Strategic Option Weigher

The capstone of the AI Lab. Describe a strategic decision in plain prose — build vs buy, market entry, vendor consolidation, M&A. Unlike every prior demo, no structured criteria are given; the model designs the analytical framework itself, weights each criterion with rationale, scores each option, runs sensitivity analysis on which assumptions, if changed, would flip the recommendation, and returns a confidence-bounded recommendation with explicit upper and lower outcomes. The first demo where the model is the strategy consultant, not the spreadsheet.

// STRATEGIC DECISION · prose, with options & context 0 chars

paste at least 400 chars or pick a preset

Architecture — what just happened

Browser
  └─→ POST /api/lab/chat
        - system: option-weighing prompt with the schema
                  below (versioned: option.v1)
        - user: the decision prose, verbatim
        - temperature: 0.3 (slightly higher — the criteria
                  framework benefits from some divergent
                  thinking; the math stays stable)
        - max_tokens: 3200
        ← single response, parsed as JSON: {
            decision_summary, category, decision_horizon,
            options[], criteria[], scores[], totals[],
            sensitivity[], recommendation
          }
        ← 6 rendered sections in executive read order:
          summary → recommendation → ranked options →
          DERIVED criteria framework → scored matrix →
          sensitivity analysis

One LLM call. The architectural twist: the model designs the criteria framework rather than receiving it. Each criterion comes with a weight (1-5) AND a rationale for that weight, surfaced in the criteria section so the executive can sanity-check the lens before trusting the scores. The sensitivity field is the consultant's discipline: which assumptions, if they shift, would change the answer? — ranked by criticality. The recommendation includes both a lower-bound outcome (what's the worst case if you pick this option?) and an upper-bound outcome (best case), plus the single most decisive factor.

Why this matters for executives and consultants

Every strategy team I've worked with builds the same artifact for every major decision: a framework, a weighted score-card, an option ranking, a sensitivity analysis, a recommendation memo. McKinsey, BCG, Bain — their entire consulting playbook is built around standardizing the production of these artifacts. A senior associate spends two weeks on this for each engagement.

This demo collapses the first draft. The model designs the framework, runs the math, drafts the sensitivity, and writes the recommendation. The executive team still does the deciding — but they start from a populated artifact, not a blank one. If the framework is wrong, you know within 30 seconds because the criteria are right there at the top with rationales. Edit the prose, re-run.

For internal strategy teams, this is the new "blank PowerPoint" replacement. For board members, it's a way to interrogate a recommendation against an alternative framing. For consultants, it's a tool for accelerating the first 60% of an engagement.

Honest caveat: the model is a fluent generalist, not a domain expert. It can produce a polished framework that misses the one criterion that actually matters in your industry. The sensitivity analysis is the model's read of which assumptions are critical — but the model doesn't know your private context (the COO who hates Salesforce, the board commitment that wasn't in the prose). Treat the output as a fast first draft a smart generalist consultant would produce; use the framework to surface disagreements, not to settle them. The confidence_band is the model's honest assessment of how clean the answer is.

Telemetry — request, response, parsed structure

Run the demo to see telemetry.