An AI readiness checklist tells you whether your organization is ready to start an AI project, and which gaps to close first. The 20 questions below cover the five areas where AI projects most often stall: data, process, people, security and budget. Score each one, add up the total and use the table at the end to decide your next step.
Use it before you talk to vendors, when you set a budget or when a pilot has stalled and you need to find out why.
How to Use This Checklist
Answer each question for one specific use case, not for "AI in general." Readiness for an invoice-processing agent and readiness for a customer-facing voice assistant are very different.
Score every question:
- 2 points: Yes, clearly, and we can show it.
- 1 point: Partly, or we think so but haven't checked.
- 0 points: No, or we don't know.
The maximum score is 40. Get the people who know the process, the data and the systems in the same room; the answers are rarely in one person's head.
Data Readiness
AI is only as good as the data it can reach. These questions test whether the data exists, is accessible and can legally be used.
- Do we know which data the use case needs, and where it lives? Name the systems, tables, folders or inboxes. "It's somewhere in the ERP" scores 1 at best.
- Can the data be accessed programmatically? APIs, database access or exports that can run on a schedule, not someone downloading spreadsheets by hand.
- Is the data accurate and complete enough for this use case? Check a real sample. Missing fields, duplicates and inconsistent formats are the most common blockers.
- Do we have examples of correct outcomes? Past tickets with good resolutions, approved invoices, labeled documents. These become the test set (evals) that proves the AI works.
If you score low here, start with data work. Our data and analytics services team builds the pipelines that make data usable, and dataset collection can fill gaps in labeled examples.
Process Readiness
AI improves a process; it can't fix one nobody has defined. These questions test whether the work is clear enough to automate or assist.
- Is the process documented, or can someone explain it step by step? Including the exceptions and who handles them.
- Is the task frequent enough to matter? A task done ten times a month rarely justifies a build; one done hundreds of times a day often does.
- Can we measure success? Time per case, error rate, response time, cost per transaction or conversion. Without a baseline, you can't prove value.
- Are the decision rules clear? When should the AI act, when should it ask and when must a human decide?
People Readiness
Many AI projects stall on adoption, not technology. These questions test ownership and change.
- Is there a named business owner? One person accountable for the outcome, with authority to make process decisions.
- Will the people who do the work today be involved? They know the edge cases and they decide whether the tool gets used.
- Do we have, or can we access, the technical skills? Someone who can integrate systems, evaluate models and run the solution after launch, in-house or through a partner.
- Is leadership willing to start small and learn? A scoped pilot with honest measurement beats a large program that has to look successful.
Security and Compliance Readiness
AI systems touch sensitive data and, increasingly, take actions. These questions test whether you can do that safely.
- Do we know which data is personal or confidential? And which rules apply: GDPR, HIPAA, India's DPDP Act, sector regulations or client contracts.
- Do we have a policy on which AI tools and providers may be used? Including where data is processed and whether it may be used for model training.
- Can we control what the AI is allowed to access and do? Role-based permissions, read-only access where possible and approval for high-impact actions.
- Can we log and audit what the AI does? Inputs, outputs, tool calls and approvals, kept long enough to investigate an issue.
For AI-specific risks such as prompt injection and data leakage, see our application security services.
Budget and Business Case Readiness
AI has build costs and running costs. These questions test whether the investment is realistic.
- Have we estimated the value of solving this problem? Hours saved, faster response, fewer errors or more revenue, even as a rough range.
- Is there a budget for a pilot and for running costs? Model usage, hosting, monitoring and improvements continue after launch.
- Have we agreed on what success looks like after the pilot? The metric, the target and who decides whether to scale.
- Do we have a plan if the pilot shows the approach doesn't work? Stopping early is a good outcome if you learn cheaply.
For a breakdown of what drives cost, read how much AI automation costs.
Scoring Table: What Your Total Means
Add up your 20 scores and find your band.
| Score | Readiness level | What it means | Recommended next step |
|---|---|---|---|
| 33–40 | Ready to build | Data, process, ownership and controls are in place | Scope a pilot with clear success metrics and start building |
| 24–32 | Ready with gaps | The use case is sound, but two or three areas need work | Run a short discovery to close the gaps, then pilot |
| 14–23 | Foundations first | Several areas are unclear or missing | Fix data access, process definition and ownership before building; consider a very small proof of concept |
| 0–13 | Not yet | The use case or organization isn't ready | Pick a different, simpler use case or invest in data and process basics first |
Also look at the pattern, not just the total. A zero on question 9 (no owner) or question 15 (no access control) is a blocker even if everything else scores well.
Which gaps matter most?
Some gaps are easy to close during a pilot; others stop a project cold.
| Gap | Can it be fixed during the pilot? |
|---|---|
| Messy but accessible data | Usually yes, for the slice the use case needs |
| No programmatic access to data | Sometimes, but plan integration work first |
| No business owner | No. Name one before you start |
| No success metric | No. Agree on it before you start |
| Unclear compliance position | No for sensitive data. Get a decision first |
| Limited in-house AI skills | Yes, with the right partner and a handover plan |
Common Mistakes When Assessing AI Readiness
- Assessing "the company" instead of a use case. Readiness is specific to the problem.
- Treating data perfection as a prerequisite. You need usable data for one use case, not a perfect data lake.
- Skipping the people questions. Technology is rarely the bottleneck.
- Ignoring running costs. Budget for operating and improving the system, not just building it.
- Choosing a partner before choosing a use case. Decide what to solve first; see how to choose an AI development company.
Get a Second Opinion on Your Score
Aaga is an AI-native engineering and IT services company that helps teams go from a readiness check to a working pilot, with senior engineers involved from the first conversation. Bring your completed checklist to a free consultation and we'll review your scores, suggest the best first use case and outline a scoped pilot. Prefer to understand budgets and team setup first? See our engagement models for how pilots, projects and dedicated teams work.
