Here is what we do and recommend. Prevention is far less costly than repairs, so you must spend time to figure out what to protect and how to protect it.
The things worth protecting form a hierarchy, from the largest system down to the smallest subsystem, and each level has its own backup tool, method and timing: a whole machine, on premises or in the cloud, the data it holds (local data, server data, and cloud data: documents and databases), the hard disks and virtual machine disk images, and the pieces easiest to forget, the system states (programs and settings) and the boot records. Match a backup to each level and you restore just the broken piece, fast and with minimal disruption, instead of rebuilding a whole server to recover one file.
A cloud account is itself one of these systems, not merely a folder of files, and part of it is already backed up on your behalf. Microsoft OneDrive and Google Drive quietly keep a deep version history of your documents, and Google Drive even lets you mark a version to keep it forever.1 Knowing it is there, and how far back it reaches, is the difference between a five-second restore and a lost afternoon.
Those systems are useless if you cannot get back into them. As passwords change through time, you may not be able to access your resources after you restore them, which is why credentials once had to be backed up as they drifted. Single sign-on has largely dissolved that chore: you route authentication through one identity provider instead of scattering dozens of changing passwords. But the task only moves. That provider is now the one door you cannot afford to have locked, and it needs its own recovery path: Microsoft tells its own customers to keep at least two cloud-only emergency accounts outside the normal sign-on.2
Newest of these systems is the AI you now work with, and it is the easiest to lose, because it forgets. Its working memory is finite, and once a long project fills it the assistant summarizes the older material and drops the detail: Anthropic, the maker of Claude, calls that summarizing compaction, and names the accuracy that decays as the window fills up: context rot.3 The assistant that knew your project last month has been partly written over. So commit what it produces to version control, the same Git that records every change and who made it. A project kept in Git can be retraced and restored long after the model has forgotten the work.
Whatever the system, its backups then obey one discipline. Because the software and data on PCs and servers have become huge and unwieldy, some of the backups must be held within the machines themselves so that recoveries can be performed quickly, and the same backups and older versions must be held remotely for redundancy. On premises data should be backed up in another location or in the cloud, and cloud data on premises or in another cloud. The industry shorthands this as 3-2-1: three copies, two kinds of storage, one off site.
Backups should be automated and incremental, so that anything can be recovered predictably in the state in which it was an hour ago, yesterday, in any of the past twelve months, or years back. With multi-terabyte drives, this is inexpensive and relatively easy. The one step most often skipped: a backup you have never restored is a hope, not a backup. The proof is a restore you have actually run, before the day you need it.
Mapping every system that matters, matching a backup to each, and proving the restore is unglamorous work, done before anything breaks rather than after. We build it into every environment we run, so the worst day at your firm is one you have already rehearsed.
The newest asset worth backing up is the memory of the AI now working beside you, and everyone forgets it first.
Concierge CIO Partners is a unified Guild of senior technologists providing dedicated, long-term fiduciary IT leadership to midmarket service firms. It offers a strategic alternative to fragile internal IT silos and to Managed Service Providers who advise you on what to buy and are paid on what you buy. With transparent unit pricing and an automated service catalog, the Guild eliminates administrative bloat and ensures every IT dollar spent and decision made directly drives your financial performance.
1 OneDrive and SharePoint, under the recommended automatic setting, keep every version for the first 30 days (up to a 500-version cap), then thin older versions to hourly copies from 30 to 60 days, daily from 60 to 180 days, and weekly beyond: Microsoft Learn. Google Drive keeps the full history of native Docs, Sheets, and Slides, and about 30 days or 100 versions of uploaded files unless one is marked "Keep forever": Google Drive API. 2 Microsoft advises keeping at least two cloud-only emergency access ("break glass") accounts, excluded from the usual sign-on and multifactor policies, in case a federation or identity-provider outage locks out normal administrators: Microsoft Learn. 3 Current top models carry a context window of up to about 1 million tokens (many far less). As a conversation approaches that limit, Anthropic's "compaction" automatically summarizes older content and drops it; the platform also documents the related decay it calls "context rot": Anthropic, Compaction; Context windows.