AI readiness assessment: what to check before you automate anything
Most automation projects fail on the inputs, not the tools. This is the framework we use on every audit: six dimensions, scored honestly, ranked weakest-first.
The six dimensions
Readiness is not one number. A team can be fluent with AI tools and still fail because nobody wrote the process down.
Data hygiene
Automation inherits the quality of whatever it reads. Duplicate records and half-filled fields multiply, they don't self-correct.
Process clarity
If a process only lives in someone's head, it can't be handed to a machine, or to a new hire.
Tool stack
Connected tools make automation a config job. Disconnected ones make it a rebuild.
Team fluency
Adoption fails on habit, not on technology. Someone has to own the change.
Approvals and risk
Knowing which decisions must keep a human signature is what separates safe automation from an incident.
Repetitive volume
Automation pays back on frequency. Low-volume work is usually cheaper to keep manual.
How to run the assessment
Score the six dimensions
Rate each dimension honestly on the evidence you have, not the version of the business you intend to build next quarter. Optimism here is the single most common reason automation projects stall.
Fix the gates before the glamour
Data hygiene and process documentation gate everything downstream. Automating on top of duplicated records or an undocumented process just produces wrong answers faster and at scale.
Pick one workflow by volume, not excitement
Rank candidates by how many times a week they run multiplied by the minutes each run costs. The winner is rarely the most interesting one; it is usually the dull, high-frequency task nobody defends.
Decide the approval gates up front
Write down which steps a human must sign: anything that emails a customer, moves money, or posts publicly. Decide this before building, not after the first embarrassing message goes out.
Define the rollback
Every workflow needs an answer to 'what do we do if this breaks on a Friday night?' If the answer is silence, you have designed an outage, not an automation.
Reading the score
Ready to scale
You are past the starting line. The risk now is automating the wrong things.
First workflow live in 2 weeks. Full orchestration in 6–8 weeks.
Ready to automate
You can start building now, with one or two gaps to close alongside.
First workflow live in 2–3 weeks. Measurable time back by week 5.
Foundations first
Automating today would lock in the mess you already have.
Foundations in 2–3 weeks, then first workflow 2 weeks after.
Not yet
Honest answer: automation is not your bottleneck right now.
Revisit once volume or documentation improves.
Four ways assessments go wrong
Scoring tools instead of outcomes
Owning a team ChatGPT plan is not fluency. What matters is whether people use it inside a defined process with a known output standard.
Starting with the most visible workflow
Customer-facing work carries the highest blast radius and the least tolerance for iteration. Prove the pattern internally first.
Treating documentation as paperwork
A written process is the specification. Without it, you are asking someone to automate a guess about how the work is done.
No owner after launch
Unowned automation decays. Apps change fields, workflows drift from reality, and within six months people quietly go back to doing it manually.