Every Decision Already Has a Rule

A request for analysis describes outcomes: premium freight is up, inventory is too high, fill keeps slipping. It rarely describes the rule that produced them, because requests are written by people who see results, and results are produced by rules the requester may never have watched fire. “Figure out what’s happening” is an honest request from someone who reads the inventory report and has never sat through the planner’s morning.

The analyst’s first job is to supply the missing description. Not the numbers, but the rule producing them, written so the people who run it would recognize it.

The documented rule is rarely the incumbent

At Meridian Motors the documented rule for a dealer-facing part is a min/max: when the projected balance at a distribution center falls below the minimum, the system proposes an order up to the maximum, with both levels set from one service target applied to every item.

Follow one water-pump order through a week and a different rule appears. The system proposes four units. The planner, who remembers a winter when a cold snap cleared the shelf and three dealers called her manager, raises it to eight. The order arrives later than the system assumed, because the supplier ships to that center twice a week. Meanwhile a dealer with a vehicle on the hoist finds the center short, calls another center or buys outside, and that demand never reaches the planning system at all.

The operating rule is the combination: the proposal, the adjustment, the supplier’s calendar, and the dealer’s way around the shortage. Only the first is in the documentation.

A challenger measured against the documentation wins on paper

Every recommendation is a comparison against the incumbent. A team that never describes it is still comparing, only against an imagined baseline, usually the documented rule, and the error runs in a predictable direction. The challenger gets credit for improvements the incumbent already achieves informally, and it escapes blame for failures the informal parts prevent, because the backtest never contained the pad or the bypass. It then loses to the operating rule on the first busy week, and the people who run that rule know it long before the analyst does.

Technical people make this mistake because the documented rule is the one with data. The system’s parameters are queryable; the planner’s pad lives in a judgment and the dealer’s bypass leaves no record. Analysis gravitates to what is tabulated, and the incumbent is mostly what is not.

Six places the incumbent hides

The undocumented parts tend to sit in the same few places, and each can usually be found in an afternoon.

The unofficial spreadsheet. Between the system’s recommendation and the transaction that executes it there is often a file: a planner’s sheet, a capacity tracker, a pricing override list, a whiteboard. The system’s output goes in and something different comes out, and the something different is what the organization does.

The pad. A rule with one parameter, applied by judgment, concentrated on the items or situations where its owner has been hurt before. It shows up as a consistent gap in one direction between what was proposed and what was done.

Last quarter’s policy. Organizations issue rules more readily than they retire them. A stocking freeze from a cash squeeze, an approval threshold from one bad month years ago: these persist because revoking them requires someone to notice and take responsibility. Ask what people are not allowed to do, and since when.

The system default. Every field takes a value when nobody touches it, and in a busy process most fields are not touched. The safety-stock parameter set at implementation decides most of the cases it governs, and does not feel like a decision to anyone.

Habit. What is done because it has always been done, often a fossil of a commitment that once made sense. Listen for “we always” and “we never.”

The person who just knows. The planner who knows which supplier lies about ship dates holds a rule in every sense that matters. It is visible mainly when that person is on vacation.

At Meridian the full incumbent is several of these at once: the documented min/max, a pad on cooling and drivetrain parts, an approval rule on expedites left over from a cash crisis, a default service target nobody has revisited since installation, and a dealer habit of calling a center manager directly. The same structure appears in retail pricing overrides, underwriter exceptions to a credit score, and a call-center supervisor’s extra staff on billing-cycle days. In each, the override contains knowledge the model lacks.

The incumbent is information, not a defect

The natural reaction to finding a shadow spreadsheet is to treat it as a control weakness to be eliminated. That is usually a mistake. The undocumented parts are better read as a record of what the official rule fails to do, kept by people who could not wait for it to be fixed.

Read that way, the incumbent shows what commitments the operation has made that no system records, what its operators are scored on (pads cluster around whatever they will be blamed for), where the official data is wrong, and where a challenger has the most to gain and the most to lose. It has also survived every peak week and supplier failure in its lifetime. That does not make it optimal, since rules survive for reasons other than merit, but survival is evidence of something.

Reading it as information is not deference. Some of it is obsolete, some is wrong in ways its owners cannot see because they observe the shortages a pad prevents and never the cash it ties up, and some serves a local scoreboard at the company’s expense. Four questions sort each part: what is it protecting against, who would notice first if it stopped, when was it last wrong, and what would its owner need to see before giving it up. The tone matters practically. A planner asked what her sheet knows will explain it and may help configure the replacement; a planner told it is a control weakness will keep it privately, and the challenger will be installed on top of a rule still running underneath.

The status quo is a pick

The incumbent is the only alternative that needs nobody to choose it, which makes it easy to forget that keeping it is a choice with consequences. A meeting that ends without a commitment, a request for more analysis, a dashboard to monitor the problem: each feels like the absence of a decision, and each leaves the incumbent running on Monday.

Two things follow for the menu. The status quo belongs on it as an alternative in its own right, with its costs stated: at Meridian, a known amount of cash in slow parts, a known rate of missed jobs on parts that stop vehicles, and finance’s request still unanswered. And the burden of evidence should be symmetric. Challengers are backtested and stress-tested while the incumbent is assumed to work because it is running. Apply the same scrutiny to both. Recommending the status quo after examining it is a decision; leaving it in place because nobody examined it is a default.

The five parts expose what is unwritten

Any repeating rule has a state it sees, a decision it may take, information it learns only afterward, a transition to the next state, and an objective it accumulates. Used as a checklist on the incumbent, they show the gaps. Meridian’s documented state is the system’s fields; the operating state includes the planner’s memory of which parts burned her. The documented decision is the proposed quantity; the operating one adds the pad. The documented information is a forecast; the operating information includes dealer bypass, which hides demand and makes the forecast look better than the service dealers receive. And the documented objective is a uniform service target, while the operating rule pursues something closer to avoiding the calls that reach the manager. Each gap is a place a challenger built from the documentation will be surprised.

The test

Before any analysis, pull a few months of what the system proposed and what was actually executed for the decision you were asked to inform. Where they match, the documented rule is the incumbent. Where they diverge, someone is making the difference: find them, watch one cycle of their work, ask to see the file, and ask what happens when they are away.

Then write the incumbent as a rule in five or six sentences: who runs it, what it sees, what it does, what it does when unsure, how exceptions happen, and who owns each part. Carry one question into the first conversation: if nobody read our analysis, what would happen to these orders next Monday, and who would decide it?

If you cannot write that paragraph, you do not yet know what your challenger has to beat. If nobody can say who owns a part of it, you have already learned something about who can change it.