In April, backup vendor Veeam published a study of more than 900 senior technology and security leaders. Ninety percent said they were confident they could recover their data and operations inside the recovery targets written down in their plans. In September, an incident-response firm called Fenix24 released field data from more than 500 ransomware recoveries: across more than 800 client engagements, exactly four came anywhere close to the 24-to-48-hour recovery targets those organizations had documented in advance. Four. None of them reached full operating capacity for several weeks.

Two studies, three months apart, drawing the same gap from opposite ends. Ninety percent of leaders hold a belief. About one in three ransomware victims, by Veeam's numbers, holds an actual recovery: only 28% fully recovered all affected data, the average recovery covered 72% of it, and 44% got back less than three-quarters. The most common ending is a partial restore with a remainder that is gone for good.

The copy and the comeback

The gap has a simple anatomy. A backup is a copy: data moved somewhere else on a schedule, with a green checkmark when the move finishes. A restore is the reverse trip: putting a working business system back using those copies, with the right files, in an order that satisfies every dependency, on a clock the business can afford. The first runs every night, unattended. The second is a project, and most organizations have never run it.

Backup software does not prove the second thing. It only proves the first.

Five boring reasons restores fail

The job list was never audited. A backup reports success for the systems it was pointed at. Whether every system that matters is actually on that list is a separate question. Fenix24 reports that not one of its more than 800 clients arrived at a recovery with a complete map of their applications and dependencies, because nobody draws that map until they need it. The customer list added last year, the order database, the accounting files: all of it can sit outside the schedule while the nightly job reports green.

The data is older than the day. A restore returns the business to the moment of its last good copy. Everything done since then, the orders taken and the invoices sent, lives only in the systems under attack. Few businesses know how many hours of work sit between their snapshots, because that number only matters on the morning it is lost.

The backups live behind the door the attacker takes first. Fenix24 found that 94% of its clients ran backup infrastructure joined to the same login system the attacker compromises first. In plain terms: the accounts that can reach or delete the backups can be the very accounts the attacker already owns. A rescue copy within an attacker's reach is a hostage too, which is why one copy should live offline, or in storage even an administrator cannot alter.

Surviving copies do not carry the recovery. In 38% of Fenix24's engagements, the backups came through the attack largely intact and still could not restore the business: too old, incomplete, the wrong type, or slower than rebuilding the machine outright. The files exist. The comeback does not.

Nobody alive has practiced it. Recovery is a procedure, usually written years ago by someone who has since moved on. Under pressure, the people running it are reading it for the first time. Nothing in an automated backup tests whether the sequence works, whether the dependencies hold, or whether the team knows its parts.

What rehearsal looks like

The test is specific and small: pick one system the business would genuinely miss, restore it from the actual copies into a separate environment, and run a clock. Then answer three things with numbers: whether the data came back usable, whether the program started and served, and how long the whole trip took. Start with the login system, the one every other system checks first; when it is down, every other restore waits behind it. Rotate to a different system each quarter, and test again after any major change. A new server, a replaced application, a platform migration: each one can quietly break the restore path that worked last time.

An owner who wants the process managed right should treat the first honest result as the finding, not a failure. Most first tests are humbling, and a humbling number on no notice costs the whole afternoon; the same number discovered on the worst morning costs the week.

The questions worth asking

At the next review, ask whoever runs the backups, in-house or contracted, for the date and duration of the last full restore of a system the business would genuinely miss. Ask what is not in the backup scope. Ask which copy lives beyond the reach of administrator credentials. Real answers arrive as dates, numbers, and named systems. Hemming arrives as a dashboard of green checks with no rehearsal behind it. Board directors, insurers, and regulators increasingly ask for proof of recoverability, and a dated, timed restore test is the shortest answer there is.

The Bottom Line

Copies are a fact. Recovery is a capability, and the 2026 field data says the capability is rare: four recoveries out of more than 800 engagements came anywhere near their stated targets, and 99.2% of clients arrived with no written plan for the login system every other restore depends on. A backup proves the business has copies. Only a rehearsal proves the business can come back. If the last dated restore test anyone can name is years old, or does not exist, that finding is worth one quiet afternoon long before an attacker decides the schedule.

Story reporting: Veeam Data Trust and Resilience Report 2026 press release, April 14, 2026; Fenix24 The State of Recoverability 2026 via PR Newswire, September 15, 2026.