Is Your Data Ready for AI? An Honest Checklist

Here's the uncomfortable truth behind most failed AI projects: the algorithm was fine. The data wasn't. Models are only as good as what they learn from, and in most businesses, the data an AI project needs is scattered across systems, inconsistently captured, and quietly contradictory. The good news: data readiness is fixable, and fixing it pays off even before any AI arrives.

The five questions that predict success

• Can you get to it? If the data lives in one vendor's system with no export, or in spreadsheets on personal drives, access comes first.
• Do the numbers agree? If sales and finance report different revenue for the same month, a model will faithfully learn the confusion.
• Is history deep enough? Forecasting and propensity models want at least 18–24 months of consistent history to learn seasonality and behaviour.
• Is it captured consistently? Free-text fields, optional columns, and "we changed how we code that in 2024" all degrade what a model can learn.
• Is it governed? POPIA applies to model training and AI processing like any other use, ownership, lawful basis, retention, and access controls need answers.

What "fixing it" actually involves

This is data engineering: pipelines that pull your sources together automatically, a warehouse or lake where the numbers reconcile, definitions everyone shares, and quality rules that catch problems at the source. It's unglamorous, and it's the single highest-leverage investment on the AI journey, because every model, dashboard, and automation you ever build stands on it.

The staged path that works

• Stage 1, consolidate: automated pipelines from your core systems into one governed platform.
• Stage 2, see: dashboards and reporting on the clean foundation. Immediate business value, and it flushes out remaining quality issues fast.
• Stage 3, predict: now models have something trustworthy to learn from, forecasting, propensity, anomaly detection.

Notice that stages 1 and 2 deliver value on their own. That's why we sequence AI work this way for clients: you're never spending months on plumbing with nothing to show. If you're wondering where your organisation sits on this curve, our AI readiness assessment answers it in a few weeks, honestly, including "you're not ready yet, do this first" when that's the truth.