Prioritization
Ranking is easy and useless. The decision that matters is what gets dropped, and the artifact that matters is a reason someone can disagree with in three months.
When this fires
More candidates than capacity; a request to say what ships first or what to cut; an existing priority order nobody can justify. It does not fire for a list of two items with an obvious dependency between them — say the order in one line and move on.
Procedure
- Name the constraint you are allocating. One release, one sprint, one person for two weeks, one review slot. Priority without a stated capacity is just an opinion about importance, and it is why ranked lists get agreed and then ignored.
- Make the list comparable. Candidates at wildly different sizes cannot be ranked against each other. Split the boulders into the first shippable slice, and say explicitly that what is being compared is the slice, not the ambition behind it.
- Write the cost honestly, in the unit of the constraint. Your estimate, its basis, and its uncertainty as a range where you have one. Include the costs that hide outside the build: migration, support load, the thing that now needs maintaining, the review it will consume.
- Write the value claim with its evidence grade. What changes for whom, and whether that is observed (data or traced cases), reported (someone said so), or assumed. Two items claiming equal value where one is observed and one is assumed are not equal, and collapsing that difference is the most common way a bad bet wins.
- Check dependency and sequencing before any scoring. What is blocked by what, what unblocks several others, what has an external date attached. Order that is forced is not a priority judgment — settle it first so the real tradeoff is what remains.
- Sort by reversibility, not only by size. A cheap one-way door — a public API, a data migration, a pricing change, a URL scheme, anything users will build habits on — deserves more scrutiny than an expensive change you can undo on Friday. Say which candidates are one-way.
- Ask what changes if each waits one cycle. Some costs grow (a migration through a growing dataset, an outage risk, a security gap), some shrink, most stay flat. Decay is what separates "later" from "never", and it is usually the only honest argument for doing a low-value item now.
- Make the cut, and write the line under it. Say what is in, what is out, and what is out for now with the condition that would bring it back. A list where nothing is out has not prioritized anything.
- Record the reasoning in a few lines, without ceremony. What the constraint was, which tradeoff decided it, and what evidence would flip it. Reach for a scoring framework only when many comparable items must be ranked by several people; for one team and a dozen items, a score is decoration that hides its own weights. Numbers invented to look objective are worse than a sentence that can be argued with.
- Present it as a recommendation. The people who own the roadmap decide, and commitments to customers, dates, or other teams are theirs to make — offer the cut and the reasoning, and let them take it. Do not announce, message, or record the decision anywhere outward-facing on your own initiative.
Checklist
- The capacity being allocated is stated
- Items are comparable in size, or explicitly sliced to be
- Every item has a cost with its basis, in the constraint's unit
- Every value claim carries an evidence grade
- Forced ordering (dependencies, external dates) resolved before judgment calls
- One-way doors identified as such
- Decay checked: what gets worse or cheaper by waiting
- Something is actually out, and "out for now" has a return condition
- Reasoning written in a few lines; no scoring theatre
- Delivered as a recommendation, with the decision left where it belongs
Failure handling
- Two items are genuinely tied. Say so and break it on reversibility or on which one teaches you more. Manufacturing a tenth of a point to separate them is false precision.
- Someone has already promised an item externally. Then it is a constraint, not a candidate. Move it above the line, say it was a commitment rather than a ranking, and rank what is left.
- The evidence is thin everywhere. Prioritize the cheapest thing that produces evidence, and say that is what you are optimizing for. Do not dress up an assumption as a value score.
- The list is all urgent. Ask what breaks, for whom, and when, if each waits a cycle. Urgency that survives that question is real; the rest was volume.
- You are being asked to rank someone else's domain. Rank the effect on the outcome you can see, name the expertise you lack, and route the call. A confident ranking of work you do not understand is a liability.
- The decision has already been made and you are being asked for justification. Say what the reasoning would have to be for it to hold, and whether the evidence supports it. Do not reverse-engineer a rationale.
Evidence to report
The constraint; the candidate list with cost, value and evidence grade per item; the forced ordering and why; which items are one-way doors; what was cut and what was deferred with its return condition; and the one or two facts that would change the answer. A ranked list with no capacity, no costs and nothing below the line is a preference, not a prioritization.
