US education provider
2.2% to 0.5%
landing-page view to lead conversion 2.2% → 0.5% on the main ad set while clicks and page views held.
Tracking & attribution engineeringClient work
A read-only reconstruction of the failure window: what changed, when, and what it did to your numbers, with confirmed findings kept apart.
Sounds familiar?
How I solve it
I build a read-only reconstruction of the failure window. I pull the platform data and the change history, compare a clean baseline with the broken period, and rank the causes by evidence. Confirmed findings are kept separate from unconfirmed ones. You get a written answer and a next step — and nothing is changed on your account while I look.
The process
6 steps. Scope and a fixed price are agreed before the first one.
Agree the window: when it worked, when it broke, and what "broke" means in numbers.
Pull read-only data: platform insights, event counts, and change history for the period.
Split the data by the dimensions that matter — device, placement, event type, source — to localise the break.
Test hypotheses against a control, so a drop is not mistaken for noise.
Write the verdict: what changed, what it did, and how confident I am.
Give you the ranked next checks, whether you act with me or without me.
Handover
A read-only diagnosis: what changed, when, and what it did to your numbers. Confirmed findings kept apart from unconfirmed ones. A ranked list of causes with the evidence for each, and the next checks to run. You can take it to your team or another specialist; implementation is a separate engagement.
Built with
Proof
Client work Done for real clients. Client details are anonymised.
Before → after
Before, a US education provider’s lead campaigns collapsed overnight while spend and page views held. The reconstruction localised the break after the landing-page view: lead conversion on the main ad set fell from 2.2% to 0.5% while clicks and page views did not move, reconstructed on 2026-05-14.
US education provider
2.2% to 0.5%
landing-page view to lead conversion 2.2% → 0.5% on the main ad set while clicks and page views held.
US home-safety products store
attribution drop traced to a consent change and proved with a difference-in-differences test, p = 0.0002; re-checked 2026-09-02.
Closed emergency work on Upwork
Meta attribution reading zero while sales continued, collection purchase limits, a countdown app and order notes, an app install and database move.
I gave a unclear brief as I was unsure what the tracking issue was, but [he] fixed the issue very quickly and with great communication.
What you can look at
A redrawn timeline of the failure window — baseline against failure, with the events that moved and the tests used — on synthetic numbers that keep the real ratios, with all campaigns, ids and account names removed.
Client details anonymised. Figures come from read-only platform access and client delivery records on the dates shown; some details altered for anonymity. I am the founder of Fixel Pixel.
Price and timeline
A Working Session answers one pre-chosen question in up to 60 minutes with a short written summary. A Health Check is the read-only written verdict, delivered two working days after the kickoff and confirmed access. Both are diagnosis; neither forces an implementation.
QuoteFixed before work starts
A note from Daniilbefore you decide
If your conversions are stable and someone on your team reads the platform diagnostics every day, you don’t need me. A read-only diagnosis is the wrong buy when you already know the cause and only need it fixed — that is a Focused Fix instead. And if the drop is clearly a bad ad rather than a broken measurement, I will say so; that happened, and I declined to sell a rebuild.
— Daniil
Questions
No. The diagnosis is read-only: access to read data, and nothing that can write. If a fix is needed, it is a separate, approved job.
The Health Check clock starts once the kickoff is done and access is confirmed sufficient — not before. That keeps the two-working-day promise honest.
Then the report says so. You get the confirmed findings, the open hypotheses, and the specific checks that would settle each one — not a confident story.
Often, yes, for a first read. It limits the confidence level, and the report will say which findings are weaker because of it.
The first answer is free, within one working day. Or write directly: next@taskfordaniel.com