AI Limitations

clariBI's AI is a powerful analysis tool, but it has real limitations. Understanding them helps you use it effectively and avoid costly misinterpretations.

The AI Can Make Mistakes

The AI engine is a language model that generates statistically probable text. When applied to data analysis, this means:

• Calculations can be wrong. The AI sometimes makes arithmetic errors, especially with percentages, growth rates, and complex aggregations. Always verify important numbers against your source data.
• Correlations are not causation. If the AI says "revenue increased because of the marketing campaign," that is its interpretation, not a proven causal link.
• It can hallucinate details. The AI may state specific numbers or trends that are not actually present in your data. The data table in each response lets you cross-check.

Data Quality Matters

The AI's output quality depends directly on your input data quality:

• Missing values can skew averages, counts, and trends. The AI does not always flag when data is incomplete.
• Duplicate rows inflate counts and sums without warning.
• Inconsistent formats (e.g., dates in multiple formats, mixed text/number columns) confuse the AI's interpretation.
• Stale data produces outdated analysis. Check your data source sync status before running analyses.

Data Size Limits

The AI has a context window that limits how much data it can analyze at once. For very large datasets (millions of rows), clariBI samples or aggregates the data before sending it to the AI engine. This means:

• Row-level details may be lost in the aggregation
• Rare outliers may not be captured in the sample
• The AI may not see every row of your data for very large tables

For precise analysis on large datasets, use specific filters and date ranges to narrow the data before querying.

What the AI Cannot Do

• Access external data. It only analyzes data you have connected to clariBI. It cannot browse the internet or access other systems.
• Modify your data. Analysis is read-only. The AI cannot write back to your databases or change source files.
• Predict the future reliably. While it can identify trends, AI predictions are extrapolations, not forecasts. They do not account for market changes, new competitors, or external events.
• Replace domain expertise. The AI does not know your business context, industry dynamics, or competitive landscape. Its recommendations are generic and should be filtered through your knowledge.
• Guarantee privacy. Your data is sent to the AI processing service for analysis. While the service has data handling policies, be aware that data leaves your clariBI instance during analysis. Do not analyze data you cannot send to a third-party API.

Common Accuracy Issues

Percentages and Growth Rates

The AI sometimes confuses percentage change with percentage points, or calculates growth rates incorrectly. When a report says "revenue grew 50%," verify the base and final numbers yourself.

Ambiguous Column Names

If your data has columns like "amount" and "total_amount," the AI may pick the wrong one. Use specific column names in your questions to avoid this.

Time Zone Issues

Date-based analyses may shift by a day depending on how your data source stores timestamps. If exact daily numbers matter, verify the timezone handling.

Small Sample Sizes

When analyzing small datasets (under 100 rows), the AI may draw conclusions from insufficient data. Be skeptical of strong claims from small samples.

Best Practices for Reliable Results

1. Always check the data table in each response to verify the raw numbers match your expectations
2. Cross-reference important findings with your original data source
3. Use the AI as a starting point, not the final word
4. Be explicit in your questions to reduce ambiguity (see Writing Better Queries)
5. Keep data sources synced so analyses use current data
6. Clean your data before connecting it — remove duplicates, fix formats, fill obvious gaps

Related

• AI Analytics Overview
• Writing Better Queries
• Conversational Analytics