Process Improvement · AI · Lean Six Sigma
What happens when you apply AI to structured process improvement?
This site is a working exploration of that question. I'm a Lean Six Sigma practitioner and operations leader, and I've been building tools that use AI to do the analytical heavy lifting inside a DMAIC framework — from mapping processes to baselining performance to surfacing what customers actually care about.
How this works
The approach: DMAIC, accelerated with AI.
Start with the method
DMAIC — Define, Measure, Analyse, Improve, Control — is a rigorous framework for process improvement. Everything on this site sits inside that structure. AI doesn’t replace the method; it does the analytical work faster.
Process before automation
A poorly understood process automated quickly is just a faster mess. The tools here are built around understanding the process first — mapping it, measuring it, finding where it’s failing customers.
Compress the cycle
Traditional improvement projects take months. The question I’ve been exploring is how much of that time — the analysis, the synthesis, the documentation — AI can absorb without losing rigour.
Working examples, not slides
Each section below is a live, interactive example of one part of the DMAIC cycle. Enter a real industry and see what the AI produces. The tools are imperfect but they’re real.
DMAIC · Define Phase · Process Mapping
Worked example: mapping a process with AI.
In the Define phase of DMAIC, one of the first tasks is mapping the process as it actually runs — not as the procedure document says it should. Describe any process below and the AI will produce a BPMN 2.0-compliant diagram, identify handoffs and potential bottlenecks, and export a file you can import into Signavio.
Describe a process
Generates a BPMN 2.0 compliant diagram. Leave blank to use the example above.
Your BPMN 2.0 diagram will appear here
Describe your process and click Generate. The AI maps each step to a correct BPMN 2.0 element type.
BPMN 2.0 · ISO/IEC 19510:2013 · Signavio-compatible export
AI Insights
After generating a diagram, AI-identified bottlenecks, handoffs, and automation opportunities will appear here.
DMAIC · Define Phase · Voice of Customer
Worked example: VOC to CTQ.
One of the most common problems in improvement projects is that the targets get set internally — by operations, finance, or management — rather than derived from what customers actually said. VOC-to-CTQ is the technique that fixes that. Here’s how it works, with AI doing the analytical work.
Gather the verbatim
The starting point is always the raw, unfiltered language customers use — survey responses, support tickets, NPS comments, interview transcripts. The verbatim matters because the moment you paraphrase it, you start substituting your interpretation for their reality.
Cluster into themes
Individual pieces of feedback are anecdotes. Grouped, they become signal. Clustering surfaces the recurring needs behind the noise — the things that come up again and again, regardless of how the customer happened to phrase them.
Derive a measurable CTQ
A Critical to Quality characteristic converts a customer need into something your operation can actually be measured against. It gives the improvement project a specific, numeric target that came from the customer — not from an internal benchmark someone invented.
What is a Critical to Quality characteristic?
A CTQ is the translation layer between what a customer says and what your operation can be measured against. It takes a qualitative complaint — “delivery takes too long” — and converts it into a specific, numeric requirement: “Order-to-delivery time < 48 hours for 95% of orders.” Without that translation, improvement targets are guesses. With it, you have something the customer would recognise as meaningful.
Try it with any industry
DMAIC · Measure Phase · Process Baseline
Worked example: baselining process performance.
The Measure phase is where a lot of improvement projects stall — getting clean, consistent data is harder than it sounds. This example shows what a baseline looks like once you have it: a control chart plotted against the CTQ target, with special causes flagged and the gap to requirement made explicit.
Measure what's actually happening
Most organisations have a sense of how they're performing. Far fewer have a time-series of actual measurements against a specific CTQ. The Measure phase of DMAIC is about replacing that sense with data — collected consistently, over time, in the same way.
Separate noise from real problems
Not all variation is equal. Statistical Process Control distinguishes common cause variation — the ordinary day-to-day noise of the process — from special cause events that are statistically unlikely to have occurred by chance. Only special causes point to something specific worth investigating.
See the gap clearly
Plotting the baseline against the CTQ target line makes the gap between current performance and customer requirement visible in a way that no average or summary statistic can. It's the starting point for every improvement conversation: this is where we are, this is where we need to be.
Reading a control chart
The Upper and Lower Control Limits (UCL and LCL) are set at three standard deviations from the process mean. This isn't an arbitrary choice — it reflects the statistical likelihood of a point occurring by chance. A point outside those limits has less than a 0.3% probability of being random variation. That's the definition of a special cause: something specific happened, and it's worth finding out what. Points inside the limits are common causevariation — they won't be fixed by one-off interventions, only by changing the process itself. The green target line is the CTQ requirement derived from customer feedback. Everything above it (for a lower-is-better metric) is a failure to meet the customer's need.
Try it with any industry
Your control chart will appear here
Enter an industry above to generate a realistic 18-month baseline with control limits, target line, and AI analysis.
It's not all about AI
Sometimes the best automation is just a form that actually works.
The travel company project taught me something important: before reaching for AI or complex integrations, there's massive value in simply understanding your workflow and building a single system that captures it properly.
Study the workflow first
Before touching any technology, map the actual process as it runs — not the documented version, the real one. Who touches what data? Where does information sit while waiting? What are the handoffs? Most operational friction lives in these gaps.
Identify minimal inputs and outputs
Every step in a workflow needs certain information to start, and produces certain information when done. Most businesses haven't clearly defined these. Once you know them, you can build a form that captures exactly what's needed — no more, no less.
Status as the workflow engine
A booking isn't just a record — it's a thing moving through stages. Lead → Itinerary Drafted → Quote Sent → Deposit Paid → Confirmed. Each status change represents a decision point and a handoff. When your system tracks this visibly, everyone knows what's happening without asking.
Via Francigena Tours — CMS & Business Admin Platform
A walking tour operator managing the full lifecycle of customer bookings. Before this system, operations ran across disconnected tools — spreadsheets, emails, PDFs, and memory.
Every booking follows a defined status lifecycle: Lead → Itinerary Drafted → Estimate Ready → Quote Sent → Deposit Paid → Confirmed. At each stage, the system prompts for exactly the information needed — customer details, walking stages, hotel preferences, transfer requirements. No more hunting through email threads for missing details.
The dashboard shows every booking's current status, what's blocking it, and what comes next. Hotel inquiries pending responses. Quotes awaiting customer acceptance. Planning fees unpaid. Everyone on the team knows what's happening without asking each other.
The impact: The majority of manual data entry and cross-system rekeying was eliminated. The operations team got real-time visibility into booking status, financial position, and outstanding actions — without having to check four different tools to piece it together.
This wasn't AI doing the work. It was simply a clear understanding of the workflow, a database that modelled it properly, and forms that captured the right information at each step. The “automation” was just the system knowing what comes next.
What the system includes
The lesson
There's a tendency to think operational improvement requires cutting-edge AI or complex automation. Often, it doesn't. The travel business saw dramatic improvement from simply having one coherent system instead of fragmented tools. Forms, statuses, and a database — basic building blocks applied thoughtfully — can be transformative. The AI features (email analysis, hotel reply detection) came later, layered on top of solid foundations. The workflow understanding came first.
Things I've noticed
Some observations from doing this work.
Not a framework. Just patterns that have come up repeatedly across different industries and different types of problems.
The process has to come before the automation
The most common way automation projects fail is that they automate a process nobody fully understood. You end up with the same confusion, just faster. Mapping it first — properly, with all the handoffs and exceptions visible — changes what you build.
AI is genuinely useful at the analytical steps
There are parts of a DMAIC project where the work is mostly structural — clustering feedback, generating process maps, formatting statistical output. That’s exactly where AI earns its keep. The judgment calls still need a human; the scaffolding doesn’t.
Small, focused systems outperform large ones
Every large transformation programme I’ve seen has moved slowly and delivered less than promised. The more interesting pattern is smaller systems that solve one problem precisely — they ship faster, break less, and the results are easier to measure.
The gap between insight and action is where value disappears
Operational teams rarely lack information about what’s wrong. They lack systems that turn that information into action quickly enough to matter. Closing that gap — making the next step obvious and easy — is where the real leverage sits.
Thought leadership
Ideas worth sharing.
AI-Accelerated DMAIC: Compressing the Improvement Cycle
Traditional DMAIC projects take months. AI can compress the analysis and synthesis phases from weeks to hours — without sacrificing the rigour that makes improvements stick.
Process Understanding Before Automation: Why Most Automation Projects Fail
Organisations rush to automate without first understanding why their processes work the way they do. The result is automated chaos. Here's the discipline that prevents it.
The Intelligent Operations Model: From Reactive to Self-Improving
Most operations teams spend their time reacting to problems that already happened. An intelligent operations model — built on continuous measurement, AI-assisted analysis, and modular automation — shifts that dynamic fundamentally.
Let's talk
If any of this is interesting to you, I'd genuinely like to hear from you.
This site is a working exploration, not a polished product. If you found something useful here, want to push back on something, or are doing similar work and want to compare notes — drop me a message.
I'm also open to conversations about roles, collaborations, or problems you're trying to solve — but no pressure on any of that. Curiosity is enough of a reason.