METHODOLOGY
How 2nd Half Measures
Promise → Fulfillment
Assistant journeys create specific promises ("SOC 2 + HIPAA, 5k endpoints, ZTNA in 90 days"). 2nd Half measures whether your landing experience keeps that promise and whether the visitor completes the intended action quickly. This page explains the data we collect, how we compute our metrics (PFR, FIS, CAT, Integrity), and the guardrails that keep results reliable and compliant.
What we measure
PFR — Promise-Fulfillment Rate (0–1)
How closely the landing experience matches the promise the assistant/ad conveyed. Combines token checks for critical facts with semantic similarity for phrasing/context.
FIS — First-Intent Success (boolean + rate)
Whether the visitor completes the promised action (e.g., request scoped assessment, start checkout) without detours.
CAT — Conversation-to-Action Time (ms)
Latency from the last assistant answer to the first qualifying action.
Integrity — Claim Integrity Rate (0–1)
Share of on-page claims (price/specs/eligibility/coverage) that are verified against a registered source (your "truth registry").
Data we use (minimal & transparent)
Promise Object (from Relay payload or reconstruction): intent, entities (region, SKU, frameworks, endpoints…), constraints (timeline, SLA, price band…), optional claims and the raw assistant text.
LP Snapshot (rendered content): hero copy, primary CTA label/URL, key module excerpts (architecture, plan timeline, price panel), structured fields (e.g., region, SKU, price).
Actions (first-party events only): request quote/demo, add-to-cart, checkout start, book assessment, form submits.
Attribution hints (if present): gclid/msclkid/fbclid + UTM parameters; we never require user identifiers.
2nd Half does not fingerprint users and does not collect sensitive content unless you explicitly sync it into the truth registry.
How we compute PFR
- 1
Select critical tokens: We extract a small, policy-controlled set of fields from the Promise Object (Identity & scope: brand/SKU, framework, coverage, endpoints, region/ZIP; Constraints: price band, SLA, timeline).
- 2
Token Match Score (0–1): For each critical token k: v, we test whether the LP Snapshot renders the same value. Token score = matches / required. Normalization and synonym maps are applied.
- 3
Semantic Match Score (0–1): We compare the Promise text against the above-the-fold LP text using vector similarity (cosine). We normalize casing, strip boilerplate, and cap length.
- 4
PFR blend and caps: PFR = w_tokens * TokenScore + w_semantic * SemanticScore, with defaults w_tokens=0.6, w_semantic=0.4. Hard-must token failures cap PFR at ≤ 0.6.
- 5
Confidence annotation: Each PFR record carries a confidence flag (High/Medium/Low) based on token completeness and promise source.
How we compute FIS
Definition: FIS = true if the first qualifying action happens within the session without intermediate navigation that indicates search/browse.
Qualifying actions: B2B: request scoped assessment, book demo, start security review, submit RFP. B2C/commerce: add to cart, checkout start, booking intent.
Detour rules: We consider a detour if any occurs before the first qualifying action: 2 pageviews that don't contain the promised intent/tokens, search box usage with unrelated keywords, category hopping that contradicts promised scope.
How we compute CAT
CAT = action_ts − last_assistant_ts. We record the assistant timestamp when the link is created (Relay payload or reconstruction from click context) and subtract it from the first qualifying action timestamp. Reported in milliseconds; dashboards show humanized bins (e.g., <30s, 30–120s, 2–10m, 10m+).
How we compute Integrity
Claim extraction: From the LP Snapshot, we parse structured statements: price, specs, frameworks, SLAs, availability, eligibility, warranties.
Verification: Each claim is checked against your truth registry (CPQ, PIM, eligibility/serviceability API, policy store). Integrity = verified_claims / total_claims_in_scope.
Scope control: You decide which modules enter the Integrity scope; we default to any claim that could trigger a legal/compliance risk if misstated.
What "good" looks like (benchmarks guidance)
PFR
Strong: 0.85–1.00
Partial: 0.65–0.85
Poor: <0.65
FIS rate
Strong: B2B > 12–18%
Partial: Commerce > 8–12%
Poor: Category-dependent
CAT
Strong: B2B < 3–7 min
Partial: Commerce < 90 sec
Poor: Baseline varies
(These are directional; your baseline may vary by vertical and funnel.)
Limitations (we call them out)
Reconstructed promises (UTMs only) lower confidence; use Relay payloads where possible.
PFR reflects above-the-fold alignment; deep content mismatches won't move PFR but will depress FIS.
Integrity depends on the completeness and freshness of your truth registry.
Privacy & security
2nd Half operates on first-party session data.
No device fingerprinting; no third-party cookies.
Promise Objects exclude PII unless you explicitly include it (not recommended).
Claims registry is read-only for verification; we do not write to your CPQ/eligibility systems.