AI Studio (Alpha)

mindzie AI Studio is a comprehensive predictive analytics platform for process mining. It empowers anyone - from data scientists to business analysts to process owners - to predict, explain, and optimize anything derivable from process data.

AI Studio Overview

Vision

AI Studio is built on three pillars:

  1. AutoML First - The machine figures out the best approach; humans focus on insights
  2. LLM-Powered Explanation - Everything is explained in plain language with generated reports
  3. Interview-Based Setup - Non-technical users configure predictions through guided conversations

How to Access AI Studio

AI Studio is available in the header navigation for tenants with PreRelease enabled.

  1. Click AI Studio (Alpha) in the header menu
  2. Select a category from the left sidebar
  3. Explore the available features

Feature Categories

AI Studio organizes its capabilities into seven main categories accessible from the left sidebar.

DATA - The Foundation

Manage your data sources and features for machine learning.

Section Description
Event Logs Import and manage event logs for training and prediction
Datasets View and manage enriched datasets ready for ML
Feature Store Reusable feature sets with version control and templates

Key Capabilities:

  • Smart data ingestion with auto-detection of columns
  • LLM-guided mapping through natural language interviews
  • Automatic data quality reports

PREDICT - Core Value

Predict what will happen in your processes.

Section Description
Outcomes Will a case succeed? Customer churn? SLA violation?
Timing Remaining time, completion date, delay probability
Next Steps What activity happens next? What path will the case take?
Resources Who will handle this? Workload forecasting, bottleneck prediction
Costs Total case cost, cost to completion, budget variance
Risks Compliance risk, fraud probability, quality risk scores

Prediction Types:

  • Binary outcomes (yes/no)
  • Multi-class outcomes
  • Probability scores (0-100%)
  • Time estimates with confidence intervals

DETECT - Find Problems

Identify issues before they become critical.

Section Description
Anomalies ML-based detection of unusual patterns in control flow, performance, and semantics
Conformance Compare actual execution against expected behavior (BPMN models, business rules, SLAs)
Drift Detect changes over time in process behavior, model performance, and data distribution

SIMULATE - Explore the Future

Test scenarios and understand potential outcomes before making changes.

What-If Analysis

Run simulations to explore how process changes would impact key metrics. Configure scenario parameters and instantly see projected results.

What-If Analysis

Scenario Configuration Options:

  • Approval Threshold - Adjust monetary thresholds for approval routing
  • Team Size - Model the impact of adding or removing staff
  • Auto-Approve Low Risk - Toggle automatic approval for low-risk cases
  • Max Queue Size - Set queue capacity limits

Simulation Results: The simulation compares your current metrics against the simulated scenario:

Metric What It Shows
Avg Cycle Time End-to-end processing time
Cases/Day Throughput capacity
SLA Compliance Percentage meeting service levels
Cost per Case Average processing cost
Resource Utilization How efficiently resources are used
Bottleneck Time Time spent waiting at bottlenecks
Error Rate Percentage of cases with errors

The Impact Visualization shows at-a-glance whether changes improve or degrade Cycle Time, Throughput, and Quality.

The Simulation Summary provides an AI-generated plain-language explanation of the results, highlighting key improvements and any trade-offs to consider.

Digital Twin

Create a visual, real-time representation of your process. The Digital Twin shows your process map with live simulation capabilities.

Digital Twin

Digital Twin Features:

  • Process Map Visualization - See your discovered process model with all variants
  • Live Simulation - Run simulations through the process to observe behavior
  • Variant Analysis - View all process variants with their frequency percentages
  • Simulation Controls - Start, stop, and monitor simulation progress

The Digital Twin enables you to:

  • Understand how cases flow through your process
  • Identify which variants are most common
  • Test hypotheses about process behavior
  • Visualize bottlenecks and parallel paths

Scenarios

Save and manage pre-built scenarios for common what-if analyses:

  • Staff reduction impact
  • Volume spike handling
  • Process redesign effects
  • Seasonal variation modeling

EXPLAIN - Understand Why

Get clear explanations for predictions and outcomes.

Feature Impact

Understand what drives your model's predictions using SHAP (SHapley Additive exPlanations) values.

Feature Impact

Global Feature Importance: The left panel shows which features have the most influence on predictions across all cases:

  • Time Since Start - How long the case has been running
  • Pending Activities - Number of activities waiting to be completed
  • Customer Priority - Priority level assigned to the customer
  • Order Amount - Monetary value of the order
  • Resource Load - Current workload of assigned resources
  • Has Escalation - Whether the case has been escalated
  • Day of Week - Which day the activity occurs
  • Region - Geographic region of the case

Case-Level Waterfall: The right panel shows how each feature contributes to a specific case's prediction:

  • Green values (+) push the prediction higher
  • Red values (-) push the prediction lower
  • The final prediction combines all feature contributions

AI-Generated Explanation: At the bottom, an AI-generated explanation describes in plain language why the model made its prediction. For example: "This case is predicted to breach SLA primarily due to the 36-hour duration and 4 pending activities. The high customer priority also increases breach likelihood. Low resource load provides a small mitigating factor."

Root Cause

Automated discovery of contributing factors when KPIs deviate from expectations. Identifies the "why" behind process problems with statistical significance.

Process Narrative

LLM-generated plain language explanations of case history. Get a story-like description of what happened in any case and why.

AUTOMATE - Continuous Intelligence

Set up automated workflows and monitoring.

Scheduled Training

Configure automatic model training to keep your predictions accurate as data evolves.

Scheduled Training

Training Configuration:

  • Dataset Selection - Choose which enriched dataset to use for training
  • Algorithm Selection - Pick from multiple ML algorithms:
    • FastForest - Fast, accurate ensemble method
    • LightGBM - Gradient boosting for large datasets
    • FastTree - Decision tree with high performance
    • Linear - Simple, interpretable linear models
  • Search Intensity - Balance between training time and model quality
  • Notification - Get notified when training completes

Activity Predictability Scan: Before training, the system scans your data to show which activities are predictable:

  • Activity - The activity to predict
  • Rating - How predictable the activity is (Recommended, Acceptable, etc.)
  • Percentage - Occurrence rate in the dataset
  • Cases - Number of cases containing this activity

This helps you select activities that will produce reliable predictions.

Alerts & Actions

Configure triggers based on predictions:

  • High-risk case detected -> Email case owner
  • SLA violation predicted -> Create task in workflow
  • Anomaly detected -> Log to investigation queue
  • Model drift detected -> Trigger retraining

Model Refresh

Automated model lifecycle management:

  • Monitor model performance over time
  • Detect when accuracy degrades
  • Trigger retraining automatically
  • Compare new models against current deployments

MODELS - Your AI Assets

Manage your trained models and deployments.

Model Registry

Catalog of all trained models with their status, performance metrics, and version history.

Deployments

Deploy trained models to make predictions available as enrichment operators.

Model Deployments

Deploy to Enrichment: Select a completed training to deploy. The model will be added as an enrichment operator that generates prediction attributes for each case.

Each trained model shows:

  • Model Name - The activity being predicted (e.g., "Imaging Ordered", "Consult Completed")
  • Enrichment - Which dataset enrichment the model was trained on
  • Completed - When training finished
  • Deploy Button - Click to deploy the model

Deployed Models: Once deployed, models appear in the Deployed Models panel. From here you can:

  • Monitor which models are active
  • View prediction capabilities
  • Manage model lifecycle

Deployed models become available as enrichment operators in your data pipelines, automatically adding prediction columns to your cases.

Performance

Track model health and accuracy over time:

  • Prediction volume and latency
  • Accuracy trends
  • Drift indicators
  • Comparison against validation data

Roadmap

AI Studio features are being released progressively. Current focus areas include:

Available Now:

  • Scheduled Training with multiple algorithms
  • Model Deployments to enrichments
  • What-If Analysis with simulation
  • Digital Twin visualization
  • Feature Impact with SHAP explanations

Coming Soon:

  • Cost Prediction - Total case cost estimation and cost driver identification
  • Outcome Prediction - Binary and multi-class outcome predictions
  • Anomaly Detection - Real-time detection of unusual patterns
  • Process Narrative - AI-generated case explanations

Providing Feedback

We welcome feedback on AI Studio! Your input helps shape these features before general release:

  • Email: support@mindzie.com
  • Subject: Include "Alpha Feedback: AI Studio"
  • Include: What you were trying to do, what happened, and what you expected