# Lead attribution analysis — secondwindconsultants.demoing.info (v2) *Run against package-v2.md (All time, 2026-04-14 → 2026-08-08, 435 leads). Every claim cites a dataset figure or a page I opened. The v2 dataset pre-flags its own measurement artifacts, which v1 had to discover by hand — the difference shows in how much faster this report gets to the point.* ## 1. The three limitations that most affect confidence 1. **93 journeys (21%) begin on a confirmation page** (limitations §"Some journeys begin on a confirmation page") — post-submission echoes recorded as journey starts, not acquisitions. They sit inside the landing table (marked) and inside Direct's 41%. 2. **Every journey is an import** — no interaction events exist anywhere, and page history is only as complete as the form platform's stored summaries. Behaviour claims below are page-level only. 3. **Paid tracking is broken upstream** — campaign coverage 1%, search terms are single letters, 66 leads carry one-letter mediums, 78 carry literal `{CampaignName}` placeholders (Data-quality findings). Nothing ad-level is analyzable this window. One journey is flagged as **staff, not a visitor** (wp-admin pages in its history); the samples exclude it from the engagement bucket. The 70.5-median-visits form row is that journey's fingerprint — read that form's row as noise. ## 2. Channels — what the volume hides | Fact | Figure (dataset) | |---|---| | Organic search is the discovery engine | Google Organic 119 leads (27%) **+ 19 assists**; transitions: *Organic → Direct, 17* | | Ads deliver volume with thin engagement | Google Ads 71 (16%), median **1.0 pages**, 14% came back, 8 assists | | Direct is the biggest label, not the biggest channel | 180 (41%) — includes confirmation echoes (§1), expired sources, own-property arrivals | | AI assistants are a real channel now | ChatGPT 22 leads (5%) — more than Social+Bing+Email combined; 2 assists | | High-intent stragglers exist | DuckDuckGo: 2 leads, median 7.0 pages, 50% came back (anecdote-sized, noted per Forms warning) | **The last-click distortion is specific and quantified:** 41 of 435 leads (9%, a floor) converted through a different channel than found them. Google Organic found 19 and was credited with none. Under last-click, organic looks ~14% smaller than its real contribution (119 credited vs 138 touched); cutting SEO spend on the 119 number would be cutting the channel that also *starts* the Ads and Direct wins. ## 3. How these people decide 78% enquire on their first visit; the median lead reads 2 pages (Headline; Channels). The sample journeys show the canonical path: land on home or a resource page → `/contact-us/` → submit, same day (leads #1302, #1303, #1305). This is a **distress purchase** — MCA-trapped owners act fast. The research minority exists (#1041: 18 days, 8 visits, returning repeatedly to an SBA calculator) but is small. The pace matters for the recommendations: pages get one shot. There is no nurture window in which a weak page gets a second chance. ## 4. The pages, as visited - **Home** (`/`) — "Resolving business distress, preserving value.", *Get Help →*, prominent phone. Matches the urgency of the traffic. 134 journeys start here (echo-marked rows excluded, per the marked landing table). - **`/solutions/…/mca-resolution/`** — now correctly surfaced by the dataset's Site list (85 journeys, 55 submissions — the strongest converting page). Headline names the exact pain ("crippled by weekly or daily withdrawals"); warning-framed video; free-consultation offer; Rise Alliance co-brand. **This page is the template the rest should follow.** - **`/l/dn/mca-relief-guide/`** (71 journey starts, 57 submissions) — the paid funnel's landing/download page. The consultation CTA is a text link under "What's Next?" while *Download Guide* gets the button. The funnel's close is its weakest element. - **`/contact-us/`** (166 submissions) — short form (name/email/phone + one select), confidentiality note. Low friction, appropriate. (Site hygiene: the plugin's debug mode is on — it exposes a `window.BWLeadAI` object to every visitor and shows internal fields to logged-in users. Turn it off.) - **`/contact-us/thank-you/`** — bare "Message Sent". For a caller in financial distress, silence after the click. Nothing says *when* contact will come or what to prepare. - A red corner error (probably reCAPTCHA rejecting this staging domain) recurs on several pages — verify absent on production. ## 5. Five changes, most valuable first 1. **Fix the Google Ads tracking template** (findings: 78 placeholder + 69 campaignless + 66 broken-medium leads). One template edit in the ad account turns 16% of lead volume from a black box into campaign-level truth. *Verify:* the dataset's "Campaigns and search terms" absence-branches disappear; Data-quality counts trend to 0 over 2–3 weeks of new leads. 2. **Give `/contact-us/thank-you/` a real next step** — "A specialist calls you within X hours; have your advance statements ready", plus the phone number. 400+ people a season land here at peak anxiety (every submission funnels through it; 93 journeys even *begin* at its echo). *Verify:* Direct's "came back" share (30%) drifts down as post-submit wandering drops. 3. **Swap the CTA hierarchy on `/l/dn/mca-relief-guide/`** — make *Schedule a FREE Consultation* the primary button, the download secondary. 57 submissions already pass through; the page currently optimises for the PDF, not the conversation. *Verify:* "Pages appearing in journeys" — submissions attributed to the `/l/dn/…` pair over 60 days. 4. **Double down on the resource engine** — organic+ChatGPT = 141 leads + 21 assists; `/resource/…` guides start 28+ journeys; assistants are already citing the content (ChatGPT column). One authoritative decision-stage guide per month, modelled on `/resource/dont-consolidate-mca-debt-eliminate-it/` (10 starts). *Verify:* Landing-pages table resource rows; ChatGPT row trend. 5. **Finish cleaning the measurement** — the dataset now flags confirmation-starts (93) and staff journeys (1) itself; the remaining step is capture-side: stop recording post-submit views as journey starts and skip logged-in staff. Then Direct becomes interpretable and every later analysis sharpens. *Verify:* "Some journeys begin on a confirmation page" count stops growing; the marked rows fade from the landing table as new journeys accumulate. ## 6. What this dataset cannot answer 1. **Any conversion rate** — converters only, twice stated in the dataset. *Needs:* all-session page capture (even sampled) to provide a denominator. 2. **Which ads/keywords produce leads** — broken tags (see #1). *Needs:* the template fix; ValueTrack params fill campaign/term from the next click onward. 3. **What the residual Direct is** — after removing echoes (93) and own-property arrivals, an unknown mix of true type-ins and expired sources remains. *Needs:* the plugin's cross-subdomain source carrying (on its roadmap) plus longer live-capture history, after which Direct's remainder is finally a real number. --- *v1 → v2 differences that mattered: the artifacts I had to deduce in v1 (§3 there) arrived pre-counted (93/1), the landing table self-marks its echo rows, and the Site list now sends the analyst to the strongest converting page instead of a "Message Sent" screen. The report above needed no detective work — which is the point of the export.* --- **Postscript (same day, after this run):** a measurement correction landed after this analysis was produced — visits are now bounded by a 30-minute session gap and pages count unique pages, applied to history via Reprocess. "Came back" figures cited above roughly halved under the corrected definition (Direct 30%→16%, Organic 19%→12%, Ads 14%→6%) and the first-visit share rose to 89%. The *relative* story — Direct's repeat share highest, Ads' lowest — survives unchanged, which is why the recommendations stand. Rerun the export for current numbers.