Process Optimization Frameworks: How to Choose and Apply One

Process Optimization Frameworks: How to Choose and Apply One

The best process optimization framework is the one that matches the problem you actually have. Use Lean or value stream mapping when work moves slowly and waits a lot. Use Six Sigma’s DMAIC when results vary and you need data to find out why. Use PDCA or Kaizen when you want steady, team-owned improvement. Most mature organizations combine two of these, and this guide shows how to pick, sequence and apply them.

TL;DR

• Lean attacks waste and slow flow. Six Sigma attacks variation and defects. Combining them is the most common recommendation in public-sector improvement guidance.
• Value stream mapping is the best first step when you do not yet know where time is lost. It works for software delivery as well as factories.
• Pick one primary KPI per process and a handful of supporting indicators, not a dashboard of thirty numbers.
• Fix and standardize the process before you automate it. Automating a broken workflow just makes the breakage faster.
• Start with one process, run a 90-day cycle, and let the result decide what comes next.

Table of Contents

• Framework, method, model: what the words mean
• The main frameworks side by side
• Lean vs Six Sigma: which one do you need?
• DMAIC: the five phases in practice
• Value stream mapping: where most projects should start
• PDCA, Kaizen and A3: the lightweight options
• Use a common process language before you benchmark
• How to choose: five questions
• A 90-day process improvement roadmap
• Where automation and AI fit
• Common mistakes
• Our perspective
• Turning your process map into working systems
• FAQ
• Sources

Framework, method, model: what the words mean

People use “framework,” “methodology,” “method” and “model” interchangeably, and search results do too. In practice there are three layers worth separating:

1. A philosophy tells you what to care about. Lean (eliminate waste, improve flow) and Six Sigma (reduce variation) are philosophies.
2. A framework or cycle tells you the order of work. DMAIC, PDCA and A3 are cycles.
3. A technique is a tool inside a cycle. Value stream mapping, SIPOC diagrams, root cause analysis and process mining are techniques.

Much confusion about which framework to choose disappears once you see that you usually need one of each layer: a philosophy for direction, a cycle for discipline, and a technique for evidence.

The main frameworks side by side

None of these is universally better. They answer different questions, which is why the comparison above is organized by question rather than by popularity.

Lean vs Six Sigma: which one do you need?

This is the comparison readers ask about most. The short version: Lean is about speed and flow, Six Sigma is about consistency.

NHS improvement guidance describes Lean as an approach that improves flow in the value stream and eliminates waste, while Six Sigma uses the DMAIC framework and statistical tools to uncover root causes of variation. The same document argues the two work best together. Lean clears the clutter so the process can be seen; Six Sigma then measures what is left.

ASQ describes the aim of Six Sigma as improving customer satisfaction by reducing and eliminating process variation that causes defects. That makes it a poor fit for a process with ten unnecessary approvals and no data, and an excellent fit for a process that mostly works but produces unpredictable errors.

A quick test:

• “It takes too long.” Start with Lean and a value stream map.
• “It’s wrong too often.” Start with DMAIC.
• “It’s both, and nobody agrees on why.” Map first, then measure, then fix. That is Lean Six Sigma in practice.

Teams under about twenty people rarely need the full statistical toolkit. Black-belt-level analysis is a poor use of time when you can read the whole process in an afternoon and run a PDCA experiment on Monday.

DMAIC: the five phases in practice

DMAIC is the backbone of Six Sigma and also works on its own. It has five phases: Define, Measure, Analyze, Improve and Control. Here is what each looks like in a business process, using invoice approval as the example.

1. Define. State the problem in one sentence with a number: “Invoices take a median of 11 days to approve, against a 5-day target.” Name the process owner and the boundary of the process.
2. Measure. Collect baseline data for a few weeks. Count cycle time, rework rate and queue time per step. If the data does not exist, building that measurement is itself a result.
3. Analyze. Find the causes with evidence. Split the data by invoice type, approver or vendor. Often a small number of exceptions cause most of the delay.
4. Improve. Test fixes on a small scale before rolling them out: a different routing rule, a threshold under which no approval is needed, a required field.
5. Control. Make the gain stick with a standard procedure, a monitored metric and a named owner. Without this phase, processes drift back within months.

The Control phase is the one teams skip, and it is the one that decides whether the project paid off a year later.

Value stream mapping: where most projects should start

If you do not know where the problem is, map before you measure anything else. Value stream mapping (VSM) draws every step from request to delivery and records how long work spends being processed versus waiting.

A software delivery guide from Harness summarizes the practice in four steps: map the current state with all stakeholders, analyze it for waste, design a future state without the non-value-adding steps, and implement improvements iteratively. It also notes that a map tracks information flow, material flow and time, so total lead time becomes visible at each step.

Value stream mapping for software development and DevOps

The technique came from manufacturing, but software teams get a lot out of it. The “material” is a code change or a feature request, and the value stream runs from idea to running in production. A typical map includes:

• Request intake and prioritization
• Design and refinement
• Development
• Code review
• Testing and QA
• Release approval
• Deployment and verification

For each step, record two numbers: process time (hands-on work) and wait time (the item sitting in a queue). Teams are usually surprised. A feature that needs two days of real effort may spend three weeks waiting for review, a test environment or a release window. That gap between process time and total lead time is where the opportunity sits.

For DevOps teams, the most common findings are slow review queues, manual environment provisioning, large batch releases and approval gates that add delay without reducing risk. Each points to a specific fix: review-time agreements, infrastructure as code, smaller releases, automated checks that replace manual sign-off.

A practical VSM session

Book two hours with the people who actually do the work, not only their managers. Walk the process on a whiteboard from the customer end backward. Use real recent examples, not the official procedure. Mark every wait, every handoff and every loop back for rework. End the session with three candidate improvements ranked by effort and expected impact. Then pick one.

PDCA, Kaizen and A3: the lightweight options

Not every improvement needs a project charter. PDCA (Plan, Do, Check, Act) is a four-step loop for running a small experiment: define the change and the expected result, try it on a limited scale, compare the result with the expectation, then adopt it or adjust. Kaizen is the habit of running those loops constantly, with the people closest to the work proposing the changes. A3 is a discipline of fitting the problem, analysis, countermeasures and follow-up on one sheet of paper so a decision can be reviewed quickly.

These suit teams that want momentum without heavy analysis. Their weakness is scope: they tend to polish individual steps and can miss a structural problem between departments. That is where a value stream map earns its place beside them.

Use a common process language before you benchmark

The frameworks above tell you how to improve. They do not tell you whether your numbers are good. For that you need benchmarks, and benchmarks need comparable definitions.

APQC recommends a standardized process framework such as its Process Classification Framework so that departments share a common language before optimization begins. This prevents comparing activities that only look alike. APQC also runs a recurring survey, and its 2025 findings on how organizations apply process frameworks are a useful reality check on adoption and the challenges involved.

Two cautions from APQC’s guidance on picking process measures and benchmarking:

• Do not clutter the dashboard. A workable limit is one primary KPI per process, supported by no more than seven secondary indicators.
• Baseline internally before comparing externally. Skipping the internal step often leads to goals that cannot be reached.

How to choose: five questions

Use these in order. The first one that gives a clear answer usually decides the starting point.

1. Do you know where the problem is? If not, map the value stream.
2. Is the issue speed or quality? Speed points to Lean; inconsistent quality points to DMAIC.
3. Do you have data? If measurements do not exist, start with PDCA experiments and collect data as you go. Do not commission a statistical analysis of numbers you do not have.
4. Who owns the process? A process that crosses three departments needs an executive sponsor, whichever framework you pick. Frameworks do not fix unclear ownership.
5. How much change can the team absorb? Heavy frameworks fail in teams that are already stretched. A one-page A3 beats an abandoned six-month program.

A 90-day process improvement roadmap

A roadmap needs to be short enough that people believe in it. This one fits a single process in a quarter.

Days 1 to 15: Scope and baseline. Choose one process with a clear owner and a visible cost, such as order-to-cash, onboarding or release management. Write the problem statement with a number. Run the value stream mapping session. Pick the primary KPI.

Days 16 to 40: Measure and diagnose. Collect data on cycle time, wait time and rework. Find the two or three causes that account for most of the loss. Skip causes that are real but small.

Days 41 to 70: Test fixes. Run PDCA experiments on the top causes. Change one thing at a time where you can, so you know what worked. Retire fixes that do not move the KPI.

Days 71 to 90: Standardize and hand over. Document the new process, train the people who run it, set up a monthly review of the KPI, and name the owner. Then decide whether to automate any step that has become stable.

The last step is where technology belongs. Automating before this point locks in whatever inefficiency remains.

Where automation and AI fit

Automation is not a framework, but it is how most optimization gains get locked in. Once a step is standardized and measured, it becomes a candidate: routing, data entry, approvals under a threshold, status notifications, reconciliation.

Process mining, which reconstructs real process flows from system event logs, is a growing bridge between the two. Analysts regularly evaluate vendors in this area, and Gartner’s process mining leaders are typically chosen by primary focus: enterprise-wide analytics, strong fit with an existing ERP landscape, or rapid automation led by robotic process automation. When you shortlist tools, matching the tool to your own systems matters more than the overall ranking. An independent directory of hands-on AI tool reviews and benchmarks can help you compare options by use case before you commit to a trial.

If AI will make or support decisions inside the optimized process, add governance from the start. Frameworks like NIST’s AI Risk Management Framework are often misread as a linear sequence. As one explainer on its four functions points out, Govern is cross-cutting and continuous, not a first stage you finish and leave behind. Our own guides cover this in more depth: mapping enterprise AI governance to NIST, ISO and the EU AI Act and a five-stage AI risk assessment.

Common mistakes

• Starting with the tool. Buying software before you have mapped the process usually digitizes the confusion.
• Optimizing a step instead of the flow. Speeding up one department can just move the queue to the next.
• No baseline. Without a before number, you cannot prove an after number, and sponsors stop funding the work.
• Too many metrics. See the APQC guidance above. A focused KPI gets managed; twenty KPIs get ignored.
• Treating it as a project. A project ends. A process needs an owner who keeps measuring after the consultants, or the task force, leave.

Our perspective

We build custom software and AI automation for a living, so we see the same pattern in nearly every engagement: the technology is rarely the hard part. The hard part is agreeing on what the process really is. Teams describe the procedure on paper, but the work follows a different path, full of workarounds, spreadsheets and side conversations.

That is why we begin with mapping and measurement, and why we treat AI as part of the process design rather than something added at the end. If a decision step can be supported by a model, it should be designed that way from the first map, with the data, the exception handling and the human review points defined. If it is bolted on afterward, it tends to sit unused.

Our advice is simple. Pick one process. Map it honestly. Choose one KPI. Fix the biggest cause, standardize, and only then automate. Do that three times and you will have a framework of your own that fits your business better than any textbook version.

— Matija

Turning your process map into working systems

When a process has been mapped, measured and stabilized, the next question is how to build it into your systems. We help companies do that through process optimization and workflow automation. Smaller, targeted automations start from 3000 EUR one-off, and you can see the scope on our AI automation page. If you want to test an idea first, proof-of-concept development starts from 5000 EUR one-off, described on our AI services page. For a look at what this looks like in practice, read our bakery route optimization case study, or our process automation guide for mid-sized enterprises.

FAQ

What is the difference between a process optimization framework and a methodology?

In everyday use, the terms overlap. A methodology is the broader philosophy and set of principles, such as Lean or Six Sigma. A framework is the structured sequence you follow, such as DMAIC or PDCA. You will usually use one of each: a philosophy for direction and a cycle for execution.

Should I use Lean, Six Sigma or both?

Use Lean when the problem is waiting, handoffs and long lead times. Use Six Sigma when the problem is defects and inconsistent results. If you have both problems, NHS improvement guidance recommends combining them, with Lean improving flow and DMAIC reducing variation.

How long does value stream mapping take?

The mapping session itself usually takes a few hours for one process, provided the right people are in the room. Collecting real wait-time and process-time data takes longer, often a few weeks. Start with a rough map and refine it as data arrives, rather than waiting for a perfect one.

Can small companies use these frameworks?

Yes, in lighter form. A small team can run a value stream mapping session, pick one KPI and test changes with PDCA without any certification or special software. The full statistical toolkit of Six Sigma is rarely needed below a certain scale of data.

When should I automate a process?

Automate after the process is mapped, simplified and stable, and when a step is repetitive, rule-based and high-volume. Automating earlier tends to hard-code workarounds. If you are unsure, test the idea on one step first.

Sources

• Lean Six Sigma: some basic concepts, NHS England
• Six Sigma definition, ASQ
• DMAIC: definition, five phases and tools
• Value stream mapping for software delivery, Harness
• APQC: benchmarking data and process frameworks
• APQC: Best practices in applying process frameworks 2025
• APQC: Picking process measures and benchmarking
• Gartner names leaders in process mining, Automation Today
• NIST AI Risk Management Framework: the four functions explained

Recommended

• Process Automation Guide for Mid-Sized Enterprises 2026
• Business Process Optimization | Workflow Automation
• Bakery Route Optimization | Case Study
• Streamline your Industry 4.0 implementation workflow