When an AI project struggles, the cause is rarely the model. Far more often, the data it depends on is incomplete, inconsistent, hard to reach, or not allowed to be used the way the project assumed. The good news is that data problems are usually visible early, if you look for them before building.
This checklist helps you assess your data in an afternoon, so you know what is ready, what needs work, and what could block the project entirely.
1. Do you know what data the project needs?
Start from the task, not from the data you happen to have.
- What information would a skilled person need to do this task well?
- Which of that information exists in a system today, and which lives only in people's heads?
- Do you have examples of the task being done well, such as good replies, correct classifications, or approved documents?
Examples of good work are often the most valuable data you have. They show the AI what "good" looks like and give you a way to test it.
2. Can you actually get to it?
Data that exists is not the same as data you can use.
- Where does it live: a database, a SaaS tool, shared drives, email, paper?
- Can it be accessed through an export or an API, or only by hand?
- Who owns each source, and will they support the project?
If the critical data sits in a system with no practical way to access it, that is worth knowing on day one.
3. Is it accurate and consistent?
- Duplicates: the same customer or product recorded several times
- Conflicts: two documents giving different answers to the same question
- Gaps: important fields that are often empty
- Stale content: outdated policies, prices, or procedures still in circulation
- Inconsistent formats: dates, names, units, or categories recorded differently across sources
You do not need perfect data. You need to know where the problems are, and to fix the ones that affect the task you are automating.
4. Is it in a usable format?
- Clean text and structured data are the easiest to work with.
- Scanned documents, images, complex tables, and slides need extra processing and are more error-prone.
- Very long documents may need to be split into sections that make sense on their own.
Test a small sample of your hardest formats early. They are where quality problems usually appear.
5. Are you allowed to use it this way?
This is the question teams most often skip, and the one that can stop a project late.
- Does the data include personal information? If so, which privacy rules apply?
- Do contracts with customers or partners limit how their data may be used?
- Is any data confidential to certain teams or roles?
- Will the data be sent to an external AI provider, and is that acceptable under your policies?
When in doubt, involve your legal or compliance contact before building, not after.
6. Will it stay up to date?
- How often does the underlying information change?
- Can updates flow into the AI system automatically?
- Who is responsible for keeping content accurate after launch?
An assistant that answers from last year's policy is worse than no assistant at all.
7. Can you measure quality with it?
Set aside a sample of real cases with known correct answers, and do not use them for anything else. This test set lets you measure whether the AI is good enough, and whether each change makes things better or worse.
Turning the checklist into a plan
After going through the questions, sort what you found into three groups:
- Ready: sources you can use as they are
- Fixable: gaps you can close with modest effort, such as cleaning up a document collection or adding an export
- Blockers: issues that could stop the project, such as no access, no permission, or no examples of good work
Start the project with the ready sources, fix the fixable ones in parallel, and resolve blockers before committing significant budget.
The bottom line
Data readiness is not a separate project you must finish before using AI. It is part of the work, and a short assessment up front saves weeks of surprises later.
If you would like help assessing whether your data is ready for a specific AI use case, we are happy to take a look with you.
