Co-written by Valentine LETZELTER , Raphaël De Neeff
Introduction
We’re no longer just improving processes — we’re redefining how work gets done.
AI is no longer just a tool for automation or analytics. With Agentic AI and digital twins, organizations can now simulate decisions, adapt operations in real time, and design self-improving systems that respond faster and smarter to change.
In this series, we will explore how AI enhances DMAIC, unlocks proactive improvement, and lays the foundation for autonomous, data-driven decision-making — accessible not just to large companies, but also SMBs, schools, and public institutions.
And finally, why AI also Stands for Accelerating Improvement!
Harnessing AI for Process Improvement and Developing Agentic AI: Enhancing DMAIC Methodology and Creating Digital Twins of Business Processes to Foster Quality, Proactivity, and Speed
When we first started leading improvement initiatives, we found ourselves spending more time analyzing than actually improving. We relied on numerical data, constrained by the limits of available tools — even powerful ones like Minitab. Inputs and outputs were manually handled, and the scope of insights was narrow.
That paradigm has changed.
In the first article of a series about next digital Transformation, we’ll explore how Artificial Intelligence (AI) — including Agentic AI — is reshaping process improvement. We’ll see how AI enhances the improvement method DMAIC (Define, Measure, Analyze, Improve, and Control) how we can start building digital twins of business processes, and how this transformation enables organizations to proactively manage quality, reduce risk, and accelerate improvement.
From Static Data to Smart Decisions
Process improvement is rooted in data. And today, organizations are swimming in it.
From manufacturing and logistics to customer service, finance, and HR — businesses generate vast volumes of operational, transactional, and behavioral data. Beyond numerical data, there’s now a goldmine of non-numerical inputs: customer surveys, social media content, image and video analysis, even sensory data like color or taste.
AI unlocks the power of all this data. It helps uncover patterns, identify root causes, simulate future scenarios, and suggest actions — at a scale and speed human teams alone could never match.
Let’s look at how AI enhances each phase of the DMAIC methodology framework.
Define
In the Define phase, AI helps identify and prioritize improvement opportunities. Machine learning models can quickly analyze historical and operational data to pinpoint areas of inefficiency or potential failure — broader and faster than traditional root cause tools.
AI also empowers subject matter experts to start building digital twins — virtual models of business processes that reflect real-world behaviour.
Measure / reduce and de Noise data
In the Measure phase, data is extracted, and the Data Scientist plays a crucial role in ensuring:
– Consistency and completeness of the data, as well as its relevance.
– Interpolation to reduce dataset size and noise, while also completing any missing information.
This foundational step is where the synergy between data scientists and AI becomes evident. For instance, I recall a project from years ago where contact center agents had to manually tag the reasons for calls—a tedious and error-prone process. Today, AI-powered systems utilize Natural Language Processing (NLP) to automatically classify call reasons based on the content of the conversations.
This advancement provides immediate, reliable input for performance metrics and analysis, enhancing the accuracy and efficiency of the Measure phase. The Data Scientist can leverage these AI tools to ensure that the data collected is not only consistent and complete but also relevant to the project’s objectives. By integrating AI in the data measurement process, we can significantly reduce manual errors and gain insights faster, ultimately driving better decision-making.
Analyze
During the Analyze phase, AI shines by revealing hidden patterns and correlations that traditional techniques often miss. Predictive analytics can anticipate future problems, helping teams address root causes before they create defects or delays.
AI also supports stronger, faster root cause analysis. It provides structured, data-driven inputs for tools like the Fishbone Diagram or 5 Whys, and can validate assumptions by comparing scenarios or historical cases.
Importantly, AI can detect process inefficiencies, bottlenecks, and low-skill repetitive tasks, allowing human experts to focus on more valuable interventions.
Improve
In the Improve phase, AI is used to simulate improvements before implementation. Digital twins allow teams to test “what-if” scenarios, predict outcomes, and optimize processes without disrupting operations — just like smart cities use digital twins to test traffic flows or public safety responses. Business knowledge remains critical here: experts are essential to validate the twin model’s responses and ensure alignment with reality.
AI can break down complex decision-making processes and use optimization algorithms to recommend the best solutions based on multiple variables. This shortens the time spent on pilots and reduces noise from uncontrolled factors that often distort traditional A/B testing in real operations.
But none of this matters if we can’t embed the improvements into daily operations. That’s where training and inference come in.
Control: From Training to Inference — Activating the Intelligent Twin
Once improvement scenarios have been tested, it’s time to make them operational — and that’s where the digital twin enters its most powerful phase: inference.
Training Phase: Building and Calibrating the Digital Twin
In this phase, the digital twin is trained using historical data, real-time inputs, and subject matter expertise. It learns how the process behaves across scenarios — normal operations, disruptions, influencing variables, and decisions made under different conditions.
Training uses machine learning methods (supervised, unsupervised, or reinforcement learning) and AI often combining numerical data with qualitative insights. Business experts guide this process to ensure that the model truly reflects operational reality.
Inference Phase: Real-Time Intelligence and Proactive Decisions
Once trained, the model transitions to the inference phase — where it begins using live data to make predictions and support decisions in real time.
Inference is the moment the model acts. It applies its trained logic to:
• Predict outcome based on new input
• Detect deviations and risks before they escalate
• Recommend corrective actions
• Trigger alerts or automated workflows
• Simulate future scenarios and outcomes
For instance:
• A manufacturing twin might detect an early-stage defect risk based on vibration and temperature data.
• A procurement twin could forecast delays due to approval bottlenecks and recommend alternate workflows.
• A customer experience twin could adapt responses in real time based on sentiment analysis during an interaction.
Inference turns the digital twin into a real-time co-pilot — one that not only observes, but adapts, reacts, and continuously improves.
It’s lightweight, optimized for speed, and often deployed at the edge — inside devices or close to the process — for maximum responsiveness.
Nevertheless, AI model relies strongly on historical data, if an unseen event happens, even highly sophisticated models can poorly perform and give a bad « advice »
Control (continued)
Automated dashboards, alerts, and closed-loop feedback systems enable consistent performance without the need for constant human intervention. But adoption is still key. Even the best AI-powered solution must be embraced and trusted by teams. That’s why communication, engagement, and change management remain essential roles for Lean Six Sigma leaders and Black Belts.
Conclusion
AI’s integration into DMAIC represents a game-changing evolution in how we improve processes. It empowers teams to:
• Analyze more deeply and quickly
• Simulate and optimize with digital twins
• Implement improvements proactively
• Sustain gains through real-time monitoring and adaptive inference
But success depends on people and adoption. As always, the role of human leadership is to guide, validate, and engage — even if the tools now make the job lighter.
What’s Next
In the coming articles, we’ll dig deeper into each DMAIC phase, sharing real-world use cases, key pitfalls to avoid, and best practices for:
• Creating digital twins of business processes
• Managing data collection and model training
• Selecting the right statistical methods for AI
• Understanding training vs. inference in operational AI
• Mapping required skills and competencies
• Defining the IT infrastructure and security needed to scale
• Making AI-driven improvement affordable and accessible for SMEs, schools, universities, and public administrations
AI doesn’t just automate — it accelerates improvement. And those who harness it will be the ones shaping the future of performance and quality.
#ArtificialIntelligence #DigitalTransformation #ProcessImprovement #DigitalTwin #AgenticAI #FutureOfWork#SmartAutomation
