T21

The claim challenges the simplicity of the scalar PMF concept ('do you have it or not?') by arguing for a vector decomposition. This is intellectually appealing but operationally risky: if sub-factors are not independently measurable, the vector decomposition is just a scalar with more labels, providing false precision. The risk of false precision in PMF assessment is high — companies have told themselves they have 'good PMF on retention but weak PMF on acquisition' when the real problem was undifferentiated positioning, not component-level PMF gaps.

PMF (Phi) as a scalar is: does the product generate the pull necessary for customers to seek it out, stick with it, and advocate for it? The scalar is correlated across dimensions but not measured precisely. A vector decomposition might include: (a) Phi_activation: do new users activate to a meaningful outcome within a defined window? (b) Phi_retention: does the cohort retain at benchmark rate by segment? (c) Phi_expansion: does the product naturally generate expansion revenue without sales pressure? (d) Phi_advocacy: do satisfied customers refer others? (e) Phi_positioning: does the market understand and agree with the value proposition? Each sub-factor is independently observable and can diverge: a product may have strong Phi_retention (churns slowly) but weak Phi_expansion (customers do not grow) and weak Phi_advocacy (do not refer). This decomposition generates different GTM interventions for each gap. But the critical condition: 'only if measured.' A vector with unmeasured components is epistemically worse than a scalar, because it introduces the illusion of precision while leaving the gaps unmeasured.

• Superhuman's 'How Superhuman Built an Engine to Find Product/Market Fit' (Rahul Vohra, 2018): decomposed PMF into segment-specific retention and advocacy signals, enabling targeted ICP refinement — a real-world vector PMF application
• Y Combinator batch analysis: companies that tracked activation rate, D30 retention, and NPS as separate PMF signals were better at diagnosing which component to fix, vs. companies with a single PMF score
• Sean Ellis test (NPS variant): the '% who would be very disappointed if your product disappeared' segmented by persona reveals vector-level PMF differences — different personas have dramatically different Phi levels

• For early-stage companies with limited instrumentation, tracking five PMF sub-factors creates measurement overhead that exceeds the signal gain
• PMF sub-factors are highly correlated for well-functioning products — decomposition adds value primarily for products with component-level gaps, which is a subset of the market
• The measurement gap register flags Phi_positioning and Phi_advocacy as particularly hard to measure with current standard tooling

Start with the scalar PMF signal (Sean Ellis test: % very disappointed, target > 40%; NPS in your ICP segment, target > 30; D90 cohort retention vs. benchmark). Only decompose into the vector when the scalar signals diverge across segments or when you have a specific hypothesis about which component is weak. The practical order of operations: (1) confirm scalar PMF is adequate; (2) if not, use the vector to diagnose which component to fix; (3) instrument that specific component with a direct measurement before claiming progress. Never use a PMF vector with unmeasured components in a board presentation — it will be mistaken for precision.