The Data Readiness Checklist: 12 Questions to Answer Before Any AI Project

Ask any experienced practitioner why analytics and AI projects fail and you will hear the same answer: the data was not ready. Not "the model was wrong" or "the tool was bad". The data was scattered, inconsistent, undocumented, or simply not there.

The good news is that data readiness is checkable. Before committing budget to any intelligence initiative, work through the questions below with your leadership team. Be honest. A "no" is not a failure, it is a work item.

Access: can you actually get to your data?

1. Do you know where your critical data lives? List the systems that hold customers, sales, inventory, finance, and HR. If the honest answer includes "in Priya's spreadsheet", write that down too. Visibility comes before quality.

2. Can you extract it without a vendor's permission? Many businesses discover their POS or ERP vendor charges heavily for exports, or offers none at all. Locked-in data is not ready data, and it should influence every future software purchase.

3. Is there one place where data comes together? It does not have to be a fancy warehouse. Even a well-managed set of scheduled exports into one database beats five systems that never meet.

Quality: can you trust what you get?

4. Are your key entities deduplicated? The same client appearing three times with three spellings will quietly poison every analysis downstream. Customer and product master data are the usual suspects.

5. Are critical fields actually filled in? Run a simple completeness check: what percentage of records have a valid email, a sector code, a closing date? Fields that staff skip in the source system cannot be conjured later.

6. Do the numbers reconcile with finance? If revenue in your reporting layer does not match the accounts, nobody will trust anything built on it. Reconciliation is the fastest trust-builder there is.

Meaning: does everyone agree what the data means?

7. Are your key metrics defined in writing? "Active customer", "gross margin", and "on-time delivery" mean different things in different departments until someone writes a definition and gets it signed off.

8. Is there a data dictionary, even a modest one? A shared document explaining what each important table and field contains saves every future project weeks of archaeology.

9. Do you know the freshness of each source? Daily, hourly, or real-time matters. A demand forecast built on data that arrives monthly will always be a month behind reality.

Governance: are you allowed to use it?

10. Do you know your legal obligations? In Mauritius, the Data Protection Act 2017 governs personal data, and if you serve European clients, GDPR follows the data. Map which datasets contain personal information before any model touches them.

11. Is access controlled and logged? Readiness cuts both ways: the right people should reach the data easily, and everyone else should not. Shared admin passwords are a red flag worth fixing this quarter.

12. Is someone accountable? Not a committee, a person. Every serious data effort we have seen succeed had a named owner with the authority to chase issues across departments.

Scoring yourself honestly

Count your confident yes answers.

• 9 to 12: you are ready for predictive and AI work now. Your risk is under-ambition, not overreach.
• 5 to 8: typical of most established firms. Run a focused three-month readiness sprint on your weakest area before funding an AI pilot, and scope the pilot around your strongest data.
• 0 to 4: do not buy AI yet. Spend the same money consolidating systems and cleaning master data. It will feel unglamorous and it will pay back for a decade.

The mindset shift

The checklist is not a gate you pass once. Data readiness decays: staff change, systems get replaced, definitions drift. Treat it as an annual health check, the same way you audit the accounts.

And resist the temptation to aim for perfection before starting anything. Readiness is scoped to a use case. You do not need every dataset clean, you need the two or three feeding your first project to be trustworthy. Teams that pair one readiness fix with one visible business win build momentum. Teams that launch a two-year "fix all the data" programme usually run out of political capital first. If you want an outside view, a structured readiness assessment from a partner such as nexus.mu can compress months of internal debate into a few weeks of findings.