Install the Rule

Every challenger is installed against a rule that is already running. The incumbent does not stop when a meeting adopts its replacement. It fires again on the next weekly run, from the same screen, through the same planners, protected by the same scorecards, until something specific makes it stop.

A recommendation that never enters the operating process has not been deferred. It has chosen the incumbent for another cycle, and agreement in principle with Tuesday unchanged is only a good meeting. The install is finished when the new rule is what fires on the next run.

What has to be specified

These items are all the same kind of thing, operating facts, and that is what makes them summarizable as one rule.

Owner of the running rule. Not the sponsor of the study. The person who can change a parameter without convening a task force. At Meridian Motors that is Parts Planning, not analytics. If analytics retains the parameters, the rule dies when the analyst changes seats.

Cadence. When the rule fires, and separately when people review it. Orders go out on the existing weekly run; classes recompute monthly. “Ongoing” is not a cadence. Ongoing is how freeze dates pass.

The rule itself. Thresholds, classes, freeze logic, and what the system does when it is unsure. Unsure is a case. If you do not specify it, a planner will specify it with a pad, and the pad will be invisible.

Exception path. Who may depart from the rule, what that costs, and where it is recorded. An unpriced, unrecorded exception path is already a shadow rule; the expedite trail is one, and the new rule should not install another.

Feedback. What the owner and the operators will look at to see whether the ranking still holds: holding dollars, stranded jobs, expedite dollars, and bypass share, each by class. A scoreboard for this rule, not a tour of last quarter.

Revisit triggers. Dates and conditions, not “ongoing monitoring.” The third of July, or sooner if bypass exceeds 8 percent for four weeks. A date can be met or missed. Monitoring cannot.

The items also have to point the same way: a protected class with an unpriced expedite path is two rules in conflict, and the floor will resolve it in favor of whichever is easier. And one design requirement gets checked at install time: sit with a planner through the first run, confirm every named field is populated, and confirm the degraded behavior fires when one is not. A gap found there is a change to the rule, and the ranking should be rechecked against the rule as it will actually run.

Legibility is a feature, and it is not the same as trust

A rule a planner does not understand will be routed around, and the routing will not be reported. The effective value of a rule is the value it delivers when followed multiplied by the probability that it is followed, so a simpler rule with a higher compliance rate frequently beats a better rule with a lower one. Three-class stocking is explicable in one sentence: protect the items that stop vehicles, drain the items that are only cash, and run the middle tighter than today.

Legibility gets the rule into conversation. Trust is whether running it is safer, for the person at the counter, than the pad they already have, and a perfectly tellable rule still loses to the pad if the pad protects a bonus, a fill number, or an angry phone call. Until a staged install has actually fired, the pad is the operating rule, whatever was adopted on Friday. That belongs in the ranking as an adoption constraint, not in a later add-on called change management.

Replacing an incumbent is the change

The forces that keep the incumbent in place are strongest on the day of the install, because that is the first day anyone has to act differently. Each install move answers one of them.

Give the rule an owner with standing. The incumbent never needed a defender. A challenger always does, because a rule nobody is authorized to adjust will be adjusted informally by whoever is closest to the screen. Priya Shah sponsors the decision and signs the kill condition; the planning manager in Parts Planning owns the parameters and is named on the change ticket.

Let the operators design the uncertain cases. The analyst owns the ranking. The people who will run the rule should write the parts where the ranking is least certain: what happens when an input is missing, which reason codes the exception path offers, where a new item sits before it has history. The unofficial spreadsheet exists to handle exactly those cases, and the loss an operator fears most tends to sit there too. At Meridian Motors, a part with too little history stays on today’s rule, and no part leaves the protected class on missing data. Those two lines cost a little cash on paper and remove the most common reason to keep a private list.

Make the new way easy and the old way cost effort. The challenger already owns the default. The old target stays visible for reference, but reverting to it requires a reason code that appears on the weekly list, and the informal expedite call is routed through the logged path rather than banned. Banning the old way moves it out of sight; a little effort keeps it visible and measures whether the install has taken.

Stage the install, and make early results visible. Two distribution centers, ninety days, a kill switch, and a named date for the expansion decision. A staged install without a scheduled decision about the next stage is a pilot that never ends. Send the scoreboard weekly to the planners and to dealer operations, in units they recognize, because the incumbent’s benefits are felt immediately and the challenger’s are invisible until someone shows them. Name whoever holds the kill switch before launch, so that pulling it is a procedure rather than a negotiation.

Change what the pad is paid for. The planner is paid for fill and rarely charged for the cash. Count holding dollars by class on the planner’s scorecard, charge expedites to the requesting region, and record a protected-class miss that occurs while the rule was followed as the rule’s miss, reviewed by the owner, not counted against the planner. That third change matters most: training will not remove the weight people place on losses, but who bears the loss is a design choice. Where the scorecard belongs to someone outside the decision, the ranking should say whether the challenger still wins if it does not change.

These are operating facts, specified before launch in the same document as the thresholds. They are not phases to be worked through afterward.

When you are not allowed to install

Sometimes the owner agrees and the change enters another function’s queue. Treating that as the end of your involvement is how the decision decays and returns in six months as a new project. Two inexpensive measures hold it together. Attach the write-up, with its reasons and flip conditions intact, to the ticket that represents the change, so anyone can answer why the choice was made and whether that reason still holds. And set a date on which the decision is re-examined rather than re-argued. If the change has not happened by then, the honest finding may be that the organization has chosen the status quo, and somebody should be told so in those words. “This recommendation is not installable until Parts Planning owns the class parameters” is a finding about the organization, not a personal failure to influence.

How installed rules die

Rules installed correctly still decay, usually after their installers move on, and an unmaintained rule still runs exactly as written. Four preventives belong in every specification.

A maintenance owner who is a role. Name the role and the current person, write down the maintenance procedure so a successor can follow it, and treat a change in either, including a reorganization, as a trigger to revisit.

A revisit date that survives a reorganization. It sits on the owning role’s calendar and on the change ticket with the flip conditions, not in one person’s memory.

Decay signals with alarms. A recompute job that silently stops leaves the classes as a photograph of an earlier season. An exception share creeping upward means the exceptions are becoming the rule. The rule’s own measure drifting past a threshold for several weeks, such as bypass share, should alert the owner and one level above.

The spreadsheet that returns. A private spreadsheet reappearing beside the rule is the most reliable sign of decay. When it returns, it is the incumbent again, and the rule has to win again: run the two side by side, show the result to the people who built the spreadsheet, and retire it with their agreement.

A challenger that wins becomes the next incumbent. One installed with its owner, assumptions, flip conditions, and maintenance procedure written down is easier to challenge on the merits, which is the condition under which it deserves to keep running.

A five-minute test

Describe the first Monday after the freeze in one paragraph: what number is in the box on the planning screen, what a planner does to depart from it, where that departure shows up on Friday, who reads it, and what condition would switch the rule off. If any clause is “the team will monitor” or names no person, the incumbent is still the rule that fires.