Kamishibai Audit: Keep the Gains Alive Long After Handover

The Kamishibai Audit tool defines the rotating card based audit tasks that check process adherence across random themes. Each card carries one check, one frequency, one auditor and one last result. Someone pulls a card, walks to the process, looks, and marks it pass or fail.

You use this tool in the Sustain phase of RDMAICS. By then the improvement is built, the control plan is written and the project team has stepped back. The risk is no longer design. The risk is drift. People go back to the old way one small step at a time, and nobody notices until the measure moves.

Kamishibai works because it is small and repeated. You are not auditing everything. You are sampling a few themes often enough that drift gets caught while it is still cheap to correct. The word means paper theatre, and the cards are meant to be physical or visible, pulled in a rotation rather than worked through as a list.

Reach for it when a process has been handed to an owner and you want evidence of adherence rather than belief in it.

The example used throughout this guide is fictional. It is the sample data that loads with the tool, and it describes an accounts payable process some months after handover.

Five cards cover five different themes. K-01 checks SOP adherence, with the check written as operator follows SOP, pulled weekly by the AP Manager. K-02 checks board currency, board updated today, weekly, by the Supervisor. K-03 checks exception aging, no exceptions over twenty four hours, weekly, by the AP Clerk. K-04 checks data quality, log completeness at or above ninety nine percent, weekly, by IT. K-05 checks training, all staff trained, monthly, by HR.

Notice the spread. One card looks at behaviour, one at the visual management, one at the live workload, one at the data feeding the measure, and one at capability. That is the point of random themes. If every card pointed at the same thing you would get a narrow view of adherence and miss the failure that actually matters.

Notice also the owners. Five cards, five different auditors. Nobody audits their own work alone, and the load on each person stays small.

The tool asks for a small number of fields per card. Audit card, check, auditor and last result are required. Theme, frequency, corrective action and due date complete the picture.

Write the check so it has one answer. The sample checks do this well. No exceptions over twenty four hours is either true or it is not. Log completeness at or above ninety nine percent gives a threshold. Compare that with a vague check such as data looks fine, which invites an auditor to pass everything.

Frequency should match how fast the thing can drift. Four of the five sample cards run weekly. K-05, training, runs monthly, because the staff roster and the training record move more slowly than the daily work.

Corrective action and due date stay blank while a card passes. In the sample, four cards show action None and no due date. Only K-03 carries both, because only K-03 failed.

From the five cards the tool calculates five headline figures. Cards: five. Pass: four of five. Fail: one. Failed cards with an action: one of one. Actions past due date: one.

The summary names the failure directly. One of five cards failed, K-03 exception aging, check no exceptions over twenty four hours, action clear aged exceptions and confirm escalation route, due 2025-06-06. One action is past its due date.

Read those five numbers as a pair of questions. First, is the process holding? Four passes out of five says most of it is. Second, is the audit system itself working? That is where failed cards with an action and actions past due date earn their place.

One of one failed cards has an action. That part of the system is working. Somebody saw the fail and wrote a response. But one action is past its due date, which means the response stalled. The card did its job. The follow through did not.

Each card is also shown individually with its result, action, owner, theme, check and frequency, so you can take one card to a conversation without printing the whole set. The tool offers an AI review that explains the result in plain words if you want a second read.

The first decision is where to put attention this week. K-03 is the only fail, so the aged exceptions and the escalation route are the work. The other four themes need no intervention beyond pulling the card again on schedule.

The second decision concerns the overdue action. One action past due date is a small number, but it is one out of one. Every corrective action currently open has slipped. You decide whether the AP Clerk has the authority and the time to clear aged exceptions, or whether the action was written for the wrong owner. Confirming the escalation route is part of the action itself, which hints the path was unclear.

The third decision is about the card set. Four cards pass weekly. If they keep passing for a long run you can consider whether a theme has stabilised and the frequency can ease, or whether the check has become too easy to pass. Either way, the decision is deliberate, not drift.

The fourth decision is about the owner of the process. Kamishibai gives the process owner a short, factual report for a governance review. Five cards, four passes, one named fail with a named action.

A Kamishibai system decays in predictable ways, and the sample result shows two of the signals.

The clearest signal here is the overdue action. A fail with a corrective action is healthy. A fail with an action that passes its due date means the loop is open. In the sample, the due date is 2025-06-06 and the tool counts one action past due date. If that count rises across reviews, the audits are being performed but not acted on, which is worse than not auditing at all because it builds the habit of ignoring a fail.

The quieter signal is a card that never fails. Four of the five cards pass. That is a good result on one reading. On another reading, a check that has not produced a fail in a long time may be written too loosely, or the auditor may be ticking it from memory rather than from observation. Board updated today is easy to verify. Operator follows SOP needs someone to actually watch.

A third signal does not appear in the sample but is worth naming. If the same person starts auditing every card, the rotation has collapsed. Here the five cards sit with five different owners: AP Manager, Supervisor, AP Clerk, IT and HR. Watch for that spread narrowing.

The most common mistake is writing too many cards. Five is enough to cover several themes and light enough that each auditor carries one card. A set of thirty becomes a compliance chore and the checks get rushed.

The second mistake is writing a check that cannot fail. If every pull returns a pass, you have built a ritual, not an audit. Thresholds help. Log completeness at or above ninety nine percent is a check that can go either way.

The third mistake is leaving a fail without an action. The tool counts failed cards with an action for this reason. In the sample the figure is one of one. If it ever reads zero of one, the audit produced information and nothing happened with it.

The fourth mistake is treating the due date as a suggestion. The sample shows one action past due date. That single number is the one most worth tracking review to review.

Kamishibai is a sampling method. Four passes out of five does not mean eighty percent of the process is correct. It means that on the days those four cards were pulled, those four checks passed. A card pulled on a different shift, with different staff and a different workload, could return a different result.

The tool does not verify that an audit was performed. It records a last result. If an auditor marks a pass without walking the process, the tool will show a pass. The integrity of the system sits with the people pulling the cards, not with the software.

It does not tell you why K-03 failed. It names the card, the theme, the check, the action and the due date. The cause of aged exceptions, and the reason the escalation route needs confirming, come from an investigation you run separately.

It does not track history in this view. You see the last result for each card, not the run of results behind it. Two cards both marked pass may have very different records. Keep your own record if you want to see trends.

Finally, it does not replace the control plan or the process measures. Kamishibai checks adherence to the way of working. The measure tells you whether that way of working is still producing the outcome. You need both, and the sample set of five cards is designed to sit alongside a measure, not in place of one.