10 min read

Decision Matrices: Turning Information Into Choices

The information-to-decision gap

Research produces information. Information doesn't make decisions. People do — and they usually do it by intuition because the leap from "here's what I know" to "here's what we should do" is genuinely hard.

Decision matrices close that gap. They force you to be explicit about: the options, the criteria that matter, how much each criterion matters, and the evidence behind each rating. AI is excellent at filling these in, once you set the structure.

The decision matrix pattern

Four steps:

List the options — usually 2-4. Too many options is itself a sign you haven't narrowed enough.

Define the criteria — 3-6 things that genuinely matter. Be specific. "Quality" isn't a criterion; "average defect rate per 1,000 units" is.

Weight the criteria — assign each a weight (1-3, or 1-5). What matters most to you matters most. Don't fake balance.

Score each option on each criterion — with evidence. This is where AI is most useful — pull together the evidence under each cell.

Multiply weight × score to get a weighted total for each option. The highest total is your starting recommendation. But the value isn't the number — it's that you can now see why one option scores higher.

For quick 2-option comparisons, you don't need the full matrix. Give AI the two options + your criteria + an explicit order to commit: "Based on these criteria, which one wins, and what's the single strongest argument for the other?" Without that ask, AI hedges and gives you a balanced-sounding table where both options have "strengths." With it, you get a defensible recommendation.

A real example

I'm choosing between two analytics vendors for our team. Build a decision matrix.

Options: Vendor A (Mixpanel, $24K/yr, strong in product analytics, dedicated CSM) vs. Vendor B (Amplitude, $30K/yr, broader feature set, self-serve support only).

Criteria (with my weight 1-5): cost (3), feature depth (4), ease of adoption for non-technical users (5), quality of support (4), risk (2).

Fill the matrix using what you know about typical products in this category. For any criterion where you can't compare them well from the info I gave you, mark it "need more info" and tell me what specific question I should ask on our next vendor call.

Notice what this prompt does: it doesn't ask AI to pick the vendor. It asks AI to organize the comparison so the human can decide.

Always include a "what would change this answer" question. After the matrix, ask AI: "If one piece of information were different, what would flip the recommendation?" That's where you find your blind spots — the assumptions doing the most work in your reasoning.

Quick Check

You're using a decision matrix to compare three vendors. AI fills it in cleanly and the highest score is Vendor B. What's the BEST next move?