Methodology
How LossPoint diagnoses a page.
LossPoint is not an AI that reviews your page and generates suggestions. It is a deterministic diagnostic engine that extracts facts, applies scoring rules, and identifies the single highest-impact conversion constraint — then uses AI only to explain the findings.
The same page produces the same diagnosis every time. The score is explainable. Every deduction traces to a named rule. The AI does not decide. It communicates.
Core Principle
AI is the last step. Not the first.
Most AI page tools feed your URL to a language model and return whatever it generates. Different prompt temperature, different output. Different day, different diagnosis. That is not a diagnostic system. That is an opinion generator with a clean interface.
LossPoint runs five deterministic layers before AI touches the result. By the time the language model is involved, the constraint has already been identified, ranked, and matched against known patterns. AI writes the explanation. It does not produce the diagnosis.
The six-layer pipeline.
Each layer is independent. Each produces structured output that feeds the next.
Layer 1
Fact Extraction
LossPoint fetches your page using a desktop Chrome user-agent and extracts structured facts from the server-rendered HTML. No JavaScript execution. No browser rendering. What the server sends is what the engine reads.
Sixteen facts are extracted: headline, subheadline, primary CTA, secondary CTA, testimonial count, pricing visibility, social proof presence, trust signals, value proposition, headline outcome-focus, feature-heavy language, proof near CTA, review count visibility, case studies, FAQ presence, and offer visibility.
Why this matters
The diagnostic engine never invents. Every observation is grounded in what was actually extracted from your page. If your page is JavaScript-rendered and content loads after the initial HTML, LossPoint qualifies its findings accordingly rather than stating something definitively that it cannot confirm.
Layer 2
Deterministic Scoring
The extracted facts are passed through 38 scoring rules across seven categories: Clarity, Positioning, Trust, Offer, Conversion, Psychology, and UX. Each category starts at 100. Rules deduct points based on what was or was not detected.
The rules are derived from conversion science — Ca$hvertising's 17 human drives, Cialdini's influence principles, and documented conversion failure patterns observed across hundreds of businesses. The weights are fixed. The rules do not change between runs.
Clarity
Weight: 20%
Trust
Weight: 15%
Offer
Weight: 15%
Conversion
Weight: 20%
Psychology
Weight: 10%
Positioning
Weight: 10%
UX
Weight: 10%
Why this matters
The score is not a judgment. It is a measurement. Every point lost maps to a specific rule that fired against a specific element on your page. You can verify every deduction in the Score Breakdown section of your diagnosis.
Layer 3
Constraint Ranking
Once scoring is complete, the engine groups fired rules by category and sums total deductions per category. The category with the highest total deduction is the primary constraint. Within that category, the rule with the largest single deduction names the specific issue.
Confidence is assigned based on how many rules fired in the primary constraint's category: three or more supporting signals means High Confidence. Two means Medium. One means Low. Confidence is always explained — not decorative.
Why this matters
This is the step that separates diagnosis from audit. Most tools list every issue they find. LossPoint ranks constraints by impact and surfaces the one that costs the most — the constraint that, if fixed, makes the other improvements more effective.
Layer 4
Pattern Matching
The primary constraint and extracted facts are compared against a library of seven documented conversion failure patterns. Each pattern has named trigger conditions derived from the extracted facts.
Mission Before Value: The page leads with what the company believes before what the visitor receives.
Feature Dumping: Capabilities are listed without connecting them to outcomes the visitor wants.
Weak Trust Stack: The page asks for action before establishing that others have acted and benefited.
CTA Friction: The call to action is generic or asks for commitment before value is clear.
Offer Ambiguity: The visitor cannot determine what they receive, what it costs, or what they commit to.
Proof Gap: Claims exist without supporting evidence, testimonials, or third-party validation.
Value Prop Mismatch: The headline and body content address different problems or audiences.
Why this matters
Pattern matching connects your specific constraint to observed outcomes from similar businesses. The pattern tells you not just what is wrong — but what typically happens to visitors when this pattern is present on a page.
Layer 5
Knowledge Base Retrieval
Before the AI writes a single word, the engine queries the LossPoint knowledge base for case studies, best practices, and documented leak patterns that match your primary constraint, funnel stage, and conversion principles.
The knowledge base is a structured database of real-world conversion outcomes — not generated examples. Each entry contains a documented business problem, the change made, the observed result, and operator intelligence on what actually drove the improvement. Entries are tagged with funnel stage, leak category, and conversion principle for precise retrieval.
Why this matters
The knowledge base is what separates LossPoint from a prompt. The AI prompt can be copied. The knowledge base compounds with every case study added. Over time, every diagnosis becomes more precise because it is backed by more observed evidence from similar businesses.
Layer 6
AI Explanation Layer
Only after the first five layers have completed does a language model enter the process. The AI receives the extracted facts, scoring results, primary constraint, matched pattern, and knowledge base context as structured inputs — and produces the written explanation.
The AI is explicitly constrained: it cannot change the constraint, cannot select a different pattern, cannot generate scores. It writes the evidence section, the visitor psychology, the decision moment, and the exact fix teaser. Every field it produces is checked for compliance with a 10-point quality test before the diagnosis is returned.
Why this matters
AI is excellent at communication. It is unreliable as a decision-maker for structured analysis. LossPoint uses it for what it is good at: translating structured findings into language that founders can read and act on. The engine decides. The AI writes.
Same page. Same result. Every time.
Layers 1 through 5 are entirely deterministic. No randomness, no temperature, no prompt variation. The same URL produces the same extracted facts, the same scoring output, the same constraint, and the same pattern match on every run.
Layer 6 (AI explanation) has minor variation in phrasing from run to run — but the constraint it explains, the evidence it quotes, and the pattern it references are locked by the deterministic layers above it. The diagnosis itself is stable. The language describing it may vary slightly.
What the free diagnosis does not include.
The diagnosis identifies and proves the constraint. It does not implement the fix.
Those are available in the Implementation Blueprint, accessible from inside your diagnosis report.
See the methodology in action.
The diagnosis takes under 60 seconds. No signup required.
Run a free diagnosis →Free · Deterministic · No signup