Decision Support Patterns

You have a decision in front of you right now — whether to buy or build the integration, whether to greenlight the redesign this sprint or next, whether the vendor contract is worth signing. You could talk it through with a colleague, sleep on it, or just go with your gut. All three work sometimes. None of them scale, and none of them leave a trail you can point to later when someone asks "why did we decide that?"

The four patterns in this lesson exist to make the reasoning behind a decision explicit — not to make the decision for you. AI is not the decision-maker here. It's the thing that forces your own reasoning into the open where you can inspect it, before you're standing in front of your boss defending a call you made on vibes.

Pattern 1: The Options Matrix (Weighted Tradeoff Table)

Instead of asking "should we buy or build this?" and getting back a paragraph of hedged prose, force the comparison into a table: list your options as rows (or columns), list the criteria that actually matter as the other axis, assign each criterion a weight, and have the AI score every option against every criterion. Multiply score by weight, sum per option, rank the totals.

The prompt shape:

Options: Buy a vendor tool / Build in-house / Hybrid (buy core, build integration)
Criteria and weights: Speed to ship (40%), Team buy-in (30%), Long-term cost (30%)
For each option, score 1-5 on each criterion, show the weighted contribution,
and sum to a weighted total. Do not average the criteria — respect the weights.

When it works: any decision with 2-4 real options and criteria you can name and rank by importance. It's especially good at surfacing decisions where your gut says "obviously A" but the weighted math says otherwise — that gap is worth investigating before you commit.

When it fails: when the weights themselves are the real disagreement, and the scoring just launders a pre-existing bias. If you set "speed to ship" at 40% because you already wanted to buy the vendor tool, the matrix will dutifully confirm what you already believed and dress it up as rigor. The matrix only earns its keep if the weights get argued about honestly, ideally by more than one person, before the scores go in.

Pattern 2: The Premortem

Instead of asking "what could go wrong," which tends to produce vague hedging, ask the AI to assume the decision has already failed — six months from now, this thing you're about to commit to has visibly gone wrong — and work backward to the specific, plausible causes. Failure-first framing produces sharper answers than risk-first framing, because "imagine you're already living the failure" bypasses the instinct to hedge every risk with "but it probably won't happen."

The prompt shape:

It is six months from now. We shipped [the decision] and it failed clearly enough
that leadership is asking what happened. Write the retrospective: what specifically
went wrong, in the order the failure likely unfolded. Then, for each cause, name
one concrete mitigation we could put in place BEFORE we ship, not after.

When it works: before you commit to anything with real cost to reverse — a hire, a vendor contract, an architecture choice, a launch date. Running it before the decision, not after the postmortem you're already dreading, is the entire point.

When it fails: if you stop at "imagine failure" and skip working backward to causes. A premortem that produces "it might just not work out" has generated zero mitigations, because there's nothing specific to mitigate. The exercise only pays off if the causes are concrete enough that step 4 — mitigation — writes itself.

Pattern 3: The Reversibility Check

Not every decision deserves the same amount of process. Jeff Bezos's Type 1 / Type 2 framing is the cleanest version of this: Type 1 decisions are one-way doors — hard, slow, or expensive to reverse (a senior hire, a platform migration, a public commitment). Type 2 decisions are two-way doors — cheap and fast to reverse if you're wrong (a UI copy change, a two-week experiment, an internal tool choice). The pattern is simple: before you spend a week gathering consensus, ask the AI to classify the decision and justify the classification.

The prompt shape:

Decision: [describe it]
Classify this as Type 1 (hard/costly to reverse) or Type 2 (cheap/fast to reverse),
and justify the classification in 2-3 sentences: what would it actually cost in
time and money to undo this if it turns out to be wrong?

When it works: as a triage step before you decide how much process a decision deserves. It's a fast way to catch yourself over-processing a decision that's actually trivial to reverse — or worse, rushing a decision that's genuinely a one-way door.

When it fails: when the classification itself is contested and the AI's confident-sounding justification papers over real uncertainty. Some decisions are structurally hard to classify — a hire looks Type 2 (you can always let someone go) until you account for the morale and reputational cost of doing so, which makes it closer to Type 1. Treat the AI's classification as a first draft to argue with, not a verdict.

Pattern 4: Second-Order Effects

The first-order effect of a decision is usually obvious — that's why you're considering it. The second-order effect is what happens after the obvious effect plays out, and it's where most of the surprises live. Ask explicitly for one or two hops past the first effect, not more.

The prompt shape:

Decision: [describe it]
First-order effect: [the obvious, intended outcome]
What is the second-order effect — what happens AFTER the first-order effect plays
out, that we would not have predicted just from the first-order effect alone?
Limit to effects within 1-2 hops. Do not speculate beyond that.

When it works: for decisions where the obvious win might create a less-obvious cost elsewhere — the classic example being a feature that boosts one metric while quietly degrading another one nobody was watching. It's a fast way to ask "what am I not tracking that this touches?"

When it fails: when you let it run past two hops. Chase effects far enough out and you stop reasoning about consequences and start inventing scenarios — speculative fiction that feels rigorous because it's structured, but has drifted so far from the original decision that it's no longer useful for deciding anything.

If you want the deeper mechanics of why structured, decomposed prompts outperform single-shot "what should I do" questions — the four-component prompt model, why explicit criteria beat implicit ones — that's covered in Prompt Engineering Mastery. This lesson assumes you can already write a clean prompt; it's handing you the four shapes worth reaching for on a real decision.

Bottom Line

Four patterns, one shared mechanism: turn an implicit judgment call into an explicit structure you can inspect, argue with, and hand to someone else. The Options Matrix forces criteria and weights into the open. The Premortem forces you to name causes before they happen instead of explaining them after. The Reversibility Check right-sizes how much process a decision deserves. Second-Order Effects catches the cost that shows up one step after the win you were chasing. None of them make the decision for you — they make sure the decision you make is one you can defend.