Invent the Alternatives

Most people fill the alternatives slot with whatever was in the email, plus the status quo, and call the object complete. It is not.

This is the step technical people skip, because it does not look like analysis. Generating alternatives looks like sitting still, and sitting still feels like not working. So the week starts with the two options the requester named and a third that occurred to you in the shower, and that is the menu for the rest of the project. Six weeks later somebody in the meeting names a fourth that should have been obvious in hour one, and the ranking is about the wrong object again.

The quality of a decision is usually more sensitive to the quality of the menu than to the quality of the mathematics that ranks it. That is an uncomfortable claim for people whose skill is in the ranking.

The incoming list is a first offer

Treat it the way you already treat the incoming brief. The argument the organization is already having is useful information, because it tells you which alternatives will be in the room whether you write them down or not. It is not useful as a ceiling. Organizations get stuck precisely because they are arguing inside a menu that cannot contain a good answer, and ranking harder inside that menu produces a more elaborate version of the same fight.

The test of whether you added enough is whether the menu now contains at least one commitment that nobody in the original request had words for.

Four habits that enlarge a menu

Relax a constraint that is not physics. Lead time is physics. “We have always used one service target” is not. “Classes are too political” is not. “Finance needs a single number for a slide” is not. You are not required to recommend the unconstrained version. You are required to put it on the list long enough to see whether it is better. If it is, the constraint was never a constraint; it was a failure of imagination wearing a history.

Look at what good operators already do when the official rule is wrong. The planner who quietly pads one item class and starves another is running an unofficial rule. Writing it down as a candidate stops you proposing as novel something the field has been doing in the dark, and it gives you a cheap existence proof that the organization can already differentiate. Alternatives already being run badly are usually better candidates than alternatives that exist only on paper.

Unbundle. Requests glue together things that need not travel as a set. “Cut inventory 20 percent and protect service” is two decisions wearing one sentence. “Roll it out everywhere on Friday” is a rule plus a scope plus a date. A rule at two sites is a different alternative from a rule everywhere, and a cash event this quarter is a different alternative from a policy still running in September.

Take each option to an extreme, then mix. What would you do if cash were the only thing that mattered? If stranded jobs were? Neither extreme is a recommendation. Both are ingredients, and the interesting alternatives are usually combinations the original list did not contain. Skipping this produces menus of mild versions of the same idea, which cannot be ranked because they barely differ.

Then kill most of them on purpose

A large menu is not a virtue. A large menu you then work through in full is the default week with more columns.

Generate widely so that you can screen cheaply. Screen on dominance, meaning an alternative that is worse on every interest than another one, and on what the owner can actually commit to this cycle. An option that cannot be installed before the date that matters is not competing with the others; it is competing with a different and worse alternative called “do the urgent thing now and maybe this later.”

One discipline holds the whole step together: activities are not alternatives. “Build a dashboard,” “run a study,” and “improve the forecast” are things to do, not commitments among which someone chooses. A menu of activities cannot be ranked, because none of its entries settles anything.

The deliverable

Six commitments, written on one page, in the same grammatical form, before any model exists. You will not analyze all of them. You have to see all of them before you know which ones to kill.