AI Readiness: 9 Checks to Run Before You Hire Anyone

The nine dimensions

Readiness frameworks vary in wording but converge on the same nine areas. Score each one honestly from one to five. The absolute score matters less than which ones are lowest.

• Business strategy — can you name the outcome you want, in numbers, without mentioning AI?
• Data quality — is the data that would feed this complete, current, and consistent?
• Infrastructure — do your systems have APIs, or is everything locked in a screen?
• Talent — is there anyone internally who can own this after handover?
• Governance — who approves an automated decision, and who is accountable when it is wrong?
• Culture — have previous process changes stuck, or been quietly abandoned?
• Financial planning — is there budget for the year after the build, not just the build?
• Process integration — does the output land where people already work?
• Monitoring — would you notice within a day if it started behaving badly?

Why it is almost always data quality

At low maturity levels the blocker is nearly always the same two: data quality and governance. Not model choice, not tooling, not budget. This is consistent enough across assessments that you should assume it applies to you until you have checked.

The reason is structural. A demo runs on a curated sample. Production runs on whatever is actually in your CRM, including the duplicates, the records where someone typed the phone number into the name field, and the fields three different people fill in three different ways. Automation applied to inconsistent data does not fail loudly — it produces confident, wrong output at scale, which is worse.

Fix the data foundation before buying any tooling. This is not a prerequisite people enjoy hearing, and it is the one that most reliably determines the outcome.

If the answers are mixed and you would rather work through them with someone, that is the first half of an AI consultation — the assessment happens before any build is proposed.

The access test

Here is a single question that predicts readiness better than most formal assessments: can someone non-technical in your business get a number out of your systems without asking IT or waiting for a report?

If yes, your data is accessible, someone understands where it lives, and the plumbing works. If no, then any automation project will spend its first phase building that access anyway — which is fine, as long as you have priced it and are not surprised when the first month produces no visible AI.

What good enough data actually means

Perfect data is not the bar and waiting for it is its own failure mode. The practical test is narrower than people assume, because you only need the data for the specific process you are automating to be sound — not the whole business.

• The fields the process depends on are populated in the large majority of records.
• The same thing is recorded the same way — one format for dates, one for statuses.
• There is a reliable way to tell which record is the current one.
• Someone can say where each field comes from and who maintains it.
• You can export it. If you cannot get it out, you cannot build on it.

Map where data moves before anything else

Draw the actual path: which system captures it, what touches it next, where it gets re-keyed by a person, and where it ends up. Almost every business that does this finds at least one manual re-entry step nobody had documented, and those steps are usually both the highest-value automation candidates and the source of the inconsistency in the first place.

This exercise also surfaces how long decisions currently take and where they wait. That measurement becomes your baseline, and without a baseline you will not be able to prove the return later.

Three findings that mean wait

Readiness assessment is only useful if it can return a negative. These three should stop you from hiring this quarter.

• Nobody owns the process. Automating a process with no owner makes the ambiguity faster, not smaller.
• The process changes materially every month. Automation locks in rules; if the rules are still moving, wait.
• The real problem is a decision nobody has made. AI is not a substitute for deciding how the business should run.

If you score well, what changes

A business that passes these checks gets a materially cheaper engagement, because the discovery phase is short and the first build can start almost immediately. It also gets a better roadmap, since the consultant is working from real numbers rather than estimates.

The other thing worth knowing: this checklist is about you. Vetting the consultant is a separate exercise with a different set of questions, and both are worth doing.