Tracking & attribution engineeringClient work
Lead tracking for Webflow and SPA sites: a layered listener that survives page transitions and measures a booking widget inside a cross-origin iframe.
Sounds familiar?
How I solve it
I build lead tracking that survives how the site actually works: a layered, deduplicated listener whose primary signal is the endpoint answering success, with a structural DOM fallback, a once-per-page guard and no personal data in the data layer. It survives in-page navigation, it survives the old container, and where the booking widget sits in a sandboxed cross-origin iframe I measure it with a postMessage bridge instead of a redirect that can never fire.
The process
6 steps. Scope and a fixed price are agreed before the first one.
What fires today, from which owner, for which form or booking — often without CMS access.
Network success first, structural DOM fallback, once-per-page guard, no PII in the data layer.
generate_lead becomes the GA4 key event; Ads imports it and restructures — generate_lead primary, form_submit secondary.
A postMessage bridge carries the booking confirmation out of the cross-origin widget when redirect detection is impossible.
Ads tags are blocked on 404 pages by the page's own structural id, and internal traffic is filtered.
Fifteen negative tests, five written acceptance conditions, and a change record where every change has its own one-step undo.
Handover
Built with
Proof
Client work Done for real clients. Client details are anonymised.
Before → after
Google Ads was optimising on
form_submit, the attempt, including failed submissions; it now optimises ongenerate_lead, a confirmed enquiry. In the 90 days before the work, 321 sessions reached the booking page and 5 reached confirmation — coming from an unmeasured booking widget. The fix closed with 5 of 5 written acceptance conditions met and 15 of 15 negative tests passed on 2026-08-21.
Hong Kong design studio, Webflow SPA
5/5
27 changes each with its own one-step undo; 5 of 5 written acceptance conditions met; 15 of 15 negative tests passed; booking measured from zero across a sandboxed cross-origin iframe with a postMessage bridge — with no CMS access.
Cross-domain, subdomain, iframe and user-ID tracking
keeping attribution intact when a funnel spans domains and iframes.
I hired [him] for the project on Friday and wasn't expecting him to start before Monday as the project was not urgent. However he did everything over the weekend!
What you can look at
A redrawn listener diagram — network-response primary with a DOM fallback and a once-per-page guard — plus a redacted excerpt of a success network response and the iframe postMessage bridge, rebuilt on synthetic data.
Client details anonymised. Figures read from live platform APIs in August 2026. Technical implementation, not legal advice.
Price and timeline
A bounded lead-tracking repair is a Focused Fix, from USD 450, with the scope and a fixed price agreed in writing before any work starts. If it is not yet clear what broke, the first step is a read-only Tracking Health Check (USD 395) with a written verdict, or a Working Session (USD 195) on one question. Discussing the task is free.
QuoteFixed before work starts
A note from Daniilbefore you decide
If your form already fires one clean lead on confirmed success, your booking tool is same-origin, and Ads already learns from the right event, you don’t need this — I’ll say so. And if the real problem is that the form is not converting at all rather than not being measured, that is a different job; a read-only check will tell you which one you have.
— Daniil
Questions
Yes. The only site-side change is the container snippet, delivered as paste-ready instructions you publish yourself. The container, Analytics and Ads changes are done through their APIs.
No. The listener is built to survive in-page navigation, which is exactly the failure the old container had.
Because the booking widget runs in a cross-origin iframe where the documented redirect cannot fire. A `postMessage` bridge measures the real confirmation instead.
No. The lead fires on the endpoint answering success, not on the attempt.
As few as possible, each with its own reversal path, recorded in a written change record. In the dated case there were 27.
The first answer is free, within one working day. Or write directly: next@taskfordaniel.com
