AI Readiness Check: 12 Questions Before Your Company Starts with AI
A concrete self-test with scoring: 12 questions on decision, data, ownership, regulation and measurability. At the end you know whether you are ready — or exactly where the real gap is.
Most companies that fail with AI do not fail on the technology. They start before it is clear which decision should improve and whether the conditions for it exist.
We argued in an earlier article why AI starts with a decision, not a model. This article turns that into a concrete tool: 12 questions, each scored 0, 1 or 2. Answer them honestly for one planned use case — not for "AI in general".
There is a scoring guide at the end. The point is not a good number, but an honest one.
How to score
For each question:
- 0 points – not true, or unclear
- 1 point – partially in place
- 2 points – clearly met
Maximum 24 points. The interpretation is at the end.
Block A: the decision
1. Can you phrase the use case as a decision, not a tool? "Flag risky inbound invoices before posting" counts. "Use AI in procurement" does not. A decision has a subject, an action and a moment.
2. Do you know the current cost of the problem? In hours, errors, delays, lost revenue or complaints — with a rough number, not just a feeling. If you cannot quantify the baseline cost, you cannot prove the benefit later.
3. Does a concrete action follow from the AI output? A score nobody uses is a dashboard, not value. It must be clear what happens because of the output.
Block B: the data
4. Do you know where the data this decision needs lives? Not "we have lots of data", but: exactly these five to twelve fields, in this system.
5. Is there a person who trusts that data and owns it? Data without an owner is more dangerous in AI projects than missing data. If nobody can explain why a value changes, the AI output will not be trusted.
6. Are the old realities (Excel, email, legacy ERP) part of the plan? If the plan assumes a clean data world that does not exist, it is not a plan.
Block C: ownership and the human
7. Is a person named who approves the AI-assisted action? Not "the department" — a name. Human-in-the-loop is not a disclaimer, it is a role.
8. Is it clear what happens when the AI is uncertain? Good systems show confidence and escalate to a human in case of doubt. With no escalation path, the system is not production-ready.
9. Are IT, the business unit and, where relevant, data protection involved from the start? AI projects rarely fail technically and often fail organisationally. Late involvement creates late blockers.
Block D: regulation and measurability
10. Have you checked whether the use case is regulatorily sensitive? The EU AI Act phases in: obligations for general-purpose AI have applied since 2 August 2025, transparency obligations from August 2026, and requirements for certain high-risk systems land across 2026 to 2027. A drafting assistant is assessed differently from a system that decides about people. This question needs no legal department — just a deliberate classification.
11. Is data residency and hosting region settled? With personal or confidential data, the hosting region is not a detail. It is part of readiness, not a later step.
12. Is there a business KPI that makes success measurable? Not "the AI works", but a business value with a target: handling time, error rate, first-resolution rate, lead time. Without a KPI there is no end and no proof.
Interpretation
Add up your points.
0–9 points: not ready yet — and that is the most valuable finding. You are not blocked by technology but by clarity. Before a pilot starts, the decision, the data location and ownership should be defined. A pilot here would mostly reveal what is missing — more expensive than a scoping conversation.
10–17 points: pilot-ready with targeted preparation. The core is there, but one to three gaps (often data owner, KPI or escalation path) should close before the start. Those exact gaps are the agenda of the first two pilot weeks.
18–24 points: ready for a controlled 90-day pilot. Decision, data, ownership and measurability are clear. The next step is not another concept but a small, measurable pilot with a stop criterion.
What a low score really means
A low number is not a bad sign. It is a precise one. Most expensive AI misinvestments happen because a project with 8 readiness points is started as if it had 20. The test has then done its job before budget was committed.
The most common gap in mid-sized companies is not the model, the budget or the talent. It is question 2 and question 5: nobody can quantify the baseline cost, and nobody owns the data.
Frequently asked questions
Should we run the test per use case or once overall? Per use case. "Are we AI-ready?" is not a useful question. "Are we ready for this decision?" is.
What if we are unsure on question 10 (regulation)? Uncertainty here is a 1, not a 0 — as long as you ask the question deliberately. A proper classification belongs in pilot preparation, not at the end.
Is a high score enough to start? It is necessary, not sufficient. Twenty points without a named Decision Owner are twenty points without a pilot.
How often should we reassess? Before every new use case and at the end of every pilot. Readiness is not a state but a recurring check.
Conclusion
AI readiness is measurable, but not through a tool inventory. It is measured by decision, data, ownership, regulation and KPI. These 12 questions do not replace a conversation — but they prevent the most expensive misunderstanding: starting a large project where clarity is actually missing.
Next step
You took the test and scored under 18? That is exactly where we start. An AI readiness check with OzyCore closes the concrete gaps — decision, KPI, data owner — and only then leads into a controlled 90-day pilot once the conditions are in place.
Sources
- NIST, AI Risk Management Framework — nist.gov
- European Commission, AI Act — digital-strategy.ec.europa.eu
- Bitkom, Digitalisierung der Wirtschaft: Fast jedes Unternehmen beschäftigt sich mit KI (2026) — bitkom.org
- Destatis, Companies using artificial intelligence — destatis.de