Read time: ~5 minutes · Series: Analyst advanced · Previous: Cache and Source Run · Next: Plan vs Memory
What they do
Lantide includes AI data analysis tools (t-test, chi-square, ANOVA, linear/logistic regression, correlation, forecasting, clustering, PCA, ARIMA, etc.), enabled by the Agent in chat via activate_analysis.
Key points:
- Statistics run on SQL-prepared tables, not instead of SQL
- Results appear mainly in AI chat; won't modify raw data files
- Fits "after Execute, run tests/models on this clean table?"
Full tool list and parameters: USER_GUIDE §12.9.
When you'll use them
| Scenario | Typical flow |
|---|---|
| A/B conversion difference | SQL aggregate two groups → t-test / chi-square in chat |
| Retention curve interpretation | SQL cohort table → chat interpretation + modeling if needed |
| Multi-factor correlation | SQL wide table → correlation / regression |
Agent may also suggest Plan checkpoints per analysis methodology—methodology guides how to write Plan, not skipping SQL.
Two rules analysts must know
1. No formal statistics in Planning
While revising Plan and definitions, Agent should focus on schema exploration and plan edits. If Agent rushes tests during Planning, put denominator and sample definition in Plan first.
2. Agent confirms before running
In Quick or Executing, before statistical tools Agent usually asks via interactive questions: column mapping, control group, significance threshold, etc. Don't skip confirmation—wrong parameters make fast tests meaningless.
vs Excel / Python notebook
| Chat statistical tools | Export and model yourself | |
|---|---|---|
| Definition | Upstream SQL stays in Lantide | Easy to drift from pull logic |
| Reproducibility | Same SQL + chat record | Depends on personal habit |
| Depth | Common tests and basic ML | Almost unlimited |
Suggestion: Pull and definitions in Lantide SQL; exploratory tests tied to Plan in chat tools; highly custom pipelines export or notebook, but formal delivery should point back to SQL / Report.