Le Constructeur de Matrice de Décision : Transformez une Décision Multi-Options en une Recommandation Pondérée et Défendable
Pourquoi ce prompt est important
Undocumented gut-call decisions get re-litigated the moment results are anything less than perfect, because there's no record of what was actually weighed. A written, weighted analysis with an explicit sensitivity check settles the argument the first time, and gives you a concrete artifact to revisit if the underlying assumptions change, instead of relying on someone's memory of a conversation that happened weeks earlier.
À quoi nous l'utilisons
You're facing a decision with three or more viable options — a vendor selection, an infrastructure migration path, a hiring choice between finalists, a build-versus-buy call — and need to present a recommendation that will hold up when a skeptical stakeholder asks why you didn't pick a different option.
Prompt
Role: Act as a senior decision-analysis consultant who helps executives and teams make high-stakes, multi-option decisions using structured, defensible frameworks — the kind of analysis that survives being challenged in a leadership meeting. Context: - The decision I'm facing: [DESCRIBE THE DECISION IN 1-2 SENTENCES] - The options I'm considering: [LIST 3-6 OPTIONS, ONE PER LINE] - The criteria that matter for this decision: [LIST 4-8 CRITERIA, e.g. upfront cost, time-to-value, team capacity required, reversibility, strategic fit] - Any hard constraints or dealbreakers: [LIST ANYTHING THAT AUTOMATICALLY DISQUALIFIES AN OPTION, OR WRITE "NONE"] - Who else needs to be convinced by this analysis: [E.G. "MY CO-FOUNDER," "THE BOARD," "MYSELF ONLY"] Task: 1. First, propose a weighting (out of 100 total points) across the criteria I listed, and explain your reasoning for each weight in one sentence. Pause here so I can adjust before you proceed. 2. Score each option against each criterion on a 1-10 scale, with a one-line justification for every score — no unexplained numbers. 3. Calculate the weighted total for each option. 4. Identify the top-scoring option and flag any options that are statistically too close to call (within 5% of each other). 5. Run a sensitivity check: identify which single criterion, if reweighted by plus or minus 15 points, would change the recommended option. State explicitly if this exists — a fragile recommendation should be labeled as such. 6. Write a final recommendation in plain language, including the strongest argument against your own recommendation and why you're making it anyway. Constraints: - Do not silently round close scores into a false sense of certainty — say explicitly when the decision is close. - Do not use vague criteria labels — operationalize each one in half a sentence before scoring. - If any option violates a stated hard constraint, exclude it from scoring entirely and explain why. - Keep the tone analytical and neutral — no cheerleading for any option. Output Format: 1. A markdown table: rows = options, columns = criteria plus weighted total, sorted highest to lowest. 2. A short "How I Weighted This" section (2-4 sentences). 3. A "How Close Is This" section flagging any near-ties or fragile results. 4. A final recommendation paragraph including the steelman case against it.
Résultat
How I weighted this
I weighted Time-to-Value highest (30 points) because your Friday deadline and board skepticism suggest speed-to-first-results matters more than long-term elegance right now. Risk of Outage (25 points) comes second since you're mid-fundraise and a customer-facing incident would be costly beyond the technical fix. Team Capacity Required (20 points) reflects your 12-person team's limited bandwidth. Reversibility (15 points) and Total Cost Over 12 Months (10 points) round it out — cost matters less than the other three given your current runway.
| Option | Time-to-Value (30) | Risk of Outage (25) | Team Capacity (20) | Reversibility (15) | Cost (10) | Weighted Total |
|---|---|---|---|---|---|---|
| Lift-and-shift | 9 — deployable in 2 weeks with minimal refactoring | 6 — moves existing fragility as-is, doesn't fix root causes | 8 — low engineering lift | 9 — easy to roll back to on-prem if needed | 5 — higher ongoing cloud spend from unoptimized resources | 78.5 |
| Containerize-and-migrate | 6 — 6-8 week timeline for containerization work | 8 — improves isolation and rollback granularity | 5 — requires Docker/K8s ramp-up for two engineers | 7 — moderately reversible, some rework needed | 7 — better resource utilization | 67.5 |
| Full re-architecture | 2 — 4-6 month timeline, misses your deadline entirely | 9 — addresses root causes directly | 2 — consumes most of the team's bandwidth for two quarters | 3 — hard to reverse once started | 8 — best long-term unit economics | 44.5 |
How close is this
This isn't close — lift-and-shift leads containerize-and-migrate by 11 points, well outside the 5 percent fragility threshold. The one criterion that could flip the recommendation is Time-to-Value: if you reweighted it down by 15 points (to 15) and shifted that weight to Risk of Outage, containerize-and-migrate would edge ahead by roughly 3 points. Worth flagging to your board member if they push back on speed as the top priority.
Recommendation
Go with lift-and-shift for this cycle. The strongest argument against this: it doesn't fix any of the architectural problems that got you here, and you'll likely be having this exact conversation again in 9-12 months once traffic grows. But given your Friday deadline and board skepticism, a working migration you can demo now is worth more than a better migration you can't show yet. Plan the containerize-and-migrate path as your Q2 follow-up, not a competing option for this decision.
La plupart des décisions à enjeux élevés impliquant plusieurs options sont prises à l'instinct déguisé en analyse — un dirigeant choisit l'option qui "semblait juste", puis construit une justification par ingénierie inverse si quelqu'un le demande. Cela fonctionne bien jusqu'à ce que la décision tourne mal, moment où il n'existe aucune trace de ce qui a réellement été pesé, et la conversation dégénère en "pourquoi n'avons-nous pas envisagé X" sans bonne réponse.
Pourquoi ce prompt est structuré ainsi
Le prompt demande au modèle de proposer d'abord les pondérations et de marquer une pause pour confirmation avant de noter quoi que ce soit. Cet ordre compte : si la pondération et la notation se produisent en une seule passe, il est facile pour un modèle (ou une personne) de reconstruire inconsciemment des pondérations produisant la réponse déjà souhaitée. Séparer les étapes force la logique de pondération à tenir seule, défendable indépendamment de l'option qui finira par gagner.
L'instruction de justifier chaque note individuelle en une phrase — pas seulement les totaux finaux — est la deuxième contrainte fondamentale. Une matrice pondérée avec des chiffres inexpliqués n'est pas plus rigoureuse qu'une décision à l'instinct ; c'est une décision à l'instinct déguisée en tableur. Forcer une justification d'une ligne par cellule rend le raisonnement auditable.
L'étape que la plupart des cadres omettent
La vérification de sensibilité — identifier quel critère unique, s'il était repondéré de ±15 points, inverserait la recommandation — est ce qui distingue ceci d'un modèle de notation statique. La plupart des matrices de décision présentent un instantané figé dans le temps et suggèrent une confiance illusoire. Une décision serrée qui basculerait avec une repondération modeste et défendable est fondamentalement différente d'une décision robuste sur une large gamme de pondérations raisonnables.
Là où ce prompt fait ses preuves
Ce n'est pas pour les décisions triviales — choisir un lieu pour déjeuner ne nécessite pas de matrice pondérée. Il est conçu pour la poignée de décisions à options multiples par trimestre qui coûtent cher en cas d'erreur et où un processus documenté compte autant que la réponse : sélection de fournisseurs, parcours de migration d'infrastructure, candidats concurrents pour un poste, décisions de construire ou d'acheter.