# Lead attribution analysis — secondwindconsultants.demoing.info
*Run against package-v1.md (All time, 2026-04-14 → 2026-08-08, 435 leads). Every claim cites a dataset figure or a page I opened.*

## 1. The three limitations that most affect confidence

1. **Every journey is an import.** 435 of 435 were reconstructed from form history (limitations §"Journeys imported"), so there are no interaction events anywhere, and page history is only as complete as the old records. Everything below about *behaviour* is page-level only.
2. **Direct is a bucket, not a channel.** 41% of leads are "Direct" (Headline), and the dataset itself says that figure absorbs expired sources, referrer-stripping apps and the site's own properties. I found two further contaminants inside it (below) — Direct's real "typed the address" share is materially smaller than 180.
3. **The paid-tracking fields are broken upstream.** 1% campaign coverage, terms that are single letters (`e`, `p`, `b`), 66 leads with one-letter mediums, 78 with literal `{CampaignName}` placeholders (Data-quality findings). Any statement about *which ads work* is unsupportable this window.

## 2. Channels — volume vs quality, and what last-click hides

- **Google Organic, 119 leads (27%), is the discovery engine.** It also has **19 assists** — first-touch on leads credited elsewhere — and the transitions table shows *Google Organic → Direct, 17 leads*: people who found the site by search and came back later unattributed. Under last-click, organic's true contribution is at least 119+19; it looks cheaper to cut than it is (dataset's own sentence).
- **Direct, 180 leads (41%), median 2.0 pages, 30% came back.** Partly real loyalty — but see §3: some of it is measurement.
- **Google Ads, 71 leads (16%), median 1.0 page, 14% came back.** Ad clicks read one page and enquire — thin engagement, real volume. Plus 7 assists that closed as Direct. The ad *tracking* is broken three ways (limitations §findings), so this channel's internals are invisible.
- **ChatGPT, 22 leads (5%)** — larger than Social, Bing, and Email combined. An AI assistant is citing this site's content; two of those also assisted Direct conversions.
- **Sleepers:** DuckDuckGo (2 leads, median 7.0 pages, 50% came back) and the trade referrers (thomasnet, abfjournal, abladvisor) are tiny but engaged — read as anecdote (Forms warning applies to small rows generally).

## 3. How these people decide — and two things in the data that are not people

**The genuine pattern:** 78% enquire on their first visit, median 2 pages (Headline; Channel table). Sample leads #1302–#1305: land (home or a resource page) → `/contact-us/` → submit, same day or next. This is a *distress purchase* — people in MCA trouble don't comparison-shop for weeks. The long tail is real but thin: #1041 (18 days, 8 visits, the SBA calculator repeatedly) shows the researcher minority.

**Artifact 1 — the thank-you pages.** `/contact-us/thank-you/` is the **#3 landing page (65 journeys)** and shows 32 "submitted here"; `/thank-you-social/` adds 28. I opened `/contact-us/thank-you/`: it is a bare "Message Sent" confirmation — nobody *arrives* there from marketing. These are post-submission page views being recorded as journey starts (visible in samples #1305, #1307, #1308, whose histories *begin* at the thank-you page). Roughly **90+ journeys' landing data and part of Direct's 41% is this echo**, not acquisition.

**Artifact 2 — staff browsing.** Lead #1184 (111 visits, 164 pages) includes `/wp-admin/options-general.php` in its page list — a WordPress admin screen. Both "Landing - Stop Withdrawals" leads look like internal/test traffic, and they carry the form table's absurd 70.5 median visits.

## 4. The pages, as visited

- **Home** (`/`, 134 journeys start here, 95 submissions attributed): "Resolving business distress, preserving value." — clear, matches the distress-purchase pattern; "Get Help →" and a prominent phone number. Submissions attributed to `/` suggest a site-wide form (modal/footer), which also fits the short journeys.
- **`/solutions/…/mca-resolution/`** (42 starts, 85 journeys touch it, **55 submissions** — the strongest single converting page; from the tables, as the Site list omitted it): headline speaks directly to "crippled by weekly or daily withdrawals", free-consultation offer, a warning-framed video ("Why Debt Settlement Mills Fail Your Business"), Rise Alliance co-brand. Message-match with MCA-distress intent is excellent. This page is the model.
- **`/l/dn/mca-relief-guide/`** (71 starts, 57 submitted): the guide-download funnel converts, but the page I saw is the *post-submit* variant (Download button + "Schedule a FREE Consultation" link). The consultation upsell is a text link buried under "What's Next?" — the weakest element on the money page of the paid funnel.
- **`/contact-us/`** (240 journeys touch it, 164 submissions): short form — First/Last/Email/Phone + one select — with a confidentiality line. Appropriately low-friction for high-anxiety visitors. (An unlabeled "Untitled" field visible to me is the attribution plugin's own field in debug mode — invisible to real visitors, but **debug mode is on**, which also exposes a `window.BWLeadAI` object publicly; turn it off.)
- **`/contact-us/thank-you/`**: "Message Sent… a representative will be in touch." Dead calm after a high-anxiety enquiry. Nothing tells the visitor *when* or *what happens next* beyond one sentence.
- A red error box (cut off, likely reCAPTCHA domain rejection) appears bottom-right on several pages — probably an artifact of this being a staging copy; verify on the production domain.

## 5. Five changes, most valuable first

1. **Fix the Google Ads tracking template.** Campaign arrives as `{CampaignName}`, terms as single letters, mediums as `g`/`d` (findings: 78+69+66 leads affected). Until fixed, 16% of lead volume is unanalyzable. *Verify:* "Campaigns and search terms" section stops rendering its absence branch; Data-quality counts fall toward 0 within 2–3 weeks of new leads.
2. **Put a real next step on the thank-you page** ("what happens in the next 2 hours / book a slot now"). 435 people a season see this page at maximum anxiety; sample journeys show them bouncing back through it repeatedly (#1041, #1307). *Verify:* "came back" share on Direct in Channel performance over 60 days — repeat wandering should fall.
3. **Promote the consultation CTA on `/l/dn/mca-relief-guide/`** from a text link to the primary button beside "Download Guide" — 57 submissions already flow through this page; it's the paid funnel's close. *Verify:* "Pages appearing in journeys" — submissions on the `/l/dn/…/thank-you/` step (currently 5 of 14 journeys touching it).
4. **Feed the organic/AI engine deliberately:** the `/resource/…` guides start 28+ journeys and organic+ChatGPT deliver 141 leads with 21 assists (tables). One new authoritative MCA-decision guide per month; the ChatGPT column says assistants are already citing this content. *Verify:* Landing-pages table — resource-page starts; ChatGPT row trend.
5. **Clean the measurement so the next analysis is sharper:** exclude thank-you pages from landing/submission attribution, exclude logged-in staff from capture, and turn debug mode off (observations in §3–4 — all three are visible in this dataset's own tables). *Verify:* `/contact-us/thank-you/` drops out of the landing table; the 70.5-visit form row disappears.

## 6. What this dataset cannot answer

1. **Conversion rate of anything** — no non-converting traffic. *Needs:* page-view capture for all sessions (even sampled), so converters have a denominator.
2. **Which ads and keywords produce the leads** — broken upstream tags. *Needs:* the template fix in #1; then ValueTrack fills campaign/term for every paid lead.
3. **What Direct really is** — 41% is a mixture of loyalty, expired sources, thank-you echoes and staff. *Needs:* the measurement fixes in #5 plus the plugin's cross-subdomain source-carrying (its own roadmap names this), after which the residual Direct is finally interpretable.