--- name: make-high-stakes-decisions title: "Make High Stakes Marketing Decisions" description: A decision making prompt that runs a premortem, checks for bias, and scores the options, giving you a marketing decision framework for your biggest bets. cluster: growth-strategy version: 1.1.0 --- # Make High Stakes Marketing Decisions This prompt turns a high stakes marketing decision into a structured audit before you spend the money. It surfaces the biases and emotions quietly steering the call, imagines the ways it fails, forces every plausible option through the same weighted scoreboard, and returns a defensible recommendation with the evidence you still need to gather. Use it to convert a gut call into a decision you can stand behind in front of a skeptical leadership team. ## When to use this → You are about to commit a large budget to a product launch, a new acquisition channel, a rebrand, or entry into a new market or segment. → The room already has a favored answer and you suspect conviction is outrunning evidence. → You need to compare two or more real alternatives on the same criteria instead of arguing about them one at a time. ## The prompt ```text You are a senior growth strategist and decision scientist advising me on a consequential marketing decision. Your job is not to cheerlead. Your job is to pressure test the decision: expose the biases and emotions steering it, imagine how it fails, score the real alternatives on shared criteria, and hand me a recommendation I can defend to a skeptical leadership team. CONTEXT - Decision under review: {{DECISION}} - Alternatives on the table (include "do nothing"): {{ALTERNATIVES}} - Budget and resources at stake: {{BUDGET_AND_RESOURCES}} - Reversibility (easy to undo, costly to undo, or one way door): {{REVERSIBILITY}} - Primary goal and the metric that would prove success: {{GOAL_AND_SUCCESS_METRIC}} - Guardrail metrics that must not degrade: {{GUARDRAIL_METRICS}} - Evidence we currently have (data, past results, research, customer voice): {{EVIDENCE}} - Who favors which option and why: {{STAKEHOLDER_POSITIONS}} - Deadline or window forcing the timing: {{DEADLINE}} Before analyzing, if any of these are missing or vague, ask me up to five clarifying questions and stop. Do not invent facts. In particular, refuse to proceed without a named success metric and at least one real alternative, because a decision with only one option is not a decision. Once you have enough, run this method in order. STEP 1. FRAME THE DECISION Restate the decision in one sentence: the choice, the alternatives, the target behavior or outcome, and the window. Classify reversibility. Flag if this is a one way door, because irreversibility raises the evidence bar. STEP 2. FOUR DRIVER AUDIT For the favored option (if the room is split, audit the top two contenders), examine the four forces behind any human decision and name the specific risk in each: - Bias: which prior belief or predictable distortion is steering evaluation? Screen explicitly for confirmation bias (noticing only supporting evidence), belief bias (prior conviction outweighing current evidence), omission bias (dropping a risky option or fact rather than weighing its real cost), and sunk cost (past spend justifying more spend). - Memory: which past win or loss feels similar and is being pattern matched onto this situation, correctly or not? - Reason: what is the best current explanation for why this will work, and which single piece of evidence, if we saw it, would falsify it? - Emotion: what approach or avoidance pull (fear of missing out, fear of looking timid, excitement about a shiny channel) is supplying the preference? Treat these as overlapping, not separate. A remembered win can create excitement, reinforce confirmation bias, and then get rationalized as pure logic. STEP 3. PRE MORTEM Assume it is twelve months from now and this decision failed badly. Write the obituary. List the six to ten most likely failure modes, each as a concrete causal story (what broke, in what order, and why). For each, tag it: assumption risk, execution risk, market or timing risk, or measurement risk. For each, state the earliest observable signal that it is happening and a mitigation. STEP 4. WYSIATI CHECK For each alternative, write the conclusion a decision maker would need to reach to pick it, list the facts required to reach that conclusion, and mark which of those facts we actually have versus are assuming. Highlight where a favorable case rests on inputs that are not in evidence. STEP 5. WEIGHTED SCORING Score every alternative (including do nothing) on these five criteria using a 1 to 5 ordinal scale, and show the component scores, never just the total: - Expected impact on the success metric - Strength of evidence behind the assumptions - Feasibility (budget, skills, traffic, and time available) - Learning value (what a result teaches us for future decisions) - Risk (score the danger to guardrail metrics and the cost of being wrong) Compute a priority score as: Priority = (impact x evidence x feasibility x learning value) / risk. Multiplying the upside criteria means one weak dimension drags the whole option down; dividing by risk means real danger to the guardrails cuts the score instead of nibbling at it. Present the results as a table, ranked. State plainly that this is an operating aid to structure judgment, not a precise formula. STEP 6. RECOMMENDATION Give one recommendation. State the strongest argument against it and why you still recommend it. If the evidence is too thin to commit, say so and design the smallest, fastest test that would resolve the main uncertainty before the full spend, with a clear decision rule for what result means go, kill, or iterate. STEP 7. FACT CHECK LIST Close with every load bearing fact or assumption the recommendation depends on that is not yet verified by the evidence I gave you. Rank them by how much the decision would change if they turned out false, and for each name the fastest, cheapest way to verify it before the money is committed. OUTPUT FORMAT 1. Decision frame and reversibility (one paragraph) 2. Four driver audit (four labeled bullets) 3. Pre mortem (numbered failure modes with signal and mitigation) 4. WYSIATI gaps (what each option assumes but has not proven) 5. Scoring table, ranked, with component scores 6. Recommendation, the case against it, and the next test if evidence is thin 7. Fact check list: unverified load bearing assumptions, ranked, each with the fastest way to verify it SELF CHECK before you finish: - Verify every claimed fact traces to the evidence I gave you. Label anything you assumed or inferred as an assumption, not a fact. - Confirm you scored do nothing honestly rather than dismissing it. - Confirm the recommendation names its strongest counterargument. Avoid these failure modes: rationalizing the favored option, treating stated reasons as the only evidence, scoring only the total and hiding the components, declaring the matter settled when a cheap test would resolve the core doubt. ``` ## Prompt versions The standard prompt above works on any model. Use these variants when you want a different tradeoff. ### Frontier model version Built for the most capable models (Claude Opus and beyond). States the goal, constraints, and quality bar up front, then trusts the model to choose its path. ```text You are a senior growth strategist and decision scientist. Pressure test a consequential marketing decision, do not cheerlead it, and hand me a recommendation I can defend to a skeptical leadership team. Deliverable: a defensible verdict on the decision, backed by a bias and emotion audit, a premortem, a ranked scoring of every option, and the unverified assumptions I still need to check. CONTEXT - Decision under review: {{DECISION}} - Alternatives, including do nothing: {{ALTERNATIVES}} - Budget and resources at stake: {{BUDGET_AND_RESOURCES}} - Reversibility: {{REVERSIBILITY}} - Goal and the metric that proves success: {{GOAL_AND_SUCCESS_METRIC}} - Guardrail metrics that must not degrade: {{GUARDRAIL_METRICS}} - Evidence we hold today: {{EVIDENCE}} - Who favors which option and why: {{STAKEHOLDER_POSITIONS}} - Deadline forcing the timing: {{DEADLINE}} Decision rules, non negotiable: - Refuse to proceed without a named success metric and at least one real alternative; a decision with one option is not a decision. - Score do nothing as honestly as every other option. - A one way door raises the evidence bar; weigh irreversibility explicitly. - For the favored option, name the specific bias, remembered pattern, reasoning, and emotional pull steering it, and the single piece of evidence that would falsify the case for it. - Rank every option on impact, evidence strength, feasibility, learning value, and risk, showing component scores, never just a total. Danger to guardrails must cut a score, not nibble at it. The output is excellent when every fact traces to the evidence I gave you (anything inferred is labeled an assumption), the premortem reads like a plausible obituary with early signals and mitigations, the recommendation names its strongest counterargument and still holds, and, where evidence is thin, it designs the smallest test with an explicit go, kill, or iterate rule rather than faking confidence. Close with the load bearing assumptions the recommendation leans on, ranked by how much the decision would change if they were false, each with the fastest way to verify it. Lead with the verdict on the first line, then the supporting audit, scoring, and assumptions. Do not invent data or statistics, do not pad with generic advice, and if a required input is missing or vague, ask one focused question and stop rather than guessing. ``` ### Quick version Five lines or fewer, for when speed matters more than rigor. ```text Pressure test this marketing decision and give me a verdict I can defend. Decision: {{DECISION}}. Alternatives, including do nothing: {{ALTERNATIVES}}. Success metric: {{GOAL_AND_SUCCESS_METRIC}}. Evidence I hold: {{EVIDENCE}}. Name the bias steering the favored option, the top failure mode, and rank every option; if the evidence is thin, propose the smallest test instead of a verdict. ``` ## How to customize → `{{DECISION}}`: the one line choice under review, for example "launch the new tier in Q3" or "make TikTok our primary paid channel next year." → `{{ALTERNATIVES}}`: the other real options, always including do nothing, for example "launch in Q3 / delay to Q1 / ship a beta to 5 percent first." → `{{BUDGET_AND_RESOURCES}}`: money, headcount, and time on the line, for example "$180k media plus two full time marketers for a quarter." → `{{REVERSIBILITY}}`: whether the call is easy to undo, costly to undo, or a one way door such as a rebrand or a market entry with signed contracts. → `{{GOAL_AND_SUCCESS_METRIC}}`: the outcome and the number that proves it, for example "reach 300 paid signups at under $90 blended CAC by day 90." → `{{GUARDRAIL_METRICS}}`: what must not get worse, for example "existing tier churn, gross margin, brand sentiment." → `{{EVIDENCE}}`: the data, prior test results, research, and customer quotes you already hold. → `{{STAKEHOLDER_POSITIONS}}`: who wants what and their reasoning, which exposes where politics is standing in for evidence. → `{{DEADLINE}}`: the date or window forcing timing, so the model can judge whether urgency is real or manufactured. ## What good output looks like → The four driver audit names a specific bias, memory, and emotion at work, not a generic list. You should recognize your own reasoning in it. → The pre mortem reads like a plausible obituary with causal stories, early warning signals, and mitigations, not a vague list of things that could go wrong. → The scoring table shows all five component scores per option and ranks do nothing honestly against the rest, so you can see why the winner won. → Every fact traces back to the evidence you supplied; anything inferred is labeled as an assumption. → When evidence is thin, the output refuses to fake confidence and instead specifies the smallest test with an explicit go, kill, or iterate rule. → The closing fact check list ranks the unverified assumptions the recommendation leans on, each with the fastest way to verify it, so you leave with a concrete evidence gathering agenda rather than a verdict alone. ## Related prompts → [Pick and Test Growth Channels](./pick-and-test-growth-channels.md): once the decision is a channel bet, use this to rank and stage the channel tests. → [Map Growth Loops and Flywheels](./map-growth-loops-and-flywheels.md): model where the decision moves the numbers before you commit budget to it. → [Set a North Star Metric and Quarterly OKRs](./set-a-north-star-and-okrs.md): translate the approved decision into the quarter's objectives and the levers that carry it.