Skip to main content

Deep Analysis (Iara)

Iara is the advanced analytics engine that runs comprehensive data analysis using techniques like clustering, forecasting, anomaly detection, and more — all from a natural language question.

Screenshot of the Deep Analysis page showing the hero prompt area with example question chips

How It Works​

  1. Ask — Describe what you want to analyze in plain language
  2. Plan — The AI creates an analysis plan with multiple techniques
  3. Run — Analyses execute in parallel in the background
  4. Results — View comprehensive findings with charts, tables, and explanations

Starting an Analysis​

Navigate to Deep Analysis in the sidebar.

Screenshot showing the question input area with example prompts listed below

Example Prompts​

PromptWhat It Runs
"Segment customers by purchasing behavior"Clustering + RFM Segmentation
"Products frequently bought together"Basket Analysis
"Forecast next quarter's revenue"Time Series Forecasting
"What drives my sales the most?"Causality / Driver Analysis
"Find unusual patterns in my data"Anomaly Detection

Analysis Plan​

After submitting your question, the AI generates a plan showing which analysis types it will run:

Screenshot of the analysis plan showing checkboxes for each analysis type with descriptions

You can toggle specific analyses on or off before starting the run.

Analysis Types​

1. Customer Clustering​

Groups your customers into segments based on behavioral patterns.

Screenshot showing clustering results with segment visualization and descriptions

Results include:

  • Number of clusters identified
  • Cluster descriptions (e.g., "High-value frequent buyers", "Occasional bargain hunters")
  • Data points per cluster
  • Clustering method used (K-Means, DBSCAN, etc.)

2. Basket Analysis​

Discovers products frequently purchased together (association rules).

Screenshot showing basket analysis results with top association rules

Results include:

  • Number of rules found
  • Top rules with:
    • Support — How often items appear together (%)
    • Confidence — How likely B is bought when A is bought (%)
    • Lift — How much more likely compared to random chance

3. Time Series Forecasting​

Predicts future values of a metric based on historical trends.

Screenshot showing a line chart with historical data and forecast projection

Results include:

  • Historical trend line
  • Forecast projection with confidence interval
  • Forecasting method (ARIMA, Prophet, etc.)
  • Forecast horizon (configurable)

4. Causality / Driver Analysis​

Identifies which factors most influence a target metric.

Screenshot showing driver analysis with coefficient bars and Granger test results

Results include:

  • Lasso coefficients — Which variables have the strongest impact
  • Granger causality tests — Whether one time series predicts another
  • Rankings of drivers by impact size

5. Anomaly Detection​

Finds unusual patterns and outliers in your data.

Screenshot showing detected anomalies highlighted on a time series chart

Results include:

  • Total anomalies detected
  • Detection method used
  • Table of anomalous data points with dates and values
  • Severity scoring

6. Survival Analysis​

Estimates customer lifetime and retention probabilities.

Screenshot showing a Kaplan-Meier survival curve

Results include:

  • Median survival time
  • Survival probability curve (Kaplan-Meier)
  • Retention rates at key intervals (30, 60, 90 days)

7. RFM Segmentation​

Classifies customers by Recency, Frequency, and Monetary value.

Screenshot showing RFM treemap, heatmap, and actionable segment labels

Results include:

  • Segment breakdown — Treemap showing segment sizes
  • R × F Heatmap — Recency vs Frequency matrix
  • Top/Bottom customers — With anonymized IDs
  • Actionable labels: Reward, Upsell, Engage, Win-Back, At-Risk, etc.

Interpreting Results​

Data Quality Warnings​

The AI may show warnings when data quality affects analysis:

WarningMeaningWhat to Do
Few rowsNot enough data points for statistical significanceUpload more historical data
No varianceAll values are the same — nothing to analyzeCheck data for errors
Short date rangeTime series too short for reliable forecastingWait for more data to accumulate

Suggested Alternatives​

When an analysis type produces low-quality results, the AI suggests alternative approaches:

"Clustering results are low confidence with your current data. Consider running RFM Segmentation instead, which works well with transactional data."

Analysis History​

View past analyses in the History sidebar (right side of the page):

Screenshot of the history sidebar showing past analyses grouped by batch with status chips

  • Analyses grouped by batch (each prompt creates a batch)
  • Status chips (Completed, Running, Failed)
  • Click to view results from any past analysis

Exporting Results​

Click Export PDF to download a formatted report of any completed analysis, including charts and summary text.

info

Tip: Deep analyses are most powerful when combined. For example:

  1. Run Clustering to segment customers
  2. Run Basket Analysis per segment to find segment-specific product affinities
  3. Run Forecasting on revenue per segment to predict growth