I have this problem. Which tool do I use?

Have I reached a cause I can actually act on?

Follow a chain of causes and test each link with the reverse read: does the cause actually produce the effect above it?

• —A problem keeps coming back after being fixed
• —Your root cause analysis ended at someone’s behaviour
• —You need to trace one specific failure back to its origin

5 Whys Chain · analyze · White Belt

The area looks tidy — so why does it keep going back?

Score a workspace against each of the five stages with an observation attached to every score, and see when the score is measuring your audit routine rather than the workplace.

• —A 5S programme scores well and the workplace keeps drifting
• —You need to know which of the five stages to work on next
• —Audit scores are being compared between areas or sites

5S Audit · control · White Belt

What could be causing this, other than the thing I already suspect?

Generate possible causes across six categories, mark which ones you have actually checked, and see when the diagram is telling you more about the room than about the problem.

• —A team has one favourite explanation and no evidence for it
• —You need to widen the search before choosing what to measure
• —The same problem keeps returning after each fix

Cause & Effect (Fishbone) · analyze · White Belt

These groups look different — are they, and which ones?

Compare three or more groups at once — with the pairwise comparisons on the same screen, because a significant F never says which group differs.

• —Three or more machines, suppliers, shifts or settings to compare at once
• —Somebody has run every pair as a separate t-test and found something
• —You need to know which groups differ, not just that some do

Comparing Several Groups (ANOVA) · analyze · Green Belt

Did something actually change, or is this how the process always behaves?

Build an I-MR, X-bar R, X-bar s, p, np, c or u chart with Western Electric or Nelson rules, and read the dispersion chart first — because the other chart's limits come from it.

• —You need to know whether a shift in a metric is real before acting on it
• —You are setting up ongoing monitoring rather than a one-off analysis
• —Everything is inside specification and problems keep reaching the customer

Control Chart · control · Yellow Belt

Which of these factors actually move the process — and do any of them depend on each other?

Run every combination of two-level factors and estimate every effect at once — including the interactions that changing one factor at a time is structurally unable to see.

• —Several candidate factors and no proof which ones matter
• —A suspicion that the right setting for one factor depends on another
• —Before a one-factor-at-a-time trial spends its runs finding a false optimum

Design of Experiments (2ᵏ Factorial) · improve · Green Belt

I have more factors than runs — which design, and what will it fail to tell me?

Run a fraction of the full experiment and see exactly what it costs — the alias structure is on screen before a single run is booked.

• —More candidate factors than a full factorial's run budget allows
• —Screening many inputs down to the few worth a real experiment
• —Before agreeing to a design somebody else chose, to see what it confounds

Fractional Factorial (Screening Design) · improve · Black Belt

Am I measuring the process, or measuring my gauge?

Find out how much of the variation you are looking at is the measurement system rather than the process, by the crossed ANOVA method.

• —Two people measure the same part and get different answers
• —Before trusting a capability study or a control chart built on the same gauge
• —A customer or auditor has asked for an MSA

Gage R&R · measure · Yellow Belt

What should I look for when I go and watch the work?

Plan what you will watch, what you will count and what you will ask before you go — because every failure mode of a Gemba walk is decided before you arrive.

• —You are about to go and watch a process for the first time
• —A previous visit produced impressions and no data
• —You suspect the documented process and the real one have diverged

Gemba Walk Planner · measure · White Belt

Is this difference real, or could it be chance?

Compare two groups, a before and after, or a process against a target — with the effect size and the confidence interval next to the p-value rather than below a fold.

• —A change was made and you need to show whether it worked
• —Two machines, shifts or suppliers appear to differ
• —Somebody has quoted a p-value and you want to know how big the difference actually is

Hypothesis Test · analyze · Green Belt

Is the average actually telling me the truth?

Get the mean, median, range and quartiles from a set of measurements, with the mean and median shown on the same axis so you can see when they disagree.

• —A report quotes an average and you want to know whether it describes anyone
• —Your data has a few very large values and you suspect they are distorting it
• —You need to describe a set of measurements to somebody who was not there

Mean, Median & Range · measure · Foundation

Why do two teams report different numbers for the same thing?

Write a measurement definition precise enough that two people, working separately, get the same number — with the gaps that usually stop that happening flagged as you type.

• —Two reports disagree and both sides are certain their data is right
• —You are about to start collecting data and want it to mean something later
• —Somebody is averaging a 1-to-5 survey scale

Operational Definition Builder · measure · Foundation

Which problems should I fix first?

Build a Pareto chart from category counts, with the cumulative line and the vital-few cut.

• —You have counts by category and need to know where to start
• —A team meeting produced ten ideas and you need evidence, not opinion
• —You suspect a few causes carry most of the problem

Pareto Chart · analyze · White Belt

Does my problem statement already contain its own answer?

Draft a problem statement and get it checked for smuggled causes, solutions and blame.

• —You are writing a project charter
• —Your problem statement contains the word "because"
• —You need to separate the problem from the goal

Problem Statement Builder · define · White Belt

Is my process actually capable of meeting the specification?

Get Cp, Cpk, Pp and Ppk from your raw measurements — and find out first whether the process is stable and normal enough for those numbers to mean anything.

• —A customer or auditor has asked for a Cpk
• —You want to know whether the specification is achievable before promising it
• —Somebody has quoted a Cpk and you want to know whether it is warranted

Process Capability (Cp, Cpk, Pp, Ppk) · measure · Yellow Belt

Where does the time actually go between the steps?

Map the steps and the waiting between them on one scale, so the gaps that carry most of the elapsed time stop being invisible.

• —A process takes days but the work in it takes minutes
• —You have a flowchart and it has not helped you find anything
• —Two teams disagree about which step is the slow one

Process Map · measure · White Belt

How does one measurement move with another — and is a straight line the right shape?

Fit a line to two measurements — with the residual plot beside it, never behind a toggle, because the fit succeeds whether or not a straight line was ever the right shape.

• —You suspect a process input drives an output and want to know how much
• —Somebody has quoted an R-squared and you want to know what the data actually looks like
• —You need to predict a value at a setting you have measured around

Regression & Correlation · analyze · Green Belt

Where is the best setting — and is the point I found actually the best?

Fit a curved surface to two factors and locate where the gradient is zero — then find out whether that point is a maximum, a minimum, or a saddle that is not an optimum at all.

• —A factorial has identified the factors and you need the settings
• —Centre points show the surface bending, so a two-level design cannot describe it
• —Somebody has reported an optimum from a quadratic fit and you want to know its kind

Response Surface (Optimisation) · improve · Black Belt

Every step hits its target, so why do so many orders go wrong?

Multiply your step yields to find the probability an order gets through the whole process with no rework, and see the gap against the average everyone quotes.

• —Each step reports a good yield but customers still see problems
• —You need to size the rework nobody has costed
• —Somebody has averaged the step yields and called it the process yield

Rolled Throughput Yield · measure · White Belt

Is this number actually going up, or is it just noise?

Plot time-ordered data with a median centre line and apply the run chart rules, so you can tell a real change from ordinary variation.

• —A metric moved and someone wants to know whether it means anything
• —You changed something and need to show whether it worked
• —A weekly report compares this week with last week and you suspect that is meaningless

Run Chart · measure · Foundation

How many do we need — and if we cannot get that many, what can this study actually see?

How many observations a comparison needs — and, when the sample is already fixed, what it could possibly show you. The question worth asking before the trial is booked rather than after it.

• —Before booking a trial, so the number is argued about in advance rather than defended afterwards
• —When the sample is already fixed by circumstance — 40 machines, 12 weeks of history — and the useful question is what it could possibly detect
• —When a trial came back 'no significant difference' and somebody is about to read that as 'the change did not work'

Sample Size and Power · measure · Master Black Belt

What are the boundaries of the process I am improving?

Map Suppliers, Inputs, Process, Outputs and Customers on one page, then check every output has a named customer.

• —Your project scope keeps expanding and nobody can say where it ends
• —Two people describe the same process and it sounds like two different processes
• —You are about to start improving something nobody has agreed the boundaries of

SIPOC Canvas · define · Foundation

Why does adjusting after every result make things worse?

Deming's funnel, running live: choose how to adjust after each result and watch what the adjustment costs. All four rules run on the same drops, so the difference is the rule and not the luck.

• —Somebody wants to act on every movement in a metric
• —This month's target is being set from last month's actual
• —You need to show a room why leaving a stable process alone is the right answer

The Funnel Experiment · foundations · Foundation

Is our best performer actually better, or just luckier?

Six workers, one bowl, and a league table nobody can influence. Watch a ranking form, then watch the control chart show there was never a difference to rank.

• —Somebody is being ranked on an outcome they do not control
• —A target has been set below what the process can deliver
• —A league table is about to decide pay, promotion or a supplier's contract

The Red Bead Experiment · foundations · Foundation

What kind of waste am I actually looking at?

Tag observed process steps against the eight wastes and see the count and time by category.

• —You are on a waste walk and need to categorise what you see
• —You want to know which type of waste dominates before choosing a fix

Waste Spotter · lean · White Belt

How much of my process time is actually worth paying for?

Tag observed process steps against the eight wastes and compute the value-add ratio.

• —You have observed a process end to end and timed each step
• —Someone has proposed hiring more people to go faster
• —You need to show where the time really goes

Waste Walk · lean · White Belt