System Design Interviews in Australia: CPS 230 and Numbers

A timed system design method for Australia, turning APRA CPS 230 tolerance levels into recovery targets, with Sydney to Melbourne replication sized.

By · · 7 min read

Illustration for the System Design Interview Framework guide

Australia edition. Written for Australian interviews at banks, insurers, super funds and the firms that serve them: APRA CPS 230 critical operations and tolerance levels, the 24 and 72-hour notices, the OAIC Notifiable Data Breaches scheme and the two Australian cloud regions. US edition

An Australian system design round uses the familiar four steps: scope and estimate, sketch the main path, go deep, then scale and trade-offs. At a bank, insurer or super fund, and at the vendors that build for them, the requirements step has a ready-made list waiting: APRA's prudential standard CPS 230 asks for three tolerance levels for every critical operation, and they map almost one to one onto recovery time, recovery point and degraded service. This edition shows how to use them, then sizes a payments design across Sydney and Melbourne. Practise with the prompts in the System Design Interview Framework (Australia).

The method, timed

StepMinutesOutput
Requirements and estimates8 to 10Features, non-goals, throughput, storage, the three tolerance levels
High-level design10 to 15Main path, APIs, data model
Detailed design15 to 20One or two components: schema, partitioning, idempotency, failure handling
Resilience and scale8 to 10Bottlenecks, degraded mode, failover, what you would monitor

CPS 230 in one table

APRA's CPS 230 Operational Risk Management (the amended version APRA released on 30 April 2026, which its operational risk page says commences on 1 July 2026; the original commenced on 1 July 2025) requires, for each critical operation, tolerance levels for:

CPS 230 tolerance (para 37)In your design
(a) the maximum period of time the entity would tolerate a disruptionRecovery time objective
(b) the maximum extent of data loss the entity would acceptRecovery point objective
(c) minimum service levels while operating under alternative arrangementsA degraded mode you must describe

Paragraph 35 lists operations an entity must treat as critical unless it can justify otherwise: for an ADI, payments, deposit-taking and management, custody, settlements and clearing; for an insurer, claims processing; for an RSE licensee (a super fund trustee), investment management and fund administration; and for all of them, customer enquiries and the systems and infrastructure needed to support critical operations. So a prompt about any of those is a prompt about tolerances.

The entity sets the tolerances; APRA can require changes. In the interview, ask for them or propose them.

Worked example: payments across two regions

Prompt. Design the payments service for an Australian bank's app, with disaster recovery.

Requirements. Accept a payment, debit the account, send it to the payment network, report status; no payment sent twice. Tolerance levels (interviewer's premises): maximum disruption 2 hours, maximum data loss 5 minutes, minimum service of balance enquiries and queued payments during a disruption.

Regions. The AWS regions list shows two Australian regions, Asia Pacific (Sydney) ap-southeast-2 and Asia Pacific (Melbourne) ap-southeast-4, each with three availability zones; Melbourne is opt-in.

Estimates (premises to agree):

peak throughput:   800 payments a second
data loss window:  5 minutes = 300 s
inside tolerance:  800 x 300 = 240,000 payments
fibre path:        assume ~1,000 km Sydney to Melbourne, ~5 µs per km
round trip:        ~10 ms before any processing

What the numbers decide.

  • Synchronous or asynchronous between cities? With about 10 ms per round trip, a strictly sequential writer that waits for Melbourne on every commit manages at most about 100 commits a second. At 800 a second you either batch and parallelise commits, or keep synchronous replication across Sydney's three zones and copy to Melbourne asynchronously.
  • The data loss tolerance sets the lag budget. Asynchronous replication is acceptable only if lag stays well under 5 minutes, with alerts long before that, and a way to reconcile anything inside the window from the payment network's records. If the interviewer sets data loss to zero, that is the argument for synchronous cross-region commits, and their cost.
  • The minimum service level becomes a feature. During failover, balances are served from a read replica and new payments are accepted into a durable queue with idempotency keys, then submitted when the primary path is back.

Notifications that start a clock

CPS 230 sets two APRA notification deadlines that your monitoring has to support:

  • 24 hours: notify APRA as soon as possible and not later than 24 hours after a disruption to a critical operation outside tolerance (para 41);
  • 72 hours: notify APRA as soon as possible and not later than 72 hours after becoming aware of an operational risk incident likely to have a material financial impact or a material impact on the ability to maintain critical operations (para 32).

For the design, that means monitoring each critical operation against its own tolerances (time down, data at risk, service level), not just CPU and error rates, so the people who must notify know which line has been crossed.

Personal information: the Notifiable Data Breaches scheme

The OAIC's guidance on the Notifiable Data Breaches scheme says a breach is eligible if it is likely to result in serious harm to any of the individuals involved; the entity must notify those individuals and the Commissioner. Where a breach is suspected, entities must take all reasonable steps to complete an assessment within 30 days, and the OAIC asks that 30 days be treated as a maximum.

Turn that into a requirement: audit logs queryable by person and date, and a data map that says which stores hold which personal information, so the assessment takes days, not weeks.

Service providers are part of the design

CPS 230 extends to material service providers, and APRA's page refers to the material service provider register entities submit. In an interview, draw the third parties: payment networks, identity verification, card processors, the cloud provider. For each, say what happens when it fails and how the critical operation stays within tolerance.

Common mistakes in Australian rounds

  • Quoting "four nines" instead of a maximum disruption time and a data loss figure.
  • Assuming synchronous replication between Sydney and Melbourne is free.
  • Leaving out the degraded mode; CPS 230 asks for minimum service levels, not just recovery.
  • Treating the OAIC 30-day assessment window as a target rather than a ceiling.

Tools for this job

Frequently asked questions

What are the three CPS 230 tolerance levels?

For each critical operation: the maximum period of disruption the entity would tolerate, the maximum extent of data loss it would accept, and the minimum service levels it would maintain under alternative arrangements. In design terms, a recovery time, a recovery point and a degraded mode.

How fast must APRA be told about an outage?

Within 24 hours of a disruption to a critical operation outside tolerance, and within 72 hours of becoming aware of an operational risk incident likely to have a material impact, in each case as soon as possible.

Should I replicate synchronously between Sydney and Melbourne?

Usually not for every write. Assuming about 10 ms per round trip, each synchronous commit pays that latency. A common answer is synchronous across zones in one region and asynchronous to the other, with the data loss tolerance setting the acceptable lag.

What does the Notifiable Data Breaches scheme need from a design?

A fast way to work out whose data was involved. Suspected breaches must be assessed within 30 days at most, and eligible breaches notified to individuals and the OAIC.

Sources and further reading

Open the System Design Interview Framework

More Interview Prep

Every tool in the Interview Prep section, free and in your browser.