Run an improvement project
Five phases, and a gate between each one that asks for evidence rather than for a document. You cannot leave a phase until the things it exists to produce exist — which is the whole difference between this and five columns with a next button.
Not signed in — this project stays in this tab.
Define
Agree what problem is being solved, for whom, and where the process starts and ends. The output is a scope somebody could hold you to.
5 things still standing between you and Measure
A problem statement with a number and a period
A statement without a quantity cannot be shown to have improved, and one containing a cause has already decided the answer before anything was measured.
Process boundaries: where it starts and where it ends
The single most valuable line in the project. Unstated boundaries expand until the work cannot be finished, and every person involved will sincerely remember agreeing to a different scope.
Each customer named, with what they actually want
Naming a customer without writing their need is how a team optimises for whatever is easiest to measure. Writing the sentence is usually the moment somebody says "actually, that is not what they care about".
Can the problem in your statement be measured with data that already exists, or that you can start collecting this week?
If the answer is no, the project cannot show it worked. That is not a reason to stop — it is a reason for the first piece of work to be building the measurement.
Is the problem statement free of a proposed solution?
A statement naming the fix commits the team to an answer before anyone has measured the question, and every subsequent step will be aimed at confirming it.
1 optional item not done
Sponsor, process owner and the people who do the work
Without a process owner the improvement decays after the project closes and reappears as next year's project.
These do not block the phase. They are listed because a deliberate omission and an oversight look identical on a checklist that hides one.
What this phase produces
A statement without a quantity cannot be shown to have improved, and one containing a cause has already decided the answer before anything was measured.
The single most valuable line in the project. Unstated boundaries expand until the work cannot be finished, and every person involved will sincerely remember agreeing to a different scope.
Naming a customer without writing their need is how a team optimises for whatever is easiest to measure. Writing the sentence is usually the moment somebody says "actually, that is not what they care about".
Without a process owner the improvement decays after the project closes and reappears as next year's project.
The tollgate questions
“No” is a legitimate answer. It means the work is not finished, not that a box is unticked — and recording it honestly is worth more than the phase moving on.
Nothing here leaves your browser. A project workspace fills up with your employer’s defect rates, your colleagues’ names and your process’s failures. You are not signed in, so this page stores nothing and sends nothing — and it will be gone when you close the tab. Signing in saves it to your account alone, still never to a third party.
The phases, and what each gate asks
The whole model in plain text — every deliverable with the reason it is required, and every tollgate question with what a bad answer usually hides.
1Define
Agree what problem is being solved, for whom, and where the process starts and ends. The output is a scope somebody could hold you to.
Produces
- A problem statement with a number and a periodA statement without a quantity cannot be shown to have improved, and one containing a cause has already decided the answer before anything was measured.problem-statement-builder
- Process boundaries: where it starts and where it endsThe single most valuable line in the project. Unstated boundaries expand until the work cannot be finished, and every person involved will sincerely remember agreeing to a different scope.sipoc-canvas
- Each customer named, with what they actually wantNaming a customer without writing their need is how a team optimises for whatever is easiest to measure. Writing the sentence is usually the moment somebody says "actually, that is not what they care about".sipoc-canvas
- Sponsor, process owner and the people who do the workoptionalWithout a process owner the improvement decays after the project closes and reappears as next year's project.
The gate asks
- Can the problem in your statement be measured with data that already exists, or that you can start collecting this week?If the answer is no, the project cannot show it worked. That is not a reason to stop — it is a reason for the first piece of work to be building the measurement.
- Is the problem statement free of a proposed solution?A statement naming the fix commits the team to an answer before anyone has measured the question, and every subsequent step will be aimed at confirming it.
2Measure
Establish what is actually happening now, with a measurement you trust and a baseline that is stable enough to compare against later.
Produces
- An operational definition two people could followUntil the measurement means the same thing to two people, every number after this is a different number depending on who took it.operational-definition
- A baseline chart with enough points to readThe comparison at the end of the project is against this. A baseline of two points cannot be compared with anything.run-chart
- A measurement systems analysisoptionalObserved variation is process variation plus measurement variation. Where the gauge has never been checked, part of what you are about to improve may be the gauge.gage-rr
The gate asks
- Is the baseline stable, or have you stated plainly that it is not?Both are acceptable answers. What is not acceptable is not knowing: a comparison against an unstable baseline compares against a moving quantity, and any result can be produced by choosing when to stop.
- Do you trust the measurement enough to make a decision on it?If two people measuring the same thing would disagree, the numbers describe the measurement system as much as the process — and no amount of improvement work downstream can recover from that.
3Analyse
Find the causes, and CHECK them. The output is not a list of candidates — it is the one or two that survived contact with data.
Produces
- Candidate causes, generated widelyA team that generates three causes generates the three it already believed. The categories exist to force a search somewhere else.fishbone
- At least one cause CHECKED against dataThis is the difference between an analysis and a meeting. Until something is counted, the longest branch of the diagram is the one the room talked about most.
The gate asks
- Is the cause you are about to fix supported by data rather than by consensus?Agreement in a room is not evidence. The cheapest check is usually a count of something already recorded, and it eliminates most of the list in an afternoon.
- Is the cause something you can actually change?A cause naming a person or an attitude means the chain stopped early. Ask what about the work made that easy to do or hard to avoid; the answer is the cause.
4Improve
Change the cause and show the change worked, on a trial small enough to reverse.
Produces
- The change, and a prediction specific enough to be wrongA prediction that cannot be false produces no learning. "This should help" will be judged true by whoever proposed it, whatever the data does.
- How much data the trial needs, decided before it runsoptionalA trial too small to detect the difference you came to find returns "no significant difference" whether the change worked or not — and that result gets written up as evidence against it. Optional because many trials are sized by circumstance rather than chosen; what is not optional is knowing what the size you have can see.sample-size
- The trial result, measured the same way as the baselineMeasured differently, the comparison shows the definition change rather than the improvement. This is why the operational definition came first.run-chart
- The benefit you expect, stated as a forecastoptionalClaimed here and VERIFIED in Control. A saving banked the week a change is made is a forecast wearing a result's clothing.
The gate asks
- Is the trial small enough to stop if it does not work?A change rolled out everywhere at once cannot be evaluated and cannot be withdrawn, so it will be defended rather than tested.
- Did the measurement move in the way you predicted, by the rule you set in advance?"It did not work" is a successful cycle — you learned something for the cost of one trial. Deciding the rule after seeing the data is not a test, it is a description.
5Control
Show the improvement held after the attention moved on, and leave behind something that keeps it there.
Produces
- Evidence the improvement held, measured after the attention moved onThe gate most projects fail and the only one that cannot be passed by writing something. A change that works while it is being watched has not been shown to work.control-chart
- The new way of working, written down where the work happensA standard in a folder is not a standard anybody works to. It belongs at the point of use, and somebody has to own keeping it current.
- The benefit as measured, next to what was forecastoptionalThe two side by side is what makes the next forecast worth anything. Reporting only the forecast is how an organisation stops believing benefit numbers.
The gate asks
- Has the improvement held for a period measured AFTER the project attention moved on?A change works while it is being watched almost by definition. The question is what happens on a night shift in week six, and only measurement answers it.
- Is there a named person responsible for the new way of working after this closes?Without one, the standard drifts, the visual control stops being updated, and this returns as next year's project with a new team.