Problem, Decision, Outcome

A typical brief reads: dealer inventory is too high, we keep missing fill rate, finance wants a 20 percent drawdown, and can we get a dashboard up for Friday.

That is several different objects in a few lines, and they get argued as if they were one.

The problem is what is wrong or under strain in the world: too much cash in dealer-facing parts, and too many jobs lost or delayed at the dealer. The decision is what someone can change: keep the current replenishment rule, take finance’s cut (every item down 20 percent by the freeze, then the current rule at targets 20 percent lower), or adopt class-based stocking. The outcome is what should improve as a consequence, in units the company values: working capital, and the cost of jobs that stop vehicles. The 20 percent is a proposed decision presented as a fact of life. The dashboard is none of the three. Fill rate, taken as the thing to move, is a proxy standing in for an outcome.

These are distinctions inside the choice sentence, not a second framework beside it. Keeping them apart is what lets the sentence mean something.

Why the mash is so natural

Treating it as sloppiness misses the mechanism, and the mechanism is what makes it stubborn.

All three can be true at once, and organizations speak a language that does not distinguish them. “We have an inventory problem” can mean the sheds are full, which is a problem; or that we should cut twenty percent, which is a decision; or that turns are below target, which is an outcome. A team that hears only the sentence tries to satisfy all three with one artifact, usually a dashboard, because a dashboard can be said to address a problem, support a decision, and track an outcome without committing to any of them.

Incentives push the same way. Functions are paid on outcomes they can report, so they request work in the language of those reports. Finance asks for inventory down. Dealer operations asks for fill up. Neither has to name a mechanism, and naming one would expose that the two requests may not be satisfiable by the same rule, or only if items are treated differently, which is a fight rather than a dashboard.

The mash is socially cheaper than the split. That is why it survives competent people.

Write the mechanism

The quickest test is one sentence: this commitment (the decision) changes this behavior in the process (the mechanism), which changes this consequence (the outcome). If the middle cannot be filled without a miracle, there is no decision yet, only a hoped-for scorecard.

At Meridian Motors, a class-based stocking rule changes which items receive cash and which receive protection, which changes holding cost and the cost of jobs that stop vehicles. A fill-rate dashboard changes what people look at in a meeting, which may or may not change an order; the middle step is a hope.

The same sentence shows how far the house metrics are from the outcome. Fill rate says nothing about whether the missed lines stopped vehicles, and turns say nothing about whether the remaining stock is the useful kind. At Meridian the cash and the missed jobs sit in different items, so a metric that averages across items hides exactly the difference the ranking depends on. “Improve fill two points and increase turns” can go on a slide without anyone naming a decision. “Reduce working capital on items that almost never stop a vehicle, and reduce the cost of jobs that do” cannot, because the sentence does not make sense until a commitment is attached.

The KPI sandwich

The characteristic failure is a brief with a metric on both sides and no decision in the middle. Fill rate is the problem, fill rate is the target, and what sits between them is “better forecasting,” which is a capability rather than a commitment.

Capabilities are not decidable. Nobody can approve “better forecasting.” They can approve a specific rule change, with a specific owner, by a specific date, and the moment that sentence exists the sandwich falls apart and the real disagreements become visible.

A stated mechanism can be wrong, and that is its value

The mechanism test does more than expose dashboards. It is how a team discovers that the decision it was asked to inform is not the one driving the outcome. When that happens, every blank in the choice sentence can move at once: the owner, the alternatives, and the date that matters.

Priya’s email implies a mechanism: the network holds the wrong total amount of stock. If that were right, the live question would be how much, and finance’s cut would be a natural candidate. Put the cash and the missed jobs side by side at the item level and the mechanism changes: the excess sits in cheap, slow parts and the misses sit in parts that stop vehicles. The decision moves from how much stock to which items get it, and a blanket cut becomes the alternative most likely to make the outcome worse. Dealer bypass sharpens the point. Demand a dealer takes elsewhere never reaches the planning system, so recorded fill looks better than the service dealers receive, and a mechanism built on recorded fill is built on a flattering number.

Summaries are where wrong mechanisms survive, because aggregation erases the moment at which the mechanism operates. The person who sees the error first is often not the analyst but whoever watches the work happen. Treat a stated mechanism as a claim, and look for the cheapest test of it, which is often one the people on the floor can describe in a sentence.

Observations are not diagnoses

A fourth distinction belongs beside the three. “Fill rate is down” is an observation: something that can be counted and can be true without saying what to change. “Fill rate is down because one uniform target overprotects slow parts and underprotects the ones that stop vehicles” is a diagnosis: a claim about the mechanism that generates the observation. The decision is the commitment that intervenes on the mechanism, and the outcome is the consequence that matters afterward.

Stating the diagnosis as a diagnosis is what allows it to be found wrong, because a diagnosis can be refuted and a list of observations cannot. A “root causes” slide that lists observations, a hoped-for outcome, and a proposed tool is the mash formatted as a diagnosis, and it survives meetings because nobody can summarize it as one thought, and therefore nobody can reject it.

The same mash under other names

The translation is always the same: which of the three is each part of the request, and where is the mechanism that connects them?

Forecast accuracy. “Error is unacceptable, improve it.” The problem may be late orders or excess stock; the decision is what the ordering or allocation rule does with the forecast. Accuracy is a diagnostic of an input until the rule that consumes it is named.

Utilization. “Utilization is too low, raise it.” The decision is fleet size, dispatch, or appointment spacing, and maximizing utilization often worsens the service the company cares about.

A classifier score. “AUC is 0.81, can we get it to 0.85?” The decision is a threshold in a named process, and the outcome is the cost of the errors on each side of it.

An agent. “We need an agent for scheduling.” The decision is who may change a sequence, with what override, at what cost of a wrong change. The agent is at most evidence or a component of the rule.

A request of the form “we need X,” where X is a tool or a metric, that will not translate does not yet describe work. It describes a purchase.

Why the split helps

It makes the argument tractable. A dispute over whether inventory is “too high” concerns the problem, a dispute over finance’s cut concerns the alternatives, and a dispute over whether fill is the right measure concerns outcomes. Conducted together they let two functions sound as if they are arguing about virtue when they are arguing about a rule.

It shows what not to model. A precise measurement of the problem is unnecessary if the alternatives are insensitive to it, and a sharper outcome metric is unnecessary if it will not change the pick.

It reveals when there is no decision at all. Some briefs, split properly, are an observation and a wish with nothing in between. That is a finding worth delivering, and it takes fifteen minutes rather than six weeks.

The test

Take the brief on your desk and label every clause problem, decision, outcome, observation, diagnosis, or deliverable. Then write the one-line mechanism from the decision to the outcome. If the middle needs a miracle, or the only decision in the brief is a tool, you have found the work: name the commitment before modeling anything. When someone insists the problem must be understood first, spend a day, enough to write a diagnosis that could be killed, and then name the decision.