ATLAS for Tableau
A decision cognition runtime for Tableau Cloud: live insight confidence, decision intent and decision risk, built with native calculations and no AI.
01
Problem
- Who
- Business leaders and analysts acting on dashboards.
- Why it mattered
- BI tools treat every number as equally reliable; confidence, intent and risk stay invisible, so decisions get made on unstable insights.
- Before
- Judgement left entirely to the analyst.
02
Constraints
Technical
Tableau Cloud only: no AI models, external APIs or third-party services.
Technical
Every signal must update live without slowing the dashboards.
03
Architecture
- Component
- Key decision (click)
Interact
- User interaction layerDashboards, filters, selections
- Perception layerInteraction-driven signals
Infer
- Intent Inference EngineRule-based intent
Govern
- Decision Risk GovernorSafety class + advisories
- Insight Trace CompilerReadable explanations
Remember
- Decision Memory StoreContext over time
Every filter or selection is treated as a decision signal: intent is inferred by rules, confidence is computed from volatility with window functions, and risk is classified from confidence, all live inside Tableau.
04
Key decisions
D1Deterministic computation instead of AI or ML
- Alternatives considered
- An ML model scoring insights
- Why
- Fully explainable, reproducible and aligned with enterprise governance, with no training data or external services.
05
Build highlights
- 01
Insight confidence from volatility
Converts the volatility of profit into a 35–90 confidence score.
sqlIF WINDOW_STD(SUM([Profit])) > 3000 THEN 35 ELSEIF WINDOW_STD(SUM([Profit])) > 1500 THEN 55 ELSEIF WINDOW_STD(SUM([Profit])) > 500 THEN 70 ELSE 90 END - 02
Decision risk classification
Turns confidence into a HIGH / MEDIUM / LOW risk advisory that updates as you explore.
sqlIF [ATLAS – Insight Confidence] < 50 THEN "HIGH RISK" ELSEIF [ATLAS – Insight Confidence] < 75 THEN "MEDIUM RISK" ELSE "LOW RISK" END - 03
Four dashboards, one runtime
Executive Decision Overview, Decision Intent Explorer, Insight Confidence & Risk Analyzer and Decision Memory work as one continuous runtime.
06
Lessons
What worked
- Intelligence in analytics does not require AI.
- Tableau's computation engine can support cognitive systems.
- Making uncertainty visible builds more trust than adding more metrics.
What I'd do differently
- Add a Tableau Extension side panel and organisation-wide decision memory.
07
