Barry Callebaut Digital Twin
An enterprise Digital Twin and AI-powered manufacturing simulator designed to optimize factory throughput, product quality, and global operational visibility.
Project Overview
Barry Callebaut operates one of the world's largest chocolate manufacturing networks, producing thousands of tons of chocolate products across multiple factories worldwide.
Manufacturing operations involve hundreds of production parameters—including ingredient ratios, equipment utilization, production schedules, machine temperatures, throughput targets, quality checkpoints, and logistics constraints. Production planners and quality teams often need to answer questions such as: "What happens if demand suddenly increases by 10%?", "Can Plant B absorb additional production?", or "Will changing a tempering configuration improve throughput without reducing quality?"
"Traditional planning methods relied heavily on spreadsheets, disconnected manufacturing systems, manual calculations, and delayed reporting. This made production planning reactive instead of predictive."
The goal of this project was to design an enterprise Digital Twin platform capable of simulating factory operations before changes were deployed in the real world. The platform combined real-time production monitoring, predictive simulations, AI-assisted recommendations, and governance workflows into a single unified experience.
Business Context
Manufacturing organizations generate enormous volumes of operational data every minute. However, data alone does not enable better decisions. Production planners constantly balance competing priorities: throughput, product quality, operational cost, machine capacity, downtime, and delivery commitments.
Volume vs. Packaging
Increasing production volume might overload downstream packaging lines.
Heat vs. Consistency
Changing temperatures may improve throughput but negatively impact chocolate quality.
Multi-Plant Allocation
Shifting production between global factories to meet localized demand spikes may increase transportation costs.
The absence of a centralized simulation platform forced users to rely on assumptions rather than evidence. Barry Callebaut wanted to transition from reactive manufacturing management toward predictive manufacturing using Digital Twin technology.
Define The User Problem
Planners and QC teams face increasing pressure to maintain high throughput, cost efficiency, and product quality—especially in the face of real-time disruptions such as demand spikes or equipment failures. Current tools are reactive, fragmented, and slow, making it difficult to:
Respond to Fluctuation
Rapidly respond to demand fluctuations (e.g., sudden 10% order increases).
Predict Change Impact
Predict how changes in production impact cost, quality, and throughput.
Optimize Configs
Optimize production configurations for evolving volume and SKU mix.
Mitigate Outage Risks
Simulate and mitigate risks from equipment downtimes or line imbalances.
Access Global Data
Access unified, real-time insights across globally distributed factories.
"Manufacturing teams lack a centralized system to test 'what-if' scenarios and foresee downstream effects before changes are made, leading to inefficiencies, delays, and quality issues."
Key Usability Pain Points (Survey Data)
Primary Objectives
The platform was designed around four primary business objectives:
1. Optimize Factory Throughput
Allow planners to simulate production changes and immediately evaluate impact on throughput, resource utilization, and cost.
2. Improve Product Quality
Enable quality teams to monitor critical parameters in real time and detect deviations before downstream packaging.
3. Global Visibility
Create a centralized Watch Tower to monitor every facility through standardized KPIs, alerts, and health metrics.
4. Strengthen Governance
Introduce structured approval workflows, simulation history, and audit trails for traceable decision compliance.
Success Metrics Mapped
Operational Goals
- • Faster production planning
- • Reduced downtime response
- • Higher planning accuracy
- • Lower production waste
Experience Goals
- • Faster scenario creation
- • Reduced cognitive load
- • Improved decision confidence
- • Cross-plant visibility
Business Goals
- • Increased planning efficiency
- • Reduced quality losses
- • Centralized executive decisions
My Design Responsibilities
As the Product Designer, I owned the end-to-end user experience for the Digital Twin platform:
UX Research & Interviews
Information Architecture
Simulation Workflow Design
High-Fidelity UIs & Prototypes
The challenge extended beyond designing dashboards—it required simplifying highly technical manufacturing processes into workflows that planners, quality controllers, and executives could confidently use under time-sensitive conditions.
Target Users
Production Planner
- Goals: Increase throughput, balance production load, allocate resources.
- Pain Points: Excel reliance, slow calculations, no downstream impact check.
Quality Controller
- Goals: Maintain consistency, detect deviations early, prevent bad batches.
- Pain Points: Late issues detection, limited real-time monitoring, fragmented logs.
Operations Director
- Goals: Monitor global factory performance, improve resource utilization.
- Pain Points: Limited cross-plant visibility, delayed reporting, no unified console.
Barry Callebaut Chocolate Factory Simulator
Adjust the tempering temperature and conching time to run simulated Digital Twin projections.
Select a preset sequence in the sidebar or adjust parameters and click execute to trigger Digital Twin telemetry projections.
Design System & Component Library
The project utilized Barry Callebaut's brand red (`#E1251B`) and rich brown (`#4B2F21`) to style a consistent, high-density interface:
Line 3 — Conching & Tempering Phase
Crystallization Telemetry
User Research
To ensure the product addresses real user needs and is usable in live environments, we focused our research on four key goals:
Understand Decision Workflows
Map how planners and QC leads currently respond to order changes, line outages, and quality dips.
Validate Simulation Value
Assess whether the ability to simulate trade-offs (cost, quality, throughput) would change user behavior compared to current reports.
Measure Wizard Usability
Determine whether users can intuitively navigate and complete the four-step scenario simulation flow without confusion.
Quantify Alert Fatigue
Identify pain points related to information overload and validate preferences for triage filters and prioritization.
Mixed-Method Methodology
We used a mixed-methods approach to gain both deep qualitative insights and broad quantitative validation:
a. Qualitative Research
• 8 total participants (5 Planners, 3 QC Leads)
• 60-min sessions via Zoom
• Walkthroughs of dashboards, mental models, low-fi prototype tests
b. Quantitative Research
• Online surveys via Google Forms
• 30 users across 4 factories
• Metric capture: pain point frequencies, KPI/alert prioritization ranking
c. Usability Testing
• Maze prototype tests
• Tasks: a. Run simulation, b. Filter alerts, c. Compare two "what-if" plans
• Metrics: completion rates, time on task, SUS scores
Synthesized Affinity Mapping
• Real-time visibility into line efficiency & bottlenecks
• Immediate cost projections tied to downtime or rush orders
• *"I simulate in Excel; it's error-prone."* — Planner, Plant 6
• Manual workarounds lead to wasted time & oversight risks
• *"Compare my last 3 simulations side-by-side."* — QC Lead
• Strong interest in benchmark tools & tracking outcomes
The Opportunity
"The research made one thing clear: manufacturing teams didn't need more dashboards—they needed better decision support."
The opportunity was to design an intelligent manufacturing platform that could transition users from monitoring to decision-making, using Digital Twin technology as the foundation.
Design Principles
Visibility Before Complexity
High-priority KPIs, alerts, and operational summaries are surfaced first, allowing users to assess factory performance at a glance.
Guided Decision-Making
Progressively guides users through monitoring, investigation, simulation, comparison, and approval to reduce cognitive load.
Context-Driven Analytics
Every chart and recommendation shows current state, historical trends, predicted outcomes, and business impact contextually.
Explainable Intelligence
AI recommendations assist—not replace—human decision-making, presenting clear confidence metrics.
Product Ecosystem & IA
Rather than a single dashboard, the solution comprises three hierarchical layers:
Layer 1: Enterprise Overview
Global factory status, plant comparison, and critical alerts watch tower designed for executive leadership reviews.
Layer 2: Operational Monitoring
Live production metrics, equipment health, line efficiency, and quality monitoring for plant managers.
Layer 3: Decision Support
What-if simulations, AI-generated recommendations, and structured governance approval workflows.
Shallow Navigation Strategy
Intentional shallow tabs structure: Dashboard, Global Watch Tower, Live Monitoring, Simulator, Reports, Notifications, Administration.
Key Design Decisions
1. Separating Monitoring from Decisions
Avoided clutter by splitting the platform into three dedicated modules (Watch Tower, Live View, Simulator) matching specific user roles.
2. Decisions Over Dashboards
Every visualization answers a specific question (e.g. "Can this factory meet next week's demand?") rather than just dumping telemetry data.
3. Progressive Disclosure
Complexity is revealed progressively: Global View → Factory View → Machine View → Simulation → Recommendations → Approval.
4. Human-Centered AI
AI acts as an advisor, explaining why a recommendation was generated and showing associated trade-offs.
Technical Considerations
Real-Time Telemetry
Supported continuously updating machine telemetry without disrupting user interactions or layout stability.
Asynchronous Sim Runs
Designed around staged result generation and loading indicators to clearly communicate computation status.
Scalable Architecture
Modular design system ensures additional plants and future analytics (energy, sustainability) plug in easily.
The Final Solution
The final solution transformed fragmented manufacturing systems into a unified enterprise platform that combines operational monitoring, predictive simulation, and AI-assisted planning. Planners can observe, simulate, compare, and approve production changes within a single closed-loop ecosystem, shifting from reactive problem-solving to proactive optimization.
Business Impact
Future Roadmap
Potential future enhancements for the Barry Callebaut Digital Twin platform:
Predictive Maintenance
Integrating machine health monitoring to forecast equipment failures before they cause line downtime.
Sustainability Analytics
Monitoring energy consumption, carbon emissions, and water efficiency across factories.
Supply Chain Visibility
Connecting procurement logistics with real-time factory scheduling.
AI-Powered Scheduling
Automatically recommending optimized daily production schedules based on raw cocoa constraints.
Key Learnings & Reflections
Working on the Barry Callebaut Digital Twin Platform reinforced that successful manufacturing products are built around **decisions, not data**. Trust in automated AI recommendations is earned not through automation alone, but through transparency, explainability, and maintaining meaningful human control over complex decisions.