AI Readiness: Is Your Data Good Enough?

Five questions that predict whether an AI project will reach production, and what to fix first if the answers are no.

When an AI pilot fails, the model is rarely the reason. The reason is usually found months earlier, in data that was incomplete, inconsistent or impossible to access. Here are the five questions we ask before any AI engagement, and what the answers mean.

1. Can you get the data out?

If the data for your use case lives in a system that only exports a PDF once a month, or needs a change request to a vendor to access, the project has a plumbing problem before it has an AI problem. Fix this first: an API, a database connection or a scheduled extract into a data platform.

2. Does it mean the same thing everywhere?

Ask three teams what "customer" or "active policy" means and you will often get three answers. AI amplifies inconsistency. A short exercise to agree definitions for the handful of fields the use case depends on is unglamorous and essential.

3. Is it complete enough?

Missing values, duplicated records and gaps in history are normal. The question is whether they affect the fields the model needs. Profile the data before deciding. A day of profiling can save months.

4. Who owns it?

Every dataset needs someone who can answer questions and approve changes. If nobody owns it, quality drifts and the AI feature quietly degrades after launch.

5. Can you measure the outcome?

If you cannot say today how you will know the AI worked, you will not know in six months either. Define the metric, measure the current baseline and agree the target before writing any code.

What to do if several answers are no

Do not cancel the AI ambition. Reorder it. A modest data platform, agreed definitions and a named owner typically take one quarter and make every later project cheaper. Our AI Readiness assessment covers exactly these questions and produces a roadmap you can act on.