AI readiness
checklist.

AI applied to a process nobody has described will encode the confusion rather than remove it.

• The process has been written down as it is actually performed, not as documented
• The people who do the work have reviewed and corrected that description
• The steps that exist only out of habit have been identified
• The decision rules a person follows can be stated explicitly
• You know how often the process runs and how long one instance takes
• The exceptions have been listed, with an agreed route for handling them

A system can only be as reliable as the information it is allowed to read.

• You know where the information the process depends on is stored
• A single authoritative version of each key document has been agreed
• Superseded and duplicate versions have been retired or archived
• Each authoritative source has a named owner
• You know how current the information is and when it was last reviewed
• Information that must never leave your environment has been identified

Integration effort is the most commonly underestimated cost in adoption projects.

• The tools involved in the process have been listed
• You know which of those tools can be connected and which cannot
• Someone has confirmed you hold the necessary admin access and licences
• Places where information is manually re-entered between systems are known
• Any system due for replacement in the next year has been flagged

Solutions without an owner degrade quietly until people stop trusting them.

• A named person will own the workflow after launch
• That person has capacity allocated, not just responsibility assigned
• There is a route for staff to report problems or request changes
• Someone can judge whether the output quality is acceptable
• A fallback exists for when the system is unavailable

Most of the rules you need already exist somewhere in the business.

• Existing obligations have been checked: client terms, regulation, retention rules
• The categories of data that must not enter an external tool are named explicitly
• The points requiring mandatory human review have been agreed
• Accountability for AI-assisted work sent to clients is clear
• An approved-tools list exists, with a route to request additions
• A review date has been set for whatever policy you write

This section is skipped most often and is the most common cause of failure.

• The people affected have been told why the change is happening, not just what is changing
• Their concerns have been asked for directly and recorded
• Time for learning has been allocated inside working hours
• Training will use the team's real tasks rather than a generic demonstration
• It is safe to report a mistake made while using a new tool
• There is agreement on what success looks like and when it will be reviewed