Lean and Six Sigma have shaped operational excellence for decades. Their core ideas (understand value from the customer’s perspective, eliminate waste, reduce variation, standardize and improve continuously) are as relevant as ever. What has often limited them is speed. Data collection takes weeks, analysis depends on scarce experts, control plans decay after projects end, and improvement cycles run quarterly when markets move weekly.
Artificial intelligence changes those constraints. Machine learning can monitor processes continuously, generative AI can summarize thousands of comments and incident reports in minutes, and AI agents can carry out routine follow-up actions. But AI also brings a new risk: automating a broken process simply produces bad outcomes faster. This playbook explains how to combine Lean discipline with AI capabilities, where AI adds most value across the improvement cycle, and how to start without falling into common traps.
Why Lean and AI Need Each Other
There is a tempting narrative that AI makes traditional improvement methods obsolete: why map a value stream when an algorithm can optimize the process? The evidence points the other way. McKinsey’s State of AI in 2026 found that nearly three-quarters of AI high performers have fundamentally redesigned workflows around AI, compared with about a quarter of other organizations. Gartner’s advice on agentic AI similarly stresses redesigning workflows rather than layering agents onto existing processes. Workflow redesign is exactly what Lean practitioners are trained to do.
In a 2026 series for Modern Distribution Management, Bart Tessel of the National Association of Wholesaler-Distributors argued that Lean Six Sigma was not wrong but early: its disciplines were sound, but organizations could not run improvement cycles fast enough to sustain them. AI removes many of those constraints. At the same time, he warned that AI does not fix unclear processes; it accelerates them, and he recommended the classic Lean sequence of stabilize, standardize, improve and only then automate.
The combination works because each side covers the other’s weaknesses:
- Lean provides problem definition, customer focus, root-cause thinking and standard work, which keep AI efforts targeted at real value
- AI provides speed, scale and continuous sensing, which keep improvement efforts from stalling and control plans from decaying
- Lean’s respect for frontline knowledge helps AI tools get adopted, while AI reduces the administrative burden that often makes improvement programs feel like extra work
AI does not fix a broken process; it runs it faster. Stabilize, standardize and improve first, then automate, and use AI to keep improving long after the project ends.
Where AI Fits in DMAIC
DMAIC (Define, Measure, Analyze, Improve, Control) remains a useful structure for AI-augmented improvement. AI contributes differently at each stage.
Define
Generative AI can analyze unstructured voice-of-customer data, such as complaints, survey comments, call transcripts and support tickets, to identify recurring issues and quantify how often they occur. This helps teams choose problems based on evidence rather than the loudest voice in the room, and draft clearer problem statements and project charters.
Measure
Process mining and process intelligence reconstruct how processes actually run from system event data, replacing much of the manual time-and-motion work traditionally needed to build a current-state value stream map. Sensors and connected equipment provide continuous measurement in manufacturing and logistics. AI can also flag data quality issues early, before they undermine the analysis.
Analyze
Machine learning can test many potential drivers of a defect or delay at once, surfacing patterns that fishbone diagrams and manual stratification may miss. Generative AI can help teams explore hypotheses and summarize findings, but statistical rigor still matters: correlation found by a model is a hypothesis to test, not a root cause.
Improve
Simulation and digital twins allow teams to test countermeasures virtually before changing the real process, from staffing patterns to scheduling rules. AI can also generate standard work drafts, training materials and checklists from the agreed future state, speeding up rollout.
Control
This is where AI may add the most value. Traditional control plans depend on periodic reviews and charts that people stop looking at once the project team moves on. AI-driven monitoring can track process behavior continuously, detect drift and trigger alerts or actions, turning control from a document into a living system.
From Projects to Continuous Intelligence
Lean has always aimed for continuous improvement, but in practice many organizations run improvement as a series of projects. AI makes it realistic to run a daily improvement loop at scale. Tessel describes a five-step loop that is a useful template:
- Sense: detect deviations early, such as process drift, forecast bias or rising rework.
- Diagnose: identify likely root causes and patterns rather than treating each exception separately.
- Simulate: test potential countermeasures before changing the business.
- Act: implement changes with guardrails and clear escalation paths.
- Recalibrate: update thresholds, models and standards as the process improves.
This loop maps naturally onto Lean daily management. Tier meetings at the frontline, supervisor and management levels can use AI-generated summaries of what changed overnight, which issues are emerging and which countermeasures worked. The human role shifts from compiling data to deciding and coaching, which is where experienced practitioners add the most value.
Keeping humans accountable
AI can recommend and even execute actions, but accountability for outcomes must remain with process owners. Make it explicit which decisions AI may take on its own (for example, re-prioritizing a work queue), which require approval (changing a pricing rule), and which remain purely human (changing a policy or a customer commitment).
The Eight Wastes, Revisited With AI
Lean’s classic wastes are a practical lens for spotting AI opportunities, and for spotting new wastes that AI can create.
- Waiting: AI can predict bottlenecks and re-sequence work, but it can also create waiting if approvals for AI outputs pile up.
- Overprocessing: AI-generated reports and summaries no one reads are a new form of overprocessing; produce only what drives decisions.
- Defects: machine learning can detect defects earlier, from visual inspection to invoice errors, while AI hallucinations introduce a new defect type that needs controls.
- Motion and transportation: in knowledge work, switching between systems is the equivalent; assistants that bring information into one place reduce it.
- Inventory: better demand forecasting reduces physical inventory, and AI triage can reduce backlogs of tickets or cases.
- Overproduction: generating content, code or analysis because it is cheap, not because it is needed, is a growing risk.
- Unused talent: AI should free experts for problem-solving and coaching, not leave them reviewing machine output all day.
Knowledge work deserves particular attention, because that is where generative AI is spreading fastest. In finance, HR, IT and customer service, waste often hides in handoffs, approvals, rework caused by incomplete requests and time spent searching for information. AI assistants can reduce some of this, for example by checking that a request is complete before it enters a queue or by retrieving the right policy at the point of need. But the bigger gains usually come from redesigning the flow itself: removing unnecessary approvals, combining steps and clarifying who decides, before deciding which remaining steps AI should support.
A useful exercise is to run a waste walk on a process that already uses AI. Many teams discover that the AI has removed some waste but added new queues, rework and reports.
A Step-by-Step Playbook
Organizations that want to combine Lean and AI can follow a staged approach.
- Pick a value stream, not a tool. Choose an end-to-end process with clear customer impact and economic value, such as order fulfillment, claims, maintenance or month-end close.
- Map the current state with data. Combine process mining with gemba walks. The data shows what happens; observation explains why.
- Stabilize and standardize. Remove obvious waste, clarify roles and define standard work before introducing AI. Measure the baseline.
- Identify AI-suitable steps. Look for high-volume, rules-plus-judgment tasks, information gathering, classification, forecasting and monitoring. Avoid steps where the underlying policy is unclear.
- Match the tool to the task. Use conventional automation for routine, deterministic steps; predictive models for forecasting and anomaly detection; generative AI for unstructured information; and agents only where multi-step decisions add value, consistent with Gartner’s advice to match the tool to the need.
- Pilot with a control group. Compare AI-supported work with the standard process, measuring quality, time, cost and employee experience.
- Build AI into the control plan. Define monitoring, thresholds, escalation and ownership for both the process and the AI components.
- Scale and teach. Capture what worked as standard work, train belts and frontline teams in AI-assisted methods, and move to the next value stream.
Building the Capability: Roles and Skills
AI changes the skills an operational excellence function needs. Black Belts and Lean coaches remain essential, but their toolkit expands.
- Improvement leaders need enough AI literacy to frame suitable use cases, question model outputs and recognize when a problem is about process design rather than technology.
- Data and AI specialists embedded in improvement teams build models, process mining views and monitoring, working alongside process experts rather than in isolation.
- Frontline teams need training on how AI tools fit their standard work, how to flag errors and how their feedback improves the system.
- Process owners must own AI-enabled processes end to end, including the performance of the AI components within them.
Belt curricula are beginning to include AI-assisted analysis, process mining and AI governance. Organizations should update their certification paths so that new practitioners learn to use these tools with the same rigor they apply to statistical methods.
Data Foundations and Governance for AI-Driven Improvement
AI-augmented improvement depends on data that many operations teams have historically treated as a by-product. Event logs, sensor feeds, quality records, maintenance histories and customer feedback need to be accessible, reasonably clean and consistently defined. A few foundations make the difference between pilots that stall and programs that scale:
- Agreed definitions for core operational metrics, such as on-time delivery, first-pass yield or cycle time, so that AI models and dashboards use the same numbers as management reviews
- Reliable event and timestamp data from core systems, which process mining and monitoring depend on
- A governed platform where operational data is landed, joined and made available to analytics and AI tools, instead of extracts held in individual spreadsheets
- Clear ownership for data quality in each process, with issues fixed at source rather than patched downstream
Governance must also cover the AI itself. Models that influence operational decisions should have owners, documented purpose, performance monitoring and a way for frontline staff to report errors. Generative AI tools used to summarize issues or draft standard work need review steps, because a confident but wrong summary can send a team chasing the wrong cause. Where AI monitors individual work, such as productivity or adherence to standard work, be transparent with employees about what is measured and why, and involve employee representatives where required. Lean’s principle of respect for people applies to AI deployments as much as to any other change.
A Worked Example: Maintenance in a Manufacturing Plant
Consider a generic mid-sized manufacturer struggling with unplanned downtime on a critical production line. A traditional Lean project would map the maintenance process, analyze failure history and introduce standard work for preventive maintenance. The team does that first, discovering that maintenance requests are logged inconsistently, spare parts are often unavailable and technicians spend significant time searching for equipment history.
After standardizing work order logging and spare parts processes, the team introduces AI in three targeted ways. A predictive model uses sensor data to flag early signs of failure on the most critical assets. A generative AI assistant gives technicians instant access to equipment manuals, past work orders and troubleshooting steps on a tablet. And the control plan is automated: downtime, mean time between failures and response times are monitored daily, with alerts when trends deteriorate.
Daily tier meetings now start with an AI-generated summary of asset health and open issues, freeing time for problem-solving. The improvement team tracks results against the pre-project baseline and uses the same approach on the next line. Importantly, the predictive model is reviewed regularly with technicians, whose feedback on false alarms improves it over time.
The sequence matters as much as the technology. Had the plant introduced predictive maintenance before fixing work order logging and spare parts availability, alerts would have arrived with no reliable way to act on them, and the model would have been trained on inconsistent failure records. By stabilizing first, the team ensured that AI amplified a working process rather than an unreliable one.
Pitfalls, Outlook and Conclusion
Pitfalls to avoid
- Automating before stabilizing, which locks in waste and variation
- Treating AI outputs as facts rather than hypotheses to verify, especially in root-cause analysis
- Building dashboards and alerts without clear owners or response standards
- Excluding frontline teams, which undermines trust and loses their knowledge
- Measuring AI activity (models built, prompts run) instead of process outcomes
Outlook
Through 2027, expect operational excellence functions to adopt process intelligence, AI-driven daily management and bounded AI agents for routine improvement actions. The organizations that benefit most will be those that keep Lean’s fundamentals at the center: customer value, respect for people, standard work and scientific problem-solving.
Lean and AI are not competing philosophies. Lean tells you what to improve and how to sustain it; AI helps you see faster, analyze deeper and act more continuously. Used together with discipline, they can turn continuous improvement from an aspiration into a daily operating rhythm.
How Hanoi Technology Can Help
Hanoi Technology combines process analysis, data engineering and AI delivery to help operations teams build AI-augmented continuous improvement programs.
Further Reading
- Lean Six Sigma Was Early in Distribution. AI Makes It Real (Part 1 of 4) – Modern Distribution Management
- The state of AI in 2026: On the road to ROI – McKinsey & Company
- Gartner: More than 40% of agentic AI projects will be canceled in the next few years – SD Times
- The Future of Lean Six Sigma: How Automation, AI and Synthetic Agents Are Redefining Process Excellence – Purdue TAP