Skip to content

Home-safety productsClient work

Proving Why Meta Attribution Dropped (p = 0.0002)

I separated consent-related journey loss, offline sales and inflated Meta purchase events for a US home-safety store, then checked recovery against orders.

The brief

Context

A US home-safety products store saw attribution deteriorate around a consent-app change. Removing the app had not made the reporting problem disappear. Sales continued, but the store and Meta no longer gave the owner a story they could reconcile.

I built an order-level evidence trail from the store’s real orders, Meta’s received events, the privacy configuration and storefront runtime. The audit was read-only.

Sounds familiar?

Symptoms

The situation Home-safety products
Tick what applies to you

These were separate defects and limits. Treating them as one attribution percentage would hide the work needed to fix each one.

Tick the lines that describe your case.

Describe my task

The investigation

Diagnosis

I compared web orders before and during the consent-app window, then separated the cohort subject to the consent requirement from a control cohort. The loss of customer journeys increased in the affected window. The report’s one-sided Fisher exact test gave p = 0.0002 for that change.

  1. The control cohort did not show the same loss during the window. That helped localise the missing journeys to the consent mechanism rather than a store-wide outage.

  2. There was an important limit to the causal claim: the privacy configuration was read after the app had been removed. No before-snapshot existed. The timing and the observed cohort difference supported the explanation, but did not prove which action first switched the requirement on.

  3. I also reconciled the purchase-event count separately. Meta received 166 purchase events against 105 real orders in the report sample. Browser and server events lacked a shared event identifier, so the received-event count could not be treated as the count of unique orders.

  4. Finally, manually entered sales explained another part of the gap. A purchase entered outside the storefront has no web checkout for a browser pixel to observe. Consent changes and browser/server deduplication cannot solve that structural limit.

Handover

What I delivered

Delivery noteDelivered Aug 2026

  1. An order-level audit that separated customer-journey loss, manual sales and event-count inflation.
  2. A consent diagnosis with a control comparison and an explicit limit on causality.
  3. A separate deduplication finding, checked against real orders rather than an attribution dashboard.
  4. A prioritised action plan and a dated follow-up after the consent requirement changed.

The follow-up mattered. An early read did not show recovery. By the later September check, every web order in the small sample since 1 September had a journey. I kept both readings in the work record rather than replacing the earlier result with a cleaner story.

Before → after

Results

Client work Done for real clients. Client details are anonymised.

  • Statistical test of the change in missing Shopify customer journeys

    Aug 2026-29

    Before
    Baseline compared with the consent-app window
    After
    One-sided Fisher exact test, p = 0.0002

  • Purchase-event count compared with real orders in the report sample

    Aug 2026-29

    Before
    105 real orders
    After
    166 purchase events received by Meta

  • Customer journeys in the later post-change web-order sample

    Sep 2026-09

    6/6

    Before
    Journey recording had deteriorated
    After
    6 of 6 orders since 1 September had a journey
The full results table
CheckObserved resultRead on
Change in missing customer journeysOne-sided Fisher exact test, p = 0.000229 August 2026
Meta purchase events versus real orders in the report sample166 versus 10529 August 2026
Post-change web orders since 1 September carrying a journey6 of 69 September 2026

Notes on the numbers

The follow-up is a small order sample, not a promise of permanent coverage. It measures Shopify journey recording, not complete recovery of Meta ad attribution. The record also leaves the exact mechanism behind the delayed recovery unresolved.

A note from Daniilbefore you decide

What this means for a similar business

When attribution drops around a consent change, I separate the records that disappeared from the sales the browser could never see. I then compare unique orders with received events. A consent fix, a server event for an offline sale and a shared deduplication key solve different problems; each needs its own evidence.

— Daniil

Situations this case answers

All situations
Read another case

See What ChatGPT Sends You: AI-Referral Reporting for a Ticketing Business

I added AI-referral reporting to a ticketing business’s own Looker Studio report and checked the underlying purchase tracking separately.
Daniil Maximkin

Hi, I’m Daniil.

I work with you from defining the problem to implementation and handover. You talk to the person who does the work. I work in English and Russian.

My story My work

Have a similar task?

Describe your task

The first answer is free, within one working day. Or write directly: next@taskfordaniel.com