Operational readiness is not a single sign-off before go-live; it is a state built and rechecked throughout a process's life. Gardner (2001) describes an operational readiness assessment as testing whether the wider operating environment, not just the system itself, can support and sustain a change: the people, the procedures, the support arrangements and the contingency plans around it. For an AI-enabled or automated process, this matters more than for a traditional system. Automation can run unattended and at volume, so a gap in documentation, communication, reporting, backup or downtime planning is more likely to be exposed at scale, and less likely to be noticed quickly, than the same gap in a manual process.
Documentation review checks that the paperwork behind a process matches what actually happens, not what was true at launch. This includes the process map, the data sources and their owners, the model or rule logic, known limitations, and who is accountable for each part of the workflow. ISO/IEC 42001, the international standard for AI management systems, treats documented information as a core control, expecting organisations to keep records of decisions, roles and monitoring across an AI system's whole lifecycle, not only at implementation (ISO, 2023). A readiness audit should test this directly: pick a recent change and check whether the documentation was updated to match it, and whether a new team member could pick up the process safely from the written record alone.
Update communication is about who is told, and when, before a change goes live. Automated workflows often touch several teams, so a change that looks small to the people making it can still surprise everyone downstream. Good practice sets out in advance who needs to know about a planned change, a scheduled downtime window or a live incident, and through which channel, rather than deciding this in the moment (Atlassian, n.d.). Where an automated process supports a decision about a person, the Cabinet Office's AI Playbook for UK government adds that people affected by, or working alongside, the system should understand when it is involved and know how to raise a concern (Cabinet Office, 2025). An audit checks that this communication plan exists, names an owner, and was actually followed the last time something changed.
Error reporting processes determine whether a fault is caught early or discovered late, often by a customer. Reason (1997) argues that organisations only learn from failure when people feel safe reporting it: a blame-driven culture pushes errors and near misses underground, while a just, blameless approach treats an honest report as useful information. Google's engineering teams apply this directly to automated systems through the postmortem, a structured, blameless review that records what happened, why, and what will change, shared widely rather than filed away (Lunney and Lueder, 2017). For AI-enabled processes specifically, this reporting needs to catch more than crashes. NIST's AI Risk Management Framework highlights slower, quieter failures such as model drift and degrading performance, which need ongoing monitoring rather than a one-off test to surface at all (NIST, 2023). An audit should check that a fault, however it is spotted, has a clear route to being logged, reviewed and fixed.
Backup processes protect the data an automated process depends on, and ensure the ability to prove it can be restored. The National Cyber Security Centre advises organisations not to rely solely on a cloud provider's own recovery features, and to keep an independent, tested copy of essential data in a separate location or service (NCSC, 2024). ISO 22301, the international standard for business continuity, goes further, asking organisations to identify which data and systems are critical, agree how quickly each must be recovered, and rehearse that recovery rather than assume it will work when needed (ISO, 2019). An audit that only checks a backup exists is incomplete: the real test is whether someone has recently restored from it and confirmed the result is usable.
Downtime procedures set out what happens when an automated process stops, whether through a planned change or an unplanned failure. GOV.UK's service standard asks teams to deploy changes without significant downtime where possible, and to have a proportionate plan, scaled to the impact on users, for when something goes wrong regardless (GDS, 2026). ISO 22301 frames the same expectation as a documented response and recovery plan, tested before it is needed, that names who can pause or roll back a process and what the safe fallback looks like in the meantime (ISO, 2019). Brought together, these five areas turn an assumption of readiness into something a leader can actually evidence.
Action Point
Pick one AI-enabled or automated process your team relies on. Check whether its documentation was updated after the last change, whether people were told about that change in advance, how a fault in it gets reported, when its backup was last successfully restored, and whether there is a tested plan for what happens if it stops working. Where the answer is unclear, treat that as a readiness gap and assign an owner and a date to close it.