Out-of-Control Action Plan: How to Hold the Gain After the Team Moves On

The Out-of-Control Action Plan tool does one job. It defines the immediate containment and investigation actions when a control chart signals an out-of-control condition. You write the plan before the signal happens, so nobody has to improvise later.

You reach for it in the Control phase of RDMAICS, after the improvements are in place and the measures are stable. By then you know what the process looks like when it behaves. You also know the ways it tends to misbehave. The plan turns that knowledge into instructions that survive a handover.

The reason this matters is simple. Project teams disband. The Green Belt moves to a new problem. The person who tuned the process takes a different role. What stays behind is the chart on the wall and the plan that says what to do when a point lands outside the limits. Without the plan, the chart becomes decoration.

Everything below uses the sample data that loads with the tool. The case is fictional. It describes an accounts payable process that has just finished an improvement project and is moving into ongoing control.

The process has four known failure modes. Error rate can climb above the upper control limit. Cycle time can trend upward. Duplicate flags can cluster. Match exceptions can spike. Each of these has a plausible cause, a way to contain the damage today, and a longer investigation that fixes the root.

Four signals is a reasonable number for a first plan. You are not trying to list every possible deviation. You are listing the ones your data and your analysis say are likely, and giving each one a response that a person on shift can follow without calling anyone.

The worksheet asks for six fields per row. Signal, likely cause, immediate containment, investigation step, owner and timeframe. Each row is one trigger condition and one complete response.

In the sample, row one covers an error rate above the upper control limit. The likely cause is OCR template drift. Containment is to switch to manual verify. The investigation step is to re-train OCR on the new template. The Green Belt owns it, same day.

Row two covers cycle time trending up, caused by an approver bottleneck. Containment is to auto-delegate low-value items. The investigation is a review of the approval matrix, owned by the AP Manager within 24 hours. Row three covers a duplicate flag cluster from vendor resubmission. Containment is to hold the duplicate batch, immediately, while the AP Clerk traces vendor and portal logs. Row four covers a match exceptions spike from PO-invoice mismatch. Containment is to route to the exception queue while the Process Owner checks master data within 48 hours.

Write the signal as something a person can observe on a chart. Write the containment as something they can do without approval. Keep the two separate. Containment protects today. Investigation protects next month.

The tool summarises the worksheet into a small set of counts. For this sample it reports four signals. It reports containment on four of four. It reports response within 24 hours on three of four. It reports the slowest response as 48 hours.

The headline line is the one that matters most. Every signal has an immediate containment step. That is the test of a usable plan. A signal without containment is a signal that will sit unaddressed while someone works out what to do.

The result then lists the four investigation steps against their owners. Re-train OCR on new template sits with the Green Belt. Review approval matrix sits with the AP Manager. Trace vendor and portal logs sits with the AP Clerk. Check master data sits with the Process Owner. Four signals, four different owners, which spreads the load rather than parking it all on one person.

The tool also offers an AI review that explains what the result means for your plan.

The first decision is whether the plan is complete enough to hand over. Four of four signals with containment says yes on that measure. If the count had read three of four, you would know exactly which row to fix before the handover meeting.

The second decision concerns speed. Three of four signals get a response within 24 hours. The slowest is 48 hours, on the match exceptions row. You can accept that, or you can argue it down. The question to ask is what happens to the invoices sitting in the exception queue during those two days. If the containment holds them safely, 48 hours is defensible. If it does not, tighten the timeframe.

The third decision is about ownership spread. The sample puts one investigation on the Green Belt. A Green Belt is often a project role rather than a permanent operational one. That is worth a conversation before the project closes, because an owner who leaves is an owner who cannot respond.

The most frequent mistake is writing containment and investigation as the same action. If your containment step is to re-train the OCR, you have a plan that takes days to execute during a live failure. Containment must be fast and reversible. Switching to manual verify meets that test. Re-training does not.

The second mistake is naming an individual instead of a role. The sample uses Green Belt, AP Manager, AP Clerk and Process Owner. Roles persist. People move. A plan written against names needs rewriting every time somebody changes job.

The third mistake is leaving the timeframe blank or vague. Soon is not a timeframe. The sample uses same day, 24 hrs, immediate and 48 hrs. Each of those can be checked. A plan you cannot audit is a plan nobody follows.

The fourth mistake is writing the plan and never testing it. Walk a real signal through the plan with the owner present. You will find the step that assumes access somebody does not have.

A plan that sits in a folder decays. Put it where the chart is. If the chart is reviewed in a daily stand-up, the plan belongs in the same place, in the same format, with the same four rows visible.

Review the plan when the process changes. New vendor portal, new approval matrix, new master data structure. Each of those can make a row obsolete or create a signal you have not listed. The sample has four rows because that reflects what the project team knew at close. It is a starting point, not a final answer.

Log what actually happens. When a signal fires, record which row applied, whether containment worked and how long the investigation took. Over time that log tells you whether 48 hours is the right timeframe or whether it needs to change.

The tool does not detect signals. It does not read your control chart, calculate limits or tell you when a point has gone out of control. You need a separate control chart for that. This tool only describes the response.

It does not verify that your likely causes are correct. OCR template drift is a hypothesis written into the worksheet. The tool counts it as present. It cannot tell you whether the real cause is something else entirely. That remains the job of your Analyse phase work.

It does not enforce the plan. Four of four signals with containment is a count of what you wrote, not evidence that anyone did it. Owners do not get reminders. Timeframes do not trigger alerts. Compliance depends on the management routine you build around the plan.

It also does not judge whether your timeframes are sensible for your risk. The result reports 48 hours as the slowest response. It does not say whether 48 hours is acceptable. That judgement belongs to you and the Process Owner.