Amey-Thakur Prioritization frameworks
Frameworks do not make decisions; they structure arguments so the real disagreements (about reach, about confidence, about strategy) surface where they can be examined instead of hiding inside gut rankings.
Method
- Score with RICE/ICE as a conversation, not a calculator. Reach (users per quarter, from data: see product-metrics), Impact (effect per user, on a coarse scale), Confidence (grounded in discovery evidence: see product-discovery's evidence grades: low confidence is a flag to run a cheaper experiment first), Effort (team-weeks from the team, not from hope). The output ordering matters less than the fights the inputs provoke: those fights are the work.
- Add cost of delay for time-sensitive calls. What does a quarter's delay cost: compliance deadlines (cliff), competitive windows (slope), steady value foregone (linear)? CD3 (cost of delay / duration) favors short valuable work and exposes the "big important thing" that could ship as three small valuable things (see user-story-writing's slicing).
- Segment before scoring. Growth bets, retention fixes, platform/debt (see tech-debt-register), and compliance are different currencies: allocate capacity across buckets first (a strategy call: see roadmap-communication), then rank within buckets. One list mixing currencies always starves the long-term bucket.
- Respect the qualitative overrides, in writing. Strategic coherence (does this compound toward the vision?), enablement (unlocks other work), and pure conviction bets exist legitimately: record when a decision overrides the scores and why (see decision-journals); frequent overrides mean the scoring inputs are missing a real variable.
- Say no with the reason attached. Every rejection names its category: "not this quarter (capacity: here is where it sits)", "not unless the evidence changes (here is what would change it)", "never (off-strategy, and why)": see feature-sunsetting for the removal sibling. A visible parked-list with reasons converts repeat lobbying into evidence-gathering.
- Re-rank on rhythm, not on noise. Quarterly full pass (see quarterly-planning), lightweight monthly touch; mid-cycle insertions pay their way by displacing something named (the trade is the decision: see code-freeze-management's same discipline for releases). Loudness is not reach; the framework exists precisely so the loudest stakeholder's pet ranks like everything else (see stakeholder dynamics in exec-briefing).
Boundaries
- Scores are made of estimates with error bars; treat close rankings as ties broken by judgment, and beware precision theater (a 47.3 vs 46.1 RICE difference is noise wearing decimals; see estimation-techniques).
- Frameworks rank options already on the table; generating better options is discovery's job (see product-discovery), and no ranking rescues a weak option set.
- Team morale and craft investment resist scoring but decide long-run throughput; leaders budget for them explicitly rather than making them compete with features.