The AI Readiness Checklist: 12 Questions Before You Build

Why most AI projects fail before they begin

A Gartner study found that over 80% of AI projects never reach production. The common explanation is "the data wasn't ready" or "the model didn't perform well enough." Both are symptoms. The underlying cause is almost always the same: the organisation started with a solution before understanding the problem, the workflow, or its own readiness to operate AI.

This checklist is designed to surface readiness gaps before you spend budget. Answer honestly. Each question is a decision gate — not an obstacle.

Section 1: Workflow clarity

1. Can you describe the workflow in process terms — not AI terms?

Before any AI discussion, write down the workflow: what triggers it, what inputs arrive, what decisions are made, who is involved, and what the output looks like. If you cannot describe it without mentioning AI or models, the workflow is not defined yet.

2. Where does human judgment currently happen in this process?

Identify every point where a person makes a decision — scoring, approving, escalating, rejecting. These are your human-in-the-loop checkpoints. AI can assist or accelerate at these points, but replacing them without understanding them creates risk.

3. What does a "good output" look like — and can you measure it?

If you cannot define success in measurable terms (accuracy rate, time saved, error reduction, cost per decision), you have no way to evaluate whether AI is working. Define the success metric before the first line of code.

Section 2: Data readiness

4. Do you have historical examples of good and bad decisions in this workflow?

AI systems learn from examples — or they retrieve from structured knowledge. If you have no historical records of how decisions were made and what outcomes followed, you are building on a blank slate. That is not impossible, but it significantly increases cost and risk.

5. Where does the data live, and can you access it programmatically?

Data in PDFs, email threads, spreadsheets, or legacy systems is not the same as accessible data. Audit where your workflow data lives and whether it can be extracted, structured and fed into an AI system without manual effort.

6. Is your data consistent enough to be trusted?

Inconsistent data — different formats, missing fields, contradictory entries — produces inconsistent AI. Before asking "can AI process this?", ask "would a new human hire understand this data well enough to do the job?" If the answer is no, the data needs cleaning first.

Section 3: Governance and security

7. What data cannot leave your environment?

Many enterprise workflows contain PII, legal documents, financial records or commercially sensitive information. Know exactly what data your AI system will touch and whether it can be sent to a third-party model API, or whether it must stay on-premises or in a private cloud.

8. Who owns the AI output — and who is accountable for errors?

When an AI-assisted hiring decision is challenged, who is responsible? When a contract analysis misses a clause, who is accountable? These are not hypothetical questions. They need answers before deployment, not after an incident.

Section 4: Organisational readiness

9. Does a senior sponsor own the AI outcome — not just the project?

AI projects sponsored by IT succeed in deployment. They fail in adoption. You need a business-side owner who cares about the workflow outcome, not just the technology delivery. Without this, AI gets deployed and then quietly ignored.

10. Have the end users of this workflow been involved in the design?

Recruiters, lawyers, analysts, operations managers — the people who will use or be affected by AI in their daily workflow need to be part of the design process. Not as stakeholders who approve requirements documents, but as active participants who test, challenge and improve the system.

11. Do you have a plan for what happens when the AI is wrong?

AI systems make mistakes. The question is not whether yours will — it will. The question is: what happens when it does? Is there a reviewer queue? A fallback process? A correction mechanism that improves future performance? If the answer is "we'll deal with it when it happens," the system is not ready.

Section 5: ROI and timeline

12. Can you articulate the business case in three numbers?

Every serious AI investment should be justified with three numbers: the cost of the current state (time, money, error rate), the expected improvement after AI deployment, and the investment required to get there. If you cannot build this calculation, either the use case is not strong enough or the workflow is not understood well enough.

How to use your results

If you answered "yes" clearly to 10 or more of these questions, you have a strong foundation. Move to workflow design and use-case scoping.

If you answered "yes" to 7–9, you have identifiable gaps. Address them before committing to a build. A short readiness workshop can close most gaps in 2–3 weeks.

If you answered "yes" to fewer than 7, stop. You are not ready to build AI — you are ready to invest in readiness. That is not a failure; it is the honest starting point that saves six months of frustration.