Root Cause Analysis

Root Cause Analysis is AskEdi's response to questions that ask why a metric moved. Rather than describing a line going up or down, AskEdi runs a real regression, ranks segments by their actual contribution to the change, tests which other metric moved with it, and scans for anomalies, then ties all four together in a single answer.

---

When Root Cause Analysis Activates

AskEdi classifies the intent of your question automatically. Questions that typically activate Root Cause Analysis include:

• "Why did revenue drop last quarter?"
• "Why is on time delivery declining in the warehouse segment?"
• "What is driving the increase in returns this month?"
• "Why did average order value fall in the second half of the year?"

You do not need a special keyword or command. There is no separate mode to select, AskEdi recognizes a causal "why" question from its phrasing in any of AskEdi's supported languages.

---

What Data This Needs

Root Cause Analysis gets more useful the more your table has of these:

• A real time span on the metric. Enough history to tell a real trend apart from normal variation. A handful of data points produces a technically valid but low-confidence answer.
• A segment column with more than one value. The Contribution & Contrastive Analysis section needs at least two distinct values to rank (regions, stores, product categories). A column already filtered to one value has nothing to contrast.
• Other numeric columns in the same table. The Correlated Drivers section checks every numeric column against your metric. A table with only the one metric and no other measures produces a thinner report, since there is no candidate driver to find.

---

What a Root Cause Analysis Response Includes

A Root Cause Analysis response is always four sections of narrative followed by a highlighted Decision & Impact Summary card. Every section is grounded in a real computation, never an estimate written by the AI.

1. Executive Observations

States the direction, size, and timeframe of the change: what the metric did, over what period, and whether the movement is a real trend or within normal variation. This is backed by a real regression fit on the metric's history, not a description of the chart.

2. Contribution & Contrastive Analysis

Ranks the segments of your data (regions, categories, stores, whatever dimension you asked about or AskEdi determines is relevant) by their real, computed share of the change. Each segment's contribution is an exact calculation, not an estimate. When your question naturally breaks the data down by both segment and time period, AskEdi additionally tests whether the difference between segments is statistically real or could be normal variation.

3. Correlated Drivers

Names the other metric in your table that moved most closely with the one you asked about. AskEdi checks every numeric column available, not just the first two it encounters, and selects the strongest match. It then tests whether that relationship is statistically significant. If the relationship could plausibly be coincidence at your sample size, the response says so rather than presenting it with false confidence.

4. Outliers & Anomalies

Surfaces the specific periods or data points that deviate furthest from the rest of the series. If nothing in your data crosses the anomaly threshold, AskEdi states that plainly, the movement reflects gradual drift rather than a single spike or break.

5. Decision & Impact Summary

The closing section renders as a highlighted card, not plain text, so it stands apart from the rest of the response. It contains five labeled lines:

A sixth line, Validation, states the specific metric to track and the condition that would confirm or reverse the recommendation.

---

What This Looks Like

A shortened example, for a question like "why is on time delivery declining?":

Executive Observations Monthly on time delivery rate moved from 41% in January to 35% in October, comparing the first half of the year against the second. The next-period projection is 56%, but trend reliability is low, so this movement is not yet distinguishable from normal variation.

Contribution & Contrastive Analysis The largest concentration of second-half pressure sits in the warehouse segment, ranked #1 with 21% of the observed volume. The absolute on time delivery rate for warehouse is 34%, versus 67% for high street.

Correlated Drivers Cancellation rate is the strongest co-mover with on time delivery rate, with a negative correlation of -0.70 that is statistically significant (p=0.03).

Decision & Impact Summary (highlighted card) Recommended action: Prioritise a recovery plan for the warehouse segment. Confidence: low, because the underlying trend is not yet statistically reliable.

The real response is longer and grounded in your own column names and numbers. This shows the shape, not the content.

---

Verifying the Numbers Behind the Answer

Every figure in a Root Cause Analysis response, the regression slope, the contribution percentages, the correlation coefficient, the anomaly z-scores, is computed server-side before the response is written. Two places let you inspect that work directly:

• Analysis view: shows the exact query that ran against your data source.
• Methodology Notes: shows the statistical method used for each section, the sample size, and the real test result, in plain language, independent of what the response text says.

---

Frequently Asked Questions

---

Next Steps

Need help? Email support@edilitics.com with your workspace, job ID, and context. We reply within one business day.