Your backup job finished successfully. That doesn't mean the data comes back.
2026 industry data shows roughly half of all restore attempts fail and most disaster-recovery tests don't pass the first time - not because the backup didn't run, but because nobody tried using it until the day they had no choice.
The job is green. That's the only thing it proves.
A 2026 industry survey found that 65% of organizations fail their disaster-recovery test on the first attempt, and only 12% hit their target recovery time when they do test. Separately, close to half of restore attempts fail to bring back usable data, and 60% of backups turn out incomplete the moment someone actually opens them. None of those backups were reported as failed. The software wrote bytes somewhere on schedule and marked the job successful - it has no way of knowing whether those bytes still mean anything six months later.
Backing up and restoring are not the same skill
Backing up is passive: it runs quietly every night whether or not anyone is watching. Restoring is active, and almost nobody rehearses it. In the meantime the schema changes, an encryption key rotates, a service account loses the one permission the restore script assumed it still had, a bucket gets renamed during a tidy-up. Each change is harmless on its own and invisible in a backup log. The only moment any of it becomes visible is the one where someone is already trying to get the data back.
Ransomware turned this into a targeted attack
Attackers who get into a network now spend their quiet weeks before encrypting anything looking for the backup credentials and snapshot history first, specifically so the safety net is already gone by the time anyone reaches for it. Manufacturers alone report losing up to $1.9 million for every day a ransomware incident keeps them down - a number that assumes the restore actually works. A backup nobody has tried to use isn't a safety net. It's an unverified claim that one exists.
The ACS approach
We don't accept a green checkmark as proof a system is backed up. Before we tell a client their data is protected, we actually restore it somewhere separate and check that what comes out matches what went in - on a schedule, not once at launch and never again. It costs an afternoon every so often. The alternative costs finding out on the one day you can't afford to be wrong.