Methodology · Scoring
How the AUG score is computed
Full transparency on the math. The composite score isn't opaque magic — it's deterministic, reproducible, and inspectable. Anyone with the per-factor scores can recompute the composite themselves.
The composite formula
AUG = 100 × Acq × Act × Eng × Ret × Adv × Mon × Perf ÷ 10⁷Each of the 7 factors is scored 1–10. The product of all 7 is divided by 10⁷ (= 10,000,000), then multiplied by 100 to yield a 0–100 composite.
Equivalently: this is a geometric mean × 10. The 7th root of the product equals the geometric mean; multiplying by 10 normalizes that mean (which would otherwise range 1–10) to the same 1–100 scale as percentile-style metrics.
Why geometric mean (multiplicative) instead of arithmetic mean (additive)?
Arithmetic mean lets one strong factor compensate for one weak one. A 9 in Acquisition + a 1 in Retention averages to 5 (“medium”). Geometric mean refuses to accept that compensation: 9 × 1 = 9, vs 5 × 5 = 25. The product punishes the zero much harder than the average does.
This matches reality. A SaaS with great acquisition and zero retention is NOT “medium-quality” — it's a leaky bucket that loses every customer it gains. Linear scoring would call it healthy; multiplicative scoring calls it broken. AUG agrees with reality.
Worked example
A SaaS with self-reported scores:
- Acquisition: 5 (~5,000 sessions/mo)
- Activation: 6 (~35% activation rate)
- Engagement: 7 (PV/session ~2.5)
- Retention: 4 (D7 ~10%)
- Advocacy: 3 (k-factor ~0.05)
- Monetization: 5 (~€30/week)
- Performance: 8 (LCP 1.4s, CWV all green)
Product = 5 × 6 × 7 × 4 × 3 × 5 × 8 = 100,800
Divide by 10⁷ = 0.01008, multiply by 100 = AUG 1.01 — “critical” band.
Note how aggressively the score drops despite 3 factors being healthy (Engagement, Performance, Activation). The Advocacy = 3 and Retention = 4 multiply through the product and drag the whole composite to ~1. That's the discipline: weak factors cannot be compensated.
Action threshold bands
| AUG range | Band | Action |
|---|---|---|
| ≥50 | Fleet champion | Scale horizontally — start the next product. |
| 30–50 | Thriving | Invest in the weakest factor weekly. |
| 15–30 | Healthy | Per-factor diagnostic; queue a P1 fix. |
| 5–15 | Needs focus | Halt new features; one-factor sprint. |
| 1–5 | Critical | Roll back recent ships; triage with senior eyes. |
| <1 | Zombie | 90-day kill-criteria — sunset or rebrand. |
How Performance is measured (PageSpeed Insights conversion)
Performance is the only factor measured automatically. The audit wizard sends the URL to Google PageSpeed Insights API and reads the lighthouseResult.categories.performance.score— a 0.0 to 1.0 value. We convert to the AUG 1–10 scale by:
AUG_perf = max(1, round(psi_score × 10))Worked: PSI 0.92 → round(9.2) → 9. PSI 0.46 → round(4.6) → 5. PSI 0.04 → max(1, round(0.4)) → 1. The max(1, ...) prevents a score of 0 — even unmeasurably-slow sites get the minimum 1 (preserves multiplicativity).
PSI uses Lighthouse's composite Performance score, which weights:
- LCP (Largest Contentful Paint) — 25%
- Total Blocking Time — 30%
- CLS (Cumulative Layout Shift) — 25%
- FCP (First Contentful Paint) — 10%
- Speed Index — 10%
The Mobile strategy is queried by default — most consumer + SaaS traffic is mobile. Real-user data (CrUX field) is used when available; lab data fallback otherwise.
How the other 6 factors are scored
The wizard asks 6 questions with multi-choice answers. Each answer maps to a 1–10 score per the rubrics published on the per-factor method pages:
- Acquisition rubric (sessions/month thresholds)
- Activation rubric (cold-traffic activation rate %)
- Engagement rubric (bounce + PV/session composite)
- Retention rubric (D7 return rate)
- Advocacy rubric (k-factor + share signals)
- Monetization rubric (€/week revenue)
Why self-report is honest enough
Critique: 6 of 7 factors are self-reported; the operator can lie. True. But the composite formula is unforgiving — even one inflated score doesn't move the composite much when other factors are honest, and the diagnostic value is in the weakest factor (which operators rarely lie about).
Honest answers produce useful scores. Best-case guessing produces vanity scores. Most founders pick the honest answer because the diagnostic is for THEM, not for an audience.
Reproducibility
Every score can be recomputed independently. The audit is deterministic: same URL + same PSI response + same 6 answers = same AUG composite. The downloadable .txt report includes all inputs for verification.
For full transparency, the wizard source is open in the deployed Next.js bundle at /audit/.
Run your audit
Apply the formula to your own SaaS. Reproducible. Deterministic. Honest.