Testing Strategic Assumptions Before You Commit Budget

The full executive answer

Take the platform’s epistemics first, because it is the layer most AI products skip. Structurally, we distrust our own outputs: every numeric claim passes provenance validation — traced to a source in the input data or flagged — before it reaches you; the six flagship analyses are generated independently by two different AI providers and compared, and where they materially disagree — below a measured agreement threshold — a third model adjudicates and the disagreement itself is surfaced rather than smoothed over. That last point is the strategically important one: cross-model disagreement is information about assumption fragility. Where two independently trained models diverge on your competitive outlook, that divergence is a flag on exactly the assumption a red team should attack — the platform gives you Philip Tetlock-style epistemic humility as an artefact, not a virtue.

For your assumptions as a strategy officer: the diagnostic modules are built to force assumptions explicit — scorecard syntheses require the underlying judgments to be entered as structured inputs rather than buried in prose, the risk-generation module systematically converts strategic assumptions into monitorable risks, and because every analysis is versioned in an append-only record, you can do the thing almost no strategy function does: go back and audit how each assumption aged against reality. That is assumption-based planning, in the RAND sense, made routine because the bookkeeping is free.

And our own product assumptions, honestly scored: the assumption that governed AI architecture could be built and CI-enforced — validated; it is in the codebase with 3,500-plus tests. The assumption that unit economics survive at self-serve prices — validated by instrumentation, with the margin engine measuring it continuously. The assumption that boards will pay for governance before regulation forces them — not yet validated; that is precisely what launch tests, and I hold it as a hypothesis with a falsification plan, not as a belief.