Why the Incumbent Wins

A team builds a replenishment rule that beats the current one by a wide margin in a twelve-month backtest. It presents well, the owner agrees it is better, and six months later the old rule is still running, sometimes underneath the new one.

Technical people find this puzzling, then frustrating, then, if they stay long enough, unsurprising. The incumbent does not need to be good to win. It is already running, its costs are familiar, the people who live with its failures have learned to work around them, and nobody has to sign anything to keep it. None of this is evidence that operators are irrational. These are regular, well-studied features of how people weigh change and how organizations assign credit and blame. The research is cited at the foot of this entry; what follows is what each force means for a challenger.

Six forces, each with a design answer

Defaults do most of the deciding. A planner reviewing two hundred proposals before a cutoff accepts most as shown. Accepting is fast, transfers responsibility to the system, carries an implied endorsement, and persists across people who inherit the screen. Whoever set the default made most of the decisions. A challenger delivered as a report beside the screen competes with the default on every line and loses most of them to effort alone. A challenger that becomes the number the screen proposes wins them without persuading anyone. Design answer: take the default.

Losses weigh more than gains, and the incumbent is the zero. Every challenger is perceived as a bundle of things gained and things lost relative to the current state, while the incumbent is perceived as neither. A planner’s pad is something she holds, and she will demand more to give it up than she would have paid to get it. A challenger can show a positive net balance on a regional ledger and still be a loss to each person it touches, because the losses are concentrated on them and the gains spread across others. Design answer: state what each party keeps, compensate directly for what they give up, and price the losses rather than netting them away.

The status quo needs no defender. A challenger needs a sponsor, a business case, a release window, a training plan, and someone willing to be blamed. The incumbent needs none of these. Harm from action is also judged more harshly than equal harm from inaction: a bad week under the old rule is blamed on circumstances, and the same week after a change is blamed on the change and its sponsor. Early failures under a new rule get attributed to it whether or not it caused them. Design answer: make the comparison symmetric on purpose, and agree before launch what will be measured, against what, over what period, and what would stop it.

Incentives pay for the pad. The people who run the incumbent are usually paid, one way or another, to keep it. When being short is discussed by name and costs a bonus while being long is a small move in a ranking, padding is the rational response, and the size of the pad is roughly what minimizes the operator’s expected personal cost rather than the company’s. The scorecard is part of the rule. A challenger that changes the logic and leaves the scorecard intact asks people to optimize one objective while being paid on another, and the usual result is the old pad applied quietly on top of the new rule. Design answer: bring the scorecard’s owner into the decision, and change what is rewarded or visible rather than exhorting.

Today beats later. The costs of a change are immediate, concrete, and concentrated on the people who run it: reconfiguration, retraining, explaining the change, the first visible miss. The benefits accrue slowly and diffusely, often to someone else’s budget. Deferral is reasonable in any given week, and a series of reasonable deferrals adds up to a permanent choice of the incumbent that nobody made. Design answer: change the shape of the timing. Do the switching work for the operators, make early benefit visible at a few sites in their own measures, and start in a quiet period rather than against a freeze.

Ownership is diffuse. The person who asked is rarely the person who can change the rule, the person who can change it rarely feels its cost, and the people who feel the cost usually cannot change it. Separate four roles: who requested the work, who owns each part of the rule, who can veto or undo a change, and who runs it on the floor. Persuade only the requester and you get a well-received report. Skip the veto holders and the change is reversed at the first opportunity. Skip the operators and the rule is formally installed and informally bypassed. Design answer: map the roles before ranking anything.

Meridian’s class-based rule, seen from the floor

From the network’s point of view, the class-based rule trades cash doing little in slow parts for service where service matters. Seen through the six forces, its position is weaker.

The uniform min/max holds the default on the planning screen, and planners accept most of its proposals. To dealers, thinner stock on slow parts is a loss of something they could count on, while better availability of water pumps is a gain noticed only on the day a pump is needed. To planners, the classes replace their judgment on exactly the parts they pad, a loss of discretion even if the classes protect the same parts. Nobody has been disciplined for the uniform rule, though it has stranded vehicles for years; the first vehicle stranded under classes will have a name attached. Planners are scored on fill and on complaints that reach their manager, and rarely charged for the cash a pad ties up. The first month is almost all cost, and the working capital comes back over the following year. And ownership is split: Parts Planning owns the targets, nobody owns the pads, finance owns the cash target, and the dealer habit of calling a center manager belongs to no one.

The challenger that can win is a different alternative from the one in the backtest. It becomes the quantity the screen proposes, explicitly protects the parts planners already protect and says so in their terms, is accompanied by a change in how planners are scored, runs first at two centers, and carries a stop condition agreed before launch. That version has different consequences from a class-based rule delivered as a report, and the ranking should compare the alternative that would actually be installed.

The failure mode is predicting a race without the head start

The dominant failure is analysis that does only the first of a challenger’s two jobs. It shows the challenger is better on the merits, measured against the incumbent as it actually runs, and stops. The second job, designing the challenger so the forces above stop working against it, is treated as change management to be handled later by someone else. It is not separate. Each design answer changes the challenger itself, and therefore its consequences and its rank.

Sometimes the head start is deserved. An incumbent that survives once its undocumented knowledge and the switching costs are priced may simply be better, and a challenger whose advantage disappears under that pricing is not worth installing. Telling the two cases apart is part of the work. An incumbent that wins because of the forces calls for a better-designed challenger; one that wins because it is good calls for a recommendation to keep it, written as plainly as any other.

The test

Take the challenger you are about to recommend and answer six questions in a sentence each. Will it be the default on the screen, or a report beside it? What does each affected party lose, and what do they keep? What comparison and stop condition have been agreed before it runs? Which scorecard rewards the behavior it replaces, and is that scorecard’s owner in the room? Who bears the first month’s cost, and what will they see early? Who requested it, who owns each part of the rule, who can undo it, and who will run it?

Any question you cannot answer is a force still working for the incumbent. If the answer to the last one shows that nobody with standing feels the incumbent’s cost, the incumbent will win regardless of the analysis, and the most useful thing to do is say so to someone who can change that.