10 Ways AI is not BI

If you were dissatisfied with the data used in your business dashboards, just wait until you see how demanding AI systems are! Here are the top ten reasons your business intelligence (BI) data stack is not suitable for AI:

1. Scale: AI requires large volumes of detailed data, which can make data movement expensive and inefficient. Dashboard pipelines that extract a few summary statistics may not be sufficient.
2. Agility: AI systems require frequent changes to data inputs and feature engineering as the world changes. In contrast, dashboard metrics rarely change.
3. Complexity: Each AI decision requires dozens, if not hundreds, of input columns, while dashboard metrics are often based on a single input or ratio of two inputs.
4. Human-in-the-loop: AI systems make decisions without human intervention, so relying on humans to apply common sense to spot errors is not practical.
5. Time-awareness: AI-ready data must be carefully joined and feature engineered in a time-aware manner.
6. Sampling: AI systems are trained on carefully constructed samples of historical data, while dashboards use totals and subtotals.
7. Computational cost: AI-ready data requires complex feature engineering that is computationally complex, while dashboards rely on common database-optimized calculations such as summation and partitioning.
8. Training versus scoring: AI systems are trained on historical data, then deployed using live data, requiring two data pipelines with different specifications that must be consistent.
9. Averages versus detail: AI systems require detailed data and cannot depend on data quality issues simply being resolved by averaging out across a broad summary metric.
10. Latency: AI systems need to make near real-time decisions, while dashboards are updated periodically and typically used for strategic decisions.