Five real brands. Real engagements.

Founder-led Shopify fashion brand running a mixed paid stack across Meta and Google Shopping. The dashboards reported double-digit ROAS on parts of the spend that, on Shopify, were not showing up as new orders. Standard founder dilemma: the platforms looked great, the bank account did not.

First causal read on the GA4 export separated the revenue the channels caused from revenue they were merely present for. It surfaced roughly 8K EUR per month of non-incremental Google spend: money buying conversions that were already going to happen.

The brand reallocated that spend. No pixel was installed and nothing was added to the storefront; the input was a GA4 export of a period that had already happened. The webstore was mid-migration to a new platform at the time, which gives the 2025 lift number an additional headwind context. Q1 2026 reflects the brand acting more aggressively on subsequent causal recommendations.

Founder considering pulling Pinterest entirely. Last-click in GA4 showed Pinterest taking credit for a small share of orders; the dashboards were arguing for cutting it. The intuition was: Pinterest is browser traffic that converts later, and the click does not get attributed.

Causal analysis on the brand's GA4 + Shopify data resolved the debate cleanly: Pinterest was driving incremental sales the last-click report missed. The conversion path was running through other touchpoints by the time the order landed, but the causal model isolated Pinterest as the channel actually moving the needle on a chunk of weekly orders that would not have happened without it.

The brand kept Pinterest live and rebalanced the rest of the mix. The 2024 revenue lift below reflects the year's cumulative effect of rejecting the obvious dashboard read and acting on the causal one instead.

Brand was scaling paid acquisition on Meta and watching last-click ROAS hold steady while gross profit dropped. The platform report read like growth; the P&L read like a leak.

The causal read surfaced what the last-click view structurally cannot: customers acquired through Meta were returning product at a materially higher rate than other channels. The platform credit-allocated revenue was real at point-of-sale and gone again by the time the returns processed two weeks later. Net-of-returns ROAS on Meta-acquired customers was substantially below the headline.

We rebuilt the bidding around net contribution rather than gross revenue and capped the Meta-acquired customer cohort at a defensible lifetime value. The 2024 revenue lift is the cumulative effect of stopping the hemorrhage and putting the spend behind cohorts the causal model identified as genuinely profitable.

Eat your own dogfood. Joris, the founder of Causality Engine, built Maison Tanger (formerly Vivid Merve) from scratch: a French womenswear dropshipping store, a brand he owns outright. The point was not the store. The point was the test.

He ran it for six months making every marketing decision blind, on the Causality Engine output alone. No gut overrides. No second-guessing the platform dashboards. If the causal read said cut a channel, he cut it. If it said scale, he scaled. Budget reallocations, channel calls, spend levels, all of it came straight from the model, the same way a customer would use it.

It is the one engagement where we can be fully transparent end to end, because we own the brand and the data. It is also the cleanest proof we have that the product is meant to be acted on, not admired: a real store, real spend, run for half a year by a founder who chose to trust the math over his own instinct.

The brand brought us in to audit channel-level attribution under the working hypothesis that meaningful revenue was being misattributed across the paid + organic split. Real engagement, GA4 + shop export, full causal read.

Honest finding: at the brand's then-current spend level, the channel-level gap the causal model surfaced sat within the confidence interval. The signal was there, but it was not yet significant enough to defend a meaningful reallocation decision against. Their existing analytics stack was already telling them most of what a EUR 99 causal read could add at that scale.

Our recommendation was to skip the continuous tier and revisit when paid spend scales. We would rather say that than pretend the product fits every ecommerce brand at every stage.