Who Undoes This?

A recommendation that improves your metric by degrading someone else’s will be reversed, and usually not by argument. It will be reversed by workaround, which is slower to detect and harder to fix, and what the workaround restores is the incumbent rule.

The people who do this are not obstructive. They are being evaluated on a local scoreboard and responding rationally to it. Everyone in the system is being reasonable, which is precisely why the system produces an unreasonable outcome. Pricing the error in units of consequence is necessary, and it is still local. The evidence is complete only when two further questions have been asked, what happens next and whom else it affects, and the answers have been checked for whether they change the ranking.

Everyone in the system is being reasonable

The wrong explanation produces the wrong recommendation, so it is worth being precise. The usual story is that people are parochial, shortsighted, or protecting budgets. Some are. But the behavior persists in organizations staffed entirely by conscientious people, which means the explanation cannot be character. Everyone is acting sensibly on the information that reaches them, and that information is partial, late, and framed by the scoreboard they are measured on.

The planner who pads an order is not sabotaging you. From that chair, the forecast has been wrong in one direction more often than the other, being short is punished loudly while being long is punished quietly, and nobody has ever shown them what the padding costs downstream. Given that view, padding is correct. The dealer who orders three water pumps when one is needed is the same case in another seat: if allocation is rationed in proportion to what is requested, asking for three is correct. The behavior the recommendation wants to eliminate is the rational response to a system the recommendation proposes to leave in place.

If the cause is character, the remedy is a stern memo, and you will be back in a year. If the cause is the information and the scoreboard, the remedy is to change what is visible or what is rewarded. Before writing a recommendation that depends on somebody behaving differently, work out what they can see when they act and what they are graded on. If the answer makes their current behavior look sensible, the recommendation has to change their situation rather than their attitude.

Local scoreboards

A plant manager measured on utilization will run long batches. A parts organization measured on holding cost will thin the tail. A sales team measured on bookings will promise dates operations cannot meet. Each function can hand over a request that is rational for it and expensive for the company, and an analyst who optimizes it as given becomes an amplifier for one scoreboard.

At Meridian Motors, finance’s cut is a local optimum for finance if missed jobs and expedite freight are ignored, and a 98 percent fill target is a local optimum for dealer operations if the location of the cash is ignored. A ranking for the company puts both on the same page in units of consequence. Sometimes the analyst really does sit in one function’s seat, and then the write-up should say so: “this note ranks alternatives for cash, reporting missed jobs as a side effect” is an honest informing job, and a document should not claim to be a decision while doing that.

The question

Before you write the recommendation, find the adjacent team that will dislike it if it works as designed, and go ask them what they will do about it.

Not whether they agree. What they will do. The answers are informative in a specific way.

If they say they will escalate, you have a disagreement, which is workable and belongs in the open.

If they say they will comply, ask what it costs them and whether anyone is watching that cost. Uncompensated compliance decays.

If they say they will work around it, you have not finished. A rule with a known, unpriced bypass is not a rule; it is a suggestion with a document attached, and the incumbent is still running underneath it.

Write the second round into the alternative

The challenger being installed is an input to everyone else’s behavior, and evidence that assumes that behavior will stay still describes a company that does not exist. A system-dynamics study is rarely needed. One pass of “and then what,” written in sentences, usually is.

At Meridian Motors, a protected class without a closed bypass path produces a second round in which dealers shop around, the class exists only on paper, and the team has added complexity without changing the rule that actually runs. That second round is a condition under which the challenger loses, and it belongs among the flip conditions as well as here. Expedites are the other loop analyses leave out: thin the stocking on an item class and expedite requests rise, restoring service at a cost that lands in a different budget line. The improvement is real in the spreadsheet and absent in the aggregate.

The remedy is to put the second round inside the alternative. “Cut inventory 20 percent” is incomplete. “Cut inventory 20 percent, after which dealers bypass the system and expedite freight rises enough to consume the cash released” is a claim that can be tested and rejected, and it is the version that belongs in the ranking.

Delay is why it oscillates

Feedback plus delay produces the swings that make supply chains miserable. Inventory is an accumulation, and the only way to change it is to change a rate and wait. The planner sees last week’s shortage, orders more, and the material arrives six weeks later; meanwhile the shortage is still visible, so they order again, and the shelf floods. Finance’s correction is over-applied for the same reason.

Two consequences belong in the recommendation. Responding faster to a delayed signal usually makes the swing worse, so if you cannot shorten the delay, damping the response is often the better intervention. And feeding a more accurate forecast into a blunt min/max rule can make things worse, because it gives the rule a cleaner signal to overshoot on. The interventions that change the behavior remove a delay, such as letting a supplier see consumption rather than orders, or change how hard the rule pulls on what it sees.

Where to push

Interventions are ordered roughly inversely to how easy they are to get approved. The weakest is adjusting a number, which the system absorbs because the structure that produced the old number is still there; finance’s cut, even carried into targets 20 percent lower, erodes for exactly this reason. Above that sit buffer sizes and delays. Above that sits who can see what: expedite freight at Meridian Motors is charged to a line the person triggering the expedite never reads, and routing the cost back to the requester has more force than any threshold you could tune. Above that sit the rules themselves, such as a rule that can tell a water pump from a trim clip, which is roughly the highest level a technical team reaches on its own authority. The strongest is changing the scoreboard. That is rarely the analyst’s to do and frequently the analyst’s to name: this rule will hold for about two quarters, after which the scoreboard will wear it down, and here is the scoreboard.

Constraints are often the system

The interactions that matter most are usually constraints: capacity, dock doors, a frozen production sequence, a contract. A recommendation that assumes a constraint will give way is a wish, and one that moves cost onto a neighbor’s constraint is a transfer rather than a gain. The test is whether including the constraint changes the ranking. If it does, the simpler model was not cheaper but wrong. At Meridian Motors the coupling that actually ranks the alternatives is often the exception path: if anyone can still get a water pump when the rule says no, the rule on paper is not the rule that runs.

The test, and the objection to it

Name one adjacent team that will dislike the recommendation if it works exactly as designed, and ask them what they will do about it. If the answer is that they will work around it, the rule is already failing, and you have learned that on a Tuesday rather than in the quarter after launch.

This is not an argument for never deciding. You have to pick a rule that survives the workaround, or close the workaround as part of the rule. What you cannot do is present evidence that is optimal for one function as a reason for a commitment across the company.

The obvious objection is that you cannot model the whole company. That is true, and it is not what the test asks for. It asks for one conversation with an adjacent team and one sentence describing the second round. If that sentence would change the pick, the work is not finished. If it would not, write it down as a condition the owner has considered and is prepared to accept.