A jewelry brand learned its retargeting added 1.3× — and reinvested the difference

Representative examplesEvery case study in this library is an illustrative composite of the kind of engagement we deliver — written to show our method and standards, not to name clients.

Where they started

Demi-fine jewelry lives on repeat gifting and trust, and the brand had grown into a solid direct business with a strong December. Retargeting was its most celebrated marketing line — the board saw the platform ROAS number quarterly and had built budget expectations around it. A new CMO arrived from a performance-marketing background, pulled the audience definitions, found cart abandoners dominating the retargeting pools, and inherited the task of telling a room its favorite slide might be arithmetic luck.

What it was costing

Retargeting was the account's celebrated star: 8× reported ROAS, the slide the board loved. The new CMO suspected the number, since most of those users had the item in their cart already.

What they could see

• The retargeting campaign's reported return dwarfed every other line — a number the board quoted back unprompted in budget meetings.
• Audience definitions showed most of the retargeting pool had the product already in cart or on a wishlist page view.
• New-customer revenue share had been drifting down for quarters while total paid revenue held steady.
• The CMO's attempts to raise the question kept dying for lack of any alternative number to argue with.

The constraints we worked inside

• The board reported on platform ROAS; changing the metric needed evidence, not opinion.
• The brand's peak season (December) made mid-year testing the only clean window.
• Geo-holdout had a cost: some regions lose the retargeting — the test had to be sized to be readable but not ruinous.

What had been tried before

The team had compared weeks with and without the retargeting line, informally, over a quarter.

Unmatched periods, seasonality, and campaign activity made the comparison noise; nobody could take that standard of evidence to a board meeting.

Attribution windows in the account were shortened to shrink the reported return.

Shortening a window moves the reported number but measures nothing — the claim under test was incrementality, which a setting cannot answer.

An analyst was asked to reconcile platform revenue against order data monthly.

Reconciliation confirmed a gap existed but couldn't say why; without a control group the board could attribute it to anything convenient.

What we proposed

We proposed answering the CMO's suspicion with a pre-registered experiment rather than an argument. Matched geo pairs would run with and without retargeting for four weeks, sized to read a result but small enough not to ruin peak-season groundwork. The design and the decision rule went to the board before launch. When the measured number returned, budget would re-weight toward prospecting creative, and the board pack would report platform ROAS and incrementality side by side from then on.

Just as important is what we ruled out, and why:

• Customer-list match-rate audits as evidence — Match rates show the audience exists, not that ads changed behavior; the board's question was causal and needed a control group.
• Waiting until after December to test — Peak season is when retargeting's harvesting effect is most confounded by demand; mid-year was the only window where a clean read was possible.
• New attribution-platform procurement — A geo-holdout with the existing account answers the question directly; a platform migration mid-year risks the peak season on a tooling bet.

How the work ran

Delivered by the growth pod — paid social specialist over 6 weeks, with working increments reviewed with the client every week.

The stack, and the reasoning

Meta Ads

The retargeting line under review lived on Meta and its audience assets were reusable; the question was the metric, not the channel.

Geo-holdout test

Matched regions with and without the line is the only design that answers incrementality causally — and it needed no new tooling to run.

Conversions API

Server-side events tightened signal quality ahead of the test so the holdout read revenue behavior, not tracking noise.

Board-level dual reporting

A standing slide showing platform ROAS beside measured incrementality turned a one-off argument into a permanent governance habit.

What went wrong

How we worked together

Cadence

Fortnightly working sessions through the test window, one pre-launch board briefing, and one post-result session where the numbers were presented in the agreed format.

Client side

The CMO designed the decision rule with us and presented to the board; the e-commerce manager handled audience builds and order-data pulls.

Decisions

The pre-registered rule — re-weight below a threshold — did the deciding; the board briefing happened before results existed, which is what made the after-meeting short.

They provided

Two seasons of order data for region matching, board time for the pre-registration, and the discipline to leave holdout regions untouched for four weeks.

What changed

The headline: retargeting's measured incremental roas versus its platform-reported one — and the honest number changed the budget — 8× → 1.3×, read from Pre-registered geo-holdout test. A second check: new-customer revenue share of paid-attributed sales, following quarter at +21%.

The board pack carries two numbers side by side now, and the room reads them without argument — the CMO calls the pre-registration the most valuable hour of the engagement. Budget conversations moved from defending the retargeting slide to funding prospecting with intent. The marketing team stopped celebrating platform ROAS internally, and new-customer growth became the shared scoreboard. The skepticism that started the project hardened into a permanent habit: every claim now expects a control.

The result was read from Pre-registered geo-holdout test against the pre-engagement baseline over the stated window, with a guardrail check on new-customer revenue share of paid-attributed sales, following quarter. Where platform-reported numbers and business outcomes differ, this record says which layer it is quoting.

What they own now

• The pre-registered test design — region pairs, sizing method, decision rule — for reuse.
• The dual-reporting board format: platform ROAS beside measured incrementality, permanently.
• The rebuilt prospecting creative briefs funded by the re-weighted budget.
• The Conversions API configuration with the EU-region modeling notes.
• The region-matching dataset built from two seasons of order history.