Twelve Questions That Decide If Your Company Can Deploy
Most AI projects are lost before the first meeting with a vendor.
Gartner’s finding is blunt. Through 2026, organizations will abandon 60% of AI projects unsupported by AI-ready data, and in the survey behind that number, 63% of data management leaders said they either lacked the right practices for AI or could not say if they had them. Two thirds of the market is uncertain about the foundation and buying on top of it anyway.
An AI readiness assessment is cheap. Finding out in month seven is not.
Readiness Is Four Things, Not One
Vendors score readiness on data because data is what they sell against. Deployment fails across four domains, and the last two kill more projects than the first two.
| Domain | The question it answers | The failure it prevents |
|---|---|---|
| Data | Can the system reach the records | An assistant answering from a stale corpus |
| Process | Is the work defined enough to hand over | Automating a process nobody agrees on |
| Authority | Who decides, and what may the system decide alone | An agent with credentials and no mandate |
| Operations | Who owns it after launch | A working system that decays into a liability |
A company can hold clean data and still fail on the other three. The pilot trap usually closes on process, not on technology.
The Twelve Questions
Answer each with yes or no. No maybes, no “mostly”.
Data
- Can you name the system of record for the data this project needs
- Is that data readable by a machine without a person reformatting it
- Does someone own its accuracy, by name, today
Process
- Can you write the target workflow in ten steps or fewer
- Do two people in the business describe that workflow the same way
- Do you have a number that says how well it runs now
Authority
- Is there a written mandate for what the system handles and what it hands back
- Has someone decided what it may do without a person in the loop
- Is there a named escalation path when it is wrong
Operations
- Is there a budget for year two, not only for the build
- Is there an owner after launch, named in the plan
- Is there an evaluation that runs on a schedule, using your data
Twelve questions. Ten minutes. Most executive teams cannot clear eight.
Readiness is not a maturity score. It is a list of people who have agreed to be responsible.
How to Read Your Score
The scoring is deliberately harsh, because the cost of being wrong is measured in quarters.
| Yes answers | What it means | What to do next |
|---|---|---|
| 10 to 12 | Deployable now | Scope one system, attach a number, build |
| 7 to 9 | One domain is weak | Fix the weak domain first, then build |
| 4 to 6 | Foundation work required | Buy nothing yet, structure the corpus and the workflow |
| 0 to 3 | A vendor will sell you a pilot | Decline it |
Every no is roughly a week of institutional work before deployment, and that work is a fraction of what a failed rollout costs. MIT’s research on enterprise pilots found that 95% produced no measurable effect on profit and loss. That statistic is a readiness statistic wearing a technology label.
Fix the Cheap Things First
Three of the twelve are usually fixable inside a month, and each one moves several others.
- Name owners. Data accuracy, the system after launch, the escalation path. Naming costs nothing and closes three questions.
- Write the workflow down. Ten steps, plain text. Disagreements surface immediately, which is the point. Structured text is the format the system will consume anyway.
- Get a baseline number. How long the work takes now, how often it is wrong, what it costs per unit. No baseline, no way to prove the system worked.
Do those three and a company at six moves to nine. That is the difference between a deployment and an expensive experiment.
Score Yourself Before a Vendor Does It For You
The market is about to sell readiness assessments as a product. Most of them will grade your data and stay silent on authority and operations, because the first is a procurement problem and the other two are management problems.
Run the twelve yourself. Take the answers into the vendor meeting. A builder worth hiring will ask harder questions than these before quoting anything, and the ones who skip straight to a demo have told you what they are selling.
Software is only the surface. Infrastructure is the rest.
Build there.
