RTO (recovery time objective) is how long a system can be down before the outage does unacceptable harm. RPO (recovery point objective) is how much data you can afford to lose, measured in time. RTO looks forward from the failure to the moment users can work again; RPO looks back from the failure to the last good copy of your data. Most firms write both numbers down, and discover in a real incident that their infrastructure delivers neither.

RTO and RPO at a glance

RTORPO
Question it answersHow long can this system be down?How much data can we lose?
MeasuredForwards, from the failure to the service being usableBackwards, from the failure to the last recoverable copy
Example“Back within 4 hours”“Lose no more than 15 minutes of data”
Mostly driven byHow much recovery infrastructure is ready, and how automated the recovery isHow often data is copied or replicated

The full definitions are in our glossary: recovery time objective and recovery point objective. The rest of this post is about what the two numbers mean for design and cost.

Why the two numbers decide your DR design

AWS describes four disaster recovery strategies in its Disaster Recovery of Workloads on AWS whitepaper. The shorter your RTO and RPO, the more has to be running before anything goes wrong, and the more it costs.

StrategyWhat it means (AWS’s wording, shortened)Typical RTOTypical RPO
Backup and restoreBack up into the recovery Region, then deploy infrastructure and restore data after a disaster. AWS calls it the least complex strategy, with a higher RTO and RPOHoursMinutes to hours, depending on how continuous the backups are
Pilot lightData replicated and always on; application servers not deployed until neededTens of minutes to hoursSeconds to minutes
Warm standbyA scaled-down but fully functional copy always runningMinutes to tens of minutesSeconds to minutes
Active/activeThe workload serves traffic from multiple RegionsNear zeroNear zero

The “typical” columns are our rules of thumb, not AWS guarantees; your own failover tests are the only reliable numbers. One AWS-stated detail is worth knowing: with automated or continuous backups and point-in-time recovery, AWS says backup and restore “can lower RPO to as low as 5 minutes in some cases”. AWS also advises choosing active/active only when business requirements demand it, as it is the most operationally complex option.

A worked example: 4-hour RTO, 15-minute RPO

(Illustrative.) A UK payments firm runs its transaction system in one AWS Region. The business sets an RTO of 4 hours and an RPO of 15 minutes. Which pattern fits?

Start with the RPO. A nightly snapshot could lose up to 24 hours of data, so it fails the 15-minute target outright. The firm needs continuous backups with point-in-time recovery, or a database replica in a second Region.

Then test the RTO. With backup and restore, a recovery has to fit into 4 hours. A realistic breakdown for this firm might be:

StepIllustrative time
Detect the failure and decide to fail over30 minutes
Rebuild networking and compute from infrastructure as code60 minutes
Restore the database90 to 150 minutes, depending on size
Validate, switch DNS, notify users30 minutes
Total3 hours 30 minutes to 4 hours 30 minutes

That sits right on the limit. Pilot light removes the slowest step, because the database replica is already running, and a typical failover test might take 45 to 60 minutes.

OptionMeets 15-minute RPO?Meets 4-hour RTO?Illustrative extra monthly cost*
Nightly backups onlyNoUnlikely1–2%
Continuous backups, restore on demandYesMarginal, depends on database size and automation2–5%
Pilot lightYesYes, with margin10–25%
Warm standbyYesYes, with more margin30–50%

*Planning assumptions as a share of the monthly production bill, not AWS prices. Replace them with figures from the AWS Pricing Calculator for your estate.

On a £10,000-a-month production bill, pilot light adds roughly £1,000 to £2,500 a month. If an hour of downtime costs this firm £15,000 in lost payments and staff time, that is an easy case. If the same system costs £500 an hour to lose, backup and restore may be the right answer. The decision comes from comparing the cost of the pattern with the cost of the outage, system by system.

The gap: three versions of the truth

Most firms hold three different views of recovery, and they rarely agree:

WhoWhat they believeWhat is usually true
The business“We’d be back within an hour”The number was never written down or agreed
IT“We have backups, so we’re covered”Backups exist, but nobody has timed a full restore
The infrastructureWhatever the architecture can actually deliverOften hours, and sometimes longer than anyone has tested

Restoring data is only part of recovery. Servers, configuration, integrations, access, validation and traffic routing all have to work too, which is why backups alone are not a DR plan.

For FCA-regulated firms, RTO is an impact tolerance in technical form

Under the FCA’s operational resilience rules, firms set an impact tolerance for each important business service: the maximum disruption it can take. Since 31 March 2025, firms must be able to remain within those tolerances, as set out in the FCA’s operational resilience policy statement, PS21/3. The RTO for the systems behind a service therefore has to sit comfortably inside its impact tolerance, with margin for the steps people forget, such as DNS changes and third-party dependencies. See FCA operational resilience for the full picture.

Tier your systems

One RTO for the whole estate is either too expensive or too weak. Group systems by how much an outage hurts:

TierExamplesTypical RTO / RPO targetUsual pattern
1: criticalPayments, client portal, tradingMinutes to a few hours / minutesWarm standby or active/active
2: importantInternal CRM, case managementSeveral hours / hoursBackup and restore or pilot light, with a tested runbook
3: tolerantArchives, reporting, test systemsDays / a dayBackup and restore

An RTO you have not tested is a guess

Test critical systems at least annually and after any significant infrastructure change, and time each step of the failover. The first test almost always finds something: a manual DNS change, a missing permission, a restore slower than assumed. Fixing those in a test is the point. For a side-by-side of the four patterns, see AWS Disaster Recovery Strategies.

Next step

Set an RTO and RPO per system, check them against your impact tolerances, and time a real recovery. A Yemberzal Disaster Recovery Assessment does this system by system, with drills that prove the numbers.