By Laas NuryBefore a company spends money on AI, one question saves the most: is this process actually ready for it? Most failed AI projects did not fail on the technology. They failed on a process nobody could describe, data nobody trusted, a system nobody could connect to, or a team that was not involved.
This self-check takes about twenty minutes per process. Answer twelve questions, add up the score, and you know whether to start now, what to fix first, or whether this process should wait.
How to use it
Pick one process, not the whole company. For example: processing purchase invoices, handling incoming orders, planning deliveries, answering customer questions about orders. Then score each question:
- 0 = no, or we do not know
- 1 = partly
- 2 = yes
Answer honestly, and ideally with the person who runs the process every day, not only with the manager.
The process
1. Can you describe the process in steps, including who does what? If the process only exists in people's heads, an AI cannot follow it. A one-page description of the steps is enough.
2. Do you know the rules and the most common exceptions? Which cases go a different way, and why? The exceptions are where AI projects succeed or fail.
3. Does it happen often enough to matter? Dozens or hundreds of times a week scores 2. A few times a month scores 0: that work is better left to a person with a good assistant.
The data
4. Is the information the process needs available digitally? E-mails, PDFs and system records all count. Paper that nobody scans does not.
5. Do you have examples of past cases where you know the right outcome? A few hundred handled cases are enough to test an AI against. Without them, nobody can say whether it works.
6. Do people trust the data in the systems involved? If the team keeps its own spreadsheet because the ERP is wrong, fix that first, or build that knowledge into the rules.
The systems
7. Can the systems involved be connected? An API or standard connector scores 2. A system that can only be reached through its screens scores 1. A system nobody may touch scores 0.
8. Is there a test environment, or a safe way to try things? Starting in the background (the AI suggests, a person decides) counts as a safe way.
9. Is it clear where the data may be processed? In the EU only, on your own servers, or anywhere with the right contract? If nobody knows, find out before you choose a solution.
People and rules
10. Is there one person who owns the process and can decide? Someone who can say "this is how we handle that exception" without a meeting.
11. Does the team that runs the process want this? Not "have they been told", but do they see what it would take off their plate? Their support decides whether it is used.
12. Is it clear whether this use involves decisions about people or sensitive data? If the process decides about applicants, staff, credit or insurance, it is probably a high-risk use under the EU AI Act, with extra obligations. Knowing that up front changes the plan.
Your score
| Score | What it means | What to do |
|---|---|---|
| 19 to 24 | Ready | Start. Design the first version for production, not as a demo, and run it in the background first. |
| 12 to 18 | Ready with work | Fix the questions that scored 0 first. Usually it is the exceptions (2), the test cases (5) or the owner (10). That work often takes weeks, not months. |
| 0 to 11 | Not yet | Pick another process, or make this one the preparation project: describe it, clean the data it needs, appoint an owner. Then check again. |
A low score is not a verdict on the company. Most companies have a few processes that score high and many that score low. The skill is starting with the right one.
The questions that matter most
If you only have time for three, ask these:
- Question 2 (the exceptions). Every AI project that stalls, stalls here.
- Question 5 (past cases with a known outcome). Without them, you cannot prove it works.
- Question 10 (one owner). Without one, nobody decides and nothing goes live.
After the check
A process that scores well is a candidate for a first build. Before you start, read from AI pilot to production: it explains how to set up the first version so that it actually goes live. And if you are still weighing whether an off-the-shelf assistant would do, Copilot, ChatGPT or a custom AI build helps with that choice.
Want a second opinion on your score? Send us the process and your answers. The first conversation is free.
