Lifted ROAS by 220% across five brand storefronts without raising media spend

A retail group operating five direct-to-consumer brands across fashion, home, and beauty had grown to £280 million in annual GMV, but the commercial leadership could no longer defend their marketing investment thesis to their investors. Paid media spend had compounded to eighteen per cent of revenue — significantly above the sector benchmark — while blended ROAS had declined from four point one in 2022 to two point four by the first quarter of 2025. The diagnosis was not that the media was poorly bought; three separate agency audits confirmed that the individual channel performance was competitive. The problem was architectural: each brand operated as a data silo, and the group had no ability to identify customers who purchased across brands, attribute revenue across touchpoints that spanned brand boundaries, or reallocate spend toward the channels that were genuinely generating incremental lifetime value.

We opened the engagement by auditing the group's customer data infrastructure. The five brands collectively ran on seven distinct ecommerce platforms — three on Shopify Plus, two on a custom Magento-derived stack, one on Commerce Cloud, and one on a legacy in-house engine scheduled for replacement. Customer identities existed in each platform's own database, with no shared key. A customer who bought a bath towel from one brand and a skincare product from another appeared as two separate people in every system the commercial team consulted, including the paid media platforms that were targeting them. The first workstream, therefore, was an identity resolution layer: we deployed Segment as the customer data platform, implemented deterministic identity stitching on email and phone hash, and backfilled eighteen months of historical orders so that the resulting customer profiles had a meaningful commercial history on day one.

Snowflake became the warehouse of record. We modelled a unified order fact table, a customer dimension with brand-level and group-level cohort definitions, and a media-touch event stream ingested from Meta, Google, TikTok, Pinterest, and the group's email service provider. The attribution model itself was where the engagement became contentious. The existing agency attribution was last-click within each brand's individual platform — a method that systematically under-credits upper-funnel channels and over-credits branded search. We replaced this with a Markov-chain attribution model calibrated against the group's own holdout experiments, and we insisted on running the new attribution alongside the legacy last-click reporting for a full quarter so that the commercial leadership could see, with their own eyes, which channels were being misjudged.

The activation work followed the reporting work. With unified customer profiles flowing from Snowflake back into Meta, Google, and TikTok, the group could now build cross-brand audience segments: high-LTV beauty customers as a prospecting audience for the home brand, repeat home customers as a retention audience for fashion, lapsed fashion customers as a win-back audience across the group. We ran twenty-three controlled experiments over the final four months of the engagement, each with a true incrementality holdout, and the results directly informed a reallocation of roughly thirty per cent of the group's media budget away from channels that the new attribution showed were cannibalising organic demand and toward channels that were generating verifiably incremental revenue.

The commercial outcome was the one the board had asked for. Blended ROAS rose from 2.4× to 7.7× across the group — an uplift of two hundred and twenty per cent — while total media spend was held flat. Blended customer acquisition cost fell by thirty-one per cent as the reallocated spend produced higher conversion rates. Repeat purchase rate rose by forty-seven per cent, driven almost entirely by cross-brand re-activation that had previously been impossible to execute. Perhaps most importantly, the group's CFO and head of marketing now share a single commercial dashboard — the same definition of customer, the same definition of revenue, the same definition of attribution — and the agency review conversations that had occupied leadership for the previous year have been replaced by experiment-driven planning cycles.