Improve Phase Plan: Turn a Diagnosed Cause Into a Pilot Worth Running

The Improve Phase Plan does one job. It plans the activities, owners and timeline for the Improve phase so that solution generation, piloting and validation actually happen in order.

You reach for it at the moment the Analyse phase closes. You have a cause you believe in. You have evidence behind it. What you do not yet have is a change anyone has tested. The gap between those two states is where most improvement work quietly stalls. People jump from a confirmed cause to a full rollout because the cause feels obvious. Then the rollout underdelivers and nobody can say why.

The plan forces a slower, safer path. Generate options. Screen them. Design the future state. Pilot it. Measure the pilot. Only then take it to a tollgate. Each of those steps gets a named owner and a target date, so the phase has a shape you can inspect rather than a vague intention to improve something.

Use it early in Improve, not halfway through. A plan written after the pilot has already started is a record, not a plan.

Everything in this guide is built from the sample data that loads with the tool. The case is fictional. It is there so you can see the mechanics work before you put your own project into it.

The project is an invoice handling improvement. The notes in the plan point to the intended change: optical character recognition combined with automatic routing. That is the solution concept the team wants to test. It is not yet a decision. It is a candidate that has to survive screening, design, a pilot and a statistical check.

Six activities carry the phase. Ideas are generated with brainstorming and SCAMPER. Options are screened on an effort and impact matrix. A future state process map is drawn. A two week pilot is built and run. Results are tested with a before and after t-test. A tollgate review is prepared for the sponsor.

That sequence is the point. The pilot sits fourth, not first. Three activities happen before anyone touches the live process.

The worksheet is a table with six columns. Activity, owner, target date, status and notes, numbered in sequence. You fill it in once and then maintain it as the phase runs.

Keep activity names concrete. In the sample, activity one is not called ideation. It is called generate solution ideas with brainstorm and SCAMPER. You can tell from the name what will be done and how. Activity five is not called validation. It is called validate pilot results statistically, with a before and after t-test named in the notes. A reader who was not in the room can still follow it.

Owners are roles here, not individuals. Team, Black Belt, Process Analyst, Green Belt. That works for a sample. On a live project you would name people, because a role cannot be chased. The tool checks whether an owner exists at all, so a blank is visible immediately.

Notes carry the method and the scope. Thirty two ideas generated. Effort impact matrix. OCR plus auto routing. Two week OCR pilot. Before and after t-test. Present to sponsor. Those short entries do more work than long descriptions, because they tell you what evidence the activity will produce.

The tool builds a delivery and ownership view from the worksheet. It takes the milestones from the target date column and checks them against a review date, which in the sample is 2026-10-04.

The headline reads plainly. Nothing is past its target date. One of six complete. Three counters sit underneath. Past target date is zero. No owner is zero. Complete is one of six. A status breakdown follows: one completed, one in progress, four not started.

Then the plan is laid out as a list of milestones in date order, each with its date, status and owner. Generate solution ideas, 2026-10-07, completed, Team. Screen and prioritise solutions, 2026-10-11, in progress, Black Belt. Design future state process map, 2026-10-15, not started, Process Analyst. Build and run pilot, 2026-10-25, not started, Green Belt. Validate pilot results statistically, 2026-10-30, not started, Black Belt. Prepare Improve tollgate review, 2026-11-04, not started, Green Belt.

The tool also offers an AI review that talks through what the delivery picture means. Treat it as a second reader, not as the decision.

The plan is where you decide how much you are willing to spend to find out whether the change works. In the sample, that figure is visible in the shape of the schedule rather than in a budget line.

Thirty two ideas were generated. Those thirty two are screened on an effort and impact matrix, which is a sizing step. It asks how much each option would cost to implement and how much it would move the measure. You are not committing to anything at that point. You are ranking.

The future state process map comes next. That is where the gain becomes specific. A map of OCR plus auto routing tells you which steps disappear, which steps change hands and where the time is saved. If the map does not show a meaningful change, the pilot is not worth running and you have found that out for the price of a design session.

Only then does the pilot start, and it is scoped to two weeks. A bounded pilot is a bounded bet. The statistical validation five days later decides whether the observed difference is real. The tollgate on 2026-11-04 is where the sponsor sees the evidence and chooses.

The first decision is whether the phase is actually resourced. Four of six activities are not started and the Black Belt owns two of them, including the statistical validation. If the Black Belt is stretched, the validation slips, and a slipped validation means the tollgate goes ahead on opinion.

The second decision is sequencing. The gap between the pilot ending around 2026-10-25 and the validation on 2026-10-30 is short. You can see that in the dates and ask whether five days is enough to collect and clean the pilot data.

The third decision is escalation. Zero items past target date and zero items without an owner is a clean position. It gives you a baseline. The moment one of those counters moves off zero, you have something specific to raise rather than a general sense that the project is drifting.

The fourth decision is tollgate readiness. The plan ends with a prepared review. If activities three, four and five are still not started a week out, the tollgate date is the thing that has to move, not the evidence standard.

The most common mistake is skipping the screening step. Teams generate ideas, fall in love with one, and go straight to build. The sample plan puts screening second for a reason. Thirty two ideas were generated and the matrix exists to reduce them honestly.

The second mistake is a pilot with no end date. The sample scopes a two week OCR pilot. A pilot that runs until people stop paying attention produces data nobody trusts.

The third mistake is calling the pilot a success because it felt better. Activity five exists to stop that. A before and after test either supports the change or it does not.

The fourth mistake is leaving statuses stale. The delivery view is only as accurate as the status column. If nobody updates it, the counters report a plan that no longer exists.

The Improve Phase Plan does not tell you whether your solution will work. It schedules the work that will answer that question. The pilot answers it. The statistical test checks the answer. The plan only makes sure both happen.

It does not calculate savings. There is no financial figure anywhere in the sample output. If you need a benefit case, you build that from the pilot measurements and your own cost data, not from this view.

It does not validate your root cause. If Analyse handed you the wrong cause, this plan will carry that error all the way to a well organised pilot of the wrong thing. The tool checks dates, owners and status. It does not check logic.

It does not manage dependencies or workload. It shows that the Black Belt owns two activities. It does not know whether that Black Belt is also running three other projects. It does not know that the process map must finish before the pilot can be built, beyond the order you gave it.

And it does not replace the tollgate conversation. Zero overdue items means the schedule is intact. It does not mean the change is worth implementing. That judgement stays with you and your sponsor.