AI Prompts for a Consulting Recommendation Matrix
A recommendation can look polished and still fall apart when a client asks, “Why did this option win?” A consulting recommendation matrix makes the decision logic visible, linking criteria, evidence, trade-offs, and a final point of view.
AI can speed up the structure and challenge weak reasoning. However, it can’t own the judgment call, validate a source, or resolve stakeholder disagreement for you.
Use the prompts below to turn messy inputs into a defensible decision process.
Start with a decision question, not a blank table
A matrix is only useful when it answers one decision. “Choose the best CRM” is too broad. “Select a CRM for a 60-person sales team that needs Salesforce migration, regional compliance, and launch within six months” gives the work a boundary.
Before asking AI for a table, write down the recommendation’s purpose, decision owner, options, criteria, available evidence, and non-negotiable constraints. This brief prevents a model from filling gaps with plausible but unsupported assumptions.
Separate gates from weighted criteria
Some conditions are pass-or-fail gates. A supplier may need a security certification, a market may require a local license, or a software platform may need a specific integration. Don’t bury those requirements inside a weighted score.
Weighted criteria distinguish between acceptable choices. Cost, expected value, implementation effort, strategic fit, and risk often belong here. First remove options that fail a gate, then score the remaining choices.
Use weights to reveal priorities
Weights should reflect the decision owner’s priorities, not the loudest voice in the room. Make them total 100 percent, and define what each score means before anyone rates an option.
This illustrative software-selection matrix shows the basic format:
| Criterion | Weight | Option A | Option B | Option C |
|---|---|---|---|---|
| Strategic fit | 30% | 4 | 5 | 3 |
| Total cost | 25% | 5 | 3 | 4 |
| Implementation speed | 20% | 4 | 2 | 5 |
| Integration quality | 25% | 3 | 5 | 3 |
A high score without a written rationale is fragile. Attach a source, calculation, interview finding, or documented assumption to every score that can change the outcome.
Prompt to build a consulting recommendation matrix
Use this prompt when you have a decision brief but need a clean first draft of the matrix. It produces a structure for review, not a final recommendation.
Act as a strategy consultant preparing a decision matrix.
Business context: [company, market, team, and current situation]
Decision to make: [single decision question]
Decision owner: [name or role]
Options to compare: [option 1, option 2, option 3]
Must-pass constraints: [requirements that eliminate an option]
Evaluation criteria: [criteria and definitions]
Proposed weights: [criteria with percentages that total 100]
Available evidence: [links, research notes, costs, interview findings, assumptions]
Time horizon: [period]Build a consulting recommendation matrix. First, flag options that fail a must-pass constraint. Then create a weighted scoring table using a 1 to 5 scale. Define the scale for each criterion, show weighted-score calculations, and identify evidence gaps. Do not invent facts or citations. Label any inference as an assumption. End with three questions the decision owner must answer before approving the matrix.
Provide real options and plain-language criteria. If a criterion means different things to different stakeholders, define it before generating scores. For example, “scalability” may mean user capacity, geographic expansion, operating capacity, or all three.
Clear sections improve outputs across major AI assistants. OpenAI’s prompt-engineering guidance and Anthropic’s advice on organizing context into distinct sections support the same discipline: separate facts, instructions, and requested output.
Prompt to score options with evidence
AI is useful for normalizing evidence that arrives in different formats, such as interview notes, vendor documents, cost models, and market research. It should not convert weak evidence into confident scores.
Require an evidence trail for every rating
This prompt forces the model to show its work. Give it your source material or concise extracts, with dates and authors where available. Redact sensitive client information before using any external service.
You are reviewing evidence for a recommendation matrix.
Decision: [decision question]
Options: [options]
Scoring criteria and weights: [criteria, definitions, and weights]
Evidence pack: [source excerpts, links, notes, data, and assumptions]
Scoring scale: [define 1 through 5]
Constraints: [budget, timing, legal, operational, or strategic limits]For each option and criterion, provide: the proposed score, a one-sentence rationale, the exact evidence supplied that supports it, the evidence quality (high, medium, or low), and any missing information. Keep facts, estimates, and assumptions in separate columns. Do not use outside facts. If the evidence does not support a score, write “insufficient evidence” rather than guessing. Return the result as a Markdown table followed by the five most decision-relevant evidence gaps.
Check every quoted source against the original. A model can misread a caveat, mistake a forecast for an actual result, or blend two sources. Source validation remains a consultant responsibility.
A score should be easy to challenge. If nobody can trace it to evidence or an explicit assumption, it doesn’t belong in the final matrix.
Prompt to test the recommendation under pressure
A weighted total is a starting point. It is not a decision by itself. Small weight changes can reverse the ranking, especially when options finish close together.
Run sensitivity analysis before presenting a winner
Use this prompt after you have a completed scoring table. It identifies the assumptions that matter most and gives you a less brittle recommendation.
Review the following weighted decision matrix.
Decision context: [context]
Options and scores: [paste scored matrix]
Current weights: [paste weights]
Non-negotiable constraints: [paste constraints]
Known uncertainties: [missing data, disputed assumptions, changing conditions]Test the recommendation’s sensitivity. Change each criterion weight by plus or minus [10 or 20] percentage points while keeping total weights at 100 percent. Identify ranking changes, score gaps, and criteria that drive the result. Also test these scenarios: [scenario 1], [scenario 2], and [scenario 3]. State whether the leading option is robust, conditional, or inconclusive. Show calculations in a table. Do not alter scores or introduce facts unless I provide them.
A matrix that stays stable under reasonable changes gives a stronger recommendation. If the winner changes easily, present the decision as conditional and state what additional evidence would resolve it.
Prompt to turn scores into an executive recommendation
Executives rarely want a spreadsheet read aloud. They need a clear recommendation, the trade-offs, the key risks, and the decision required.
Write a recommendation that doesn’t hide trade-offs
This prompt converts approved analysis into concise narrative. Feed it only reviewed scores and validated evidence. The resulting draft still needs your judgment on tone, context, and political realities.
Write an executive recommendation based only on the approved decision matrix below.
Audience: [CEO, investment committee, client sponsor, board, or other role]
Decision required: [decision]
Approved matrix: [paste table]
Evidence and assumptions: [paste validated notes]
Sensitivity-analysis result: [paste result]
Constraints and risks: [paste items]
Preferred output length: [for example, 250 words]Lead with the recommended option and the decision requested. Explain the two strongest reasons it ranks first. State the most material trade-off, the condition that could change the answer, and the next action. Use direct language. Do not claim certainty where the evidence is incomplete. Do not mention analysis that is not included above.
For a close decision, the right output may be “approve a pilot” rather than “select a winner.” That distinction protects the client from false precision.
Align stakeholders before the matrix becomes a verdict
A recommendation matrix can expose disagreement that meetings have hidden. One executive may value speed, while another values cost control. Those are competing priorities, not arithmetic mistakes.
Ask AI to surface conflicts, not settle them
Use this prompt with workshop notes, stakeholder interviews, or draft comments. It prepares a focused conversation, but it cannot decide whose priorities should prevail.
Analyze these stakeholder inputs for a pending recommendation.
Decision: [decision]
Stakeholder comments: [paste attributed notes]
Proposed criteria and weights: [paste matrix design]
Current recommendation: [option, if any]Identify agreements, disagreements, missing decision rights, and criteria that stakeholders define differently. Separate factual disputes from priority disputes. Suggest up to five workshop questions that would resolve the most important issues. Do not assign weights or choose a winner without explicit stakeholder direction.
Record who approved the criteria, weights, scores, and final recommendation. This decision log matters when conditions change or a client revisits the choice six months later.
Control the output format and preserve the audit trail
A strong consulting recommendation matrix is easy to update. Keep the raw evidence, scoring rules, calculations, stakeholder decisions, and version date together. Avoid a workbook where final numbers appear with no route back to their origin.
When your AI platform supports structured responses, request fixed fields for options, criteria, evidence references, scores, assumptions, and open questions. OpenAI describes how Structured Outputs can follow a JSON schema, while Claude documents schema-constrained structured outputs. Those controls reduce reformatting and make review easier.
A practical review sequence is simple:
- Confirm constraints and decision rights with stakeholders.
- Validate sources and approve scoring definitions.
- Test weights and scenarios before writing the recommendation.
- Document what would trigger a re-evaluation.
Make the Recommendation Defensible
AI can assemble tables, compare evidence, spot missing inputs, and draft a concise point of view. Consultant judgment determines whether the evidence is sound, the weights are fair, and the recommendation fits the client’s real constraints.
Use a consulting recommendation matrix to make trade-offs visible, not to disguise them behind a weighted total. When the assumptions are clear and the sensitivity test holds, the final recommendation can withstand a much harder conversation.