets planned or it does not exist: you pick the units that genuinely represent your organisation, rotate them across months and keep the evidence. For the evidence you have two dated records: restore session history, which "is retained for 366 days", and the product activity logs, at "a fixed, nonconfigurable 365 days". That holds up in the auditor's folder, and it is what we show when somebody asks whether the backups get tested.
January brings the full workload policy. It does not touch the fourteen days.
On 18 November, the day before yesterday, Microsoft updated notice MC1387526 with new dates. General availability of full workload backup — one checkbox per workload, which we wrote about when it was due in September — moves to January 2027, with rollout completing end of February. It had been mid-September, completing in October. In the documentation it is still marked preview. Microsoft gives no exact day at either end, so the honest way to count it is roughly: a little over six weeks until it starts and a bit more than three months until it finishes.
What that policy fixes is real, although the message centre notice does not spell it out and you have to go to the policy documentation to see it: "Microsoft 365 Backup checks for newly created SharePoint sites, OneDrive accounts, and Exchange mailboxes every 24 hours." Custom policies keep precedence and you can exclude up to ten thousand units in one operation. With that switched on it stops depending on somebody remembering to add yesterday's new site to the policy.
In the meantime there is a gap worth noting, because it does not hit the three workloads equally. OneDrive and Exchange have dynamic rules — also in preview — based on distribution lists or security groups, and there membership changes are "automatically reevaluated daily": the mailbox of whoever joined on Monday is covered on Tuesday. SharePoint has no such thing. Its name and URL filters — also in preview — are a bulk-add aid, up to ten names at a time, and nowhere does the documentation say they re-evaluate on their own; the sentence that does declare static rules sits in the OneDrive block, not the SharePoint one. Plainly: unless you switch on the preview, a new SharePoint site enters no policy until a person puts it there by hand. And new SharePoint sites are exactly what a fast-moving team produces. The preview is there and you can enable it today; what you have to decide is whether switching on a preview feature inside the product that holds your backups is something you want to do with the calendar in hand, or something you would rather wait until January for.
Four questions before you touch the window
| Question | Why it decides |
|---|---|
| What are you actually asking for when you ask for "a year"? | If the real request is "get me a file from three months ago", the window does not answer it. The candidates are SharePoint version history and the second-stage recycle bin, both older than any of this; but each has its own expiry — the 93 days of the recycle bin run from deletion and span both stages, and version history depends on the version limit the admin set — so this is something you check, not something you promise. |
| Who holds the SharePoint Backup Admin role? | Without that role there is no granular restore, not even inside the fourteen days. It is a role people discover on the day they need it, which is the worst day to discover it. |
| Which libraries have IRM and which sites are locked or under strict hold? | Granular restore does not support IRM-enabled libraries or locked sites. And a site under a strict SEC 17a-4(f) hold fails any in-place restore: your options are a new URL or removing the hold. |
| When did you last restore something? | With the vendor recommending no more than two tests a month per unit, a decent drill needs a calendar and a chosen set of units. With no calendar, the honest answer to this question is "never". |
And what we would not do, which is the part usually missing: we would not move the window to ten years by default, however cheap it is. If the reason is compliance, that conversation is not about backup but about Purview retention, a different product with different rules. If the reason is recovery, the deciding number is not the 10: it is the fourteen days and the weekly point. Ten years of points that only yield whole sites are not a searchable archive; they are a right to rewind that nobody will exercise in 2034.
None of this is an argument for not switching it on. We do switch it on, and for the case that matters — the day you have to put a whole organisation back to the night before — it is fast, it is cheap and it does not depend on anyone dragging terabytes down a link. What we are saying is: buy it knowing which cell of that table is yours. There is a quick way to find out without opening anything: the last file somebody asked you to recover — what date was it from? If it is more than two weeks back, you live in the right-hand column of that table. And in the right-hand column there are no files, there are whole sites.
Sources (verified at origin on 19/11/2026): the extension of the window to ten years (31/10/2026), the sentence about granular restore and the fourteen days, the recommendation of two test restores a month per protection unit, the 366 days of restore session history and the fixed, nonconfigurable 365 days of activity logs come from "What's new in Microsoft 365 Backup" and "Restore data in Microsoft 365 Backup", which is also the source of the restore point frequency table, the "Not available", the two destination options, the warning that sites are not locked read-only, the 1,000-restore limit for new URLs, the 1,000 items per Exchange restore and the considerations on IRM, locked sites and SEC 17a-4(f). The window values, the scale limits, the dynamic rules reevaluated daily and how full workload backup works (24-hour checks, custom policy precedence, 10,000 exclusions per operation) — "Create, view, and edit backup policies". The $0.15 per GiB per month rate, free restores, the restore timings (30 minutes, 2 hours, 250 units per hour, the warm-up warning and the 12 GiB benchmark) and the Exchange row reading "10 minutes for prior 52 weeks" — "Overview of Microsoft 365 Backup". The cumulative throughput of "roughly 6,000 sites and accounts per day" comes from the 18/09/2026 FAQ in "What's new", not from the overview page. That the read-only lock on the new URL can only be removed by the Global Administrator, and that backups of removed items "aren't deleted and will be charged", come from "Restore data" and "Create, view, and edit backup policies" respectively. The 93 days of the SharePoint recycle bin (spanning both stages) — "Learn about retention for SharePoint and OneDrive". The date change for full workload backup to January 2027 is the 18/11/2026 update of message centre notice MC1387526. Translations of the quotes are ours; the English originals are behind the links. The arithmetic of the 3,653 days, the 0.38 %, the 520 weekly points, the twelve hours for three thousand sites are our own calculations on those figures. Photograph: ICRC public archives in Geneva, by RomanDeckert, CC BY-SA 4.0 via Wikimedia Commons, cropped. We are not a reseller of Microsoft licences or anybody else's: we charge for designing and operating, not for what you subscribe to.
Do you know the date of the last file you were asked for?
At everyWAN we set up Backup 365 by first spelling out what it covers and what it does not, and we fit it into the rest of the disaster recovery plan rather than leaving it as a switched-on checkbox. If you want to review your window, your policies and your drill calendar, write to us.
Talk to everyWAN