Case Study•Consumer Goods

Thermos

Global Drinkware Brand

Engagement: 5 months + ongoing

+94%

Contribution margin increase

+86%

Revenue growth

+86%

Conversion increase

5mo+

Ongoing partnership

Problem

Thermos was spending confidently and growing slowly. Conversions looked healthy, ROAS looked acceptable, and yet the contribution the account returned to the P&L never matched what the dashboard promised.

The brief was not "get more revenue". It was to work out why a profitable-looking account was not behaving like a profitable one.

Evidence

The previous agency's tracking was double-counting conversions, inflating reported ROAS by roughly 40%. Every optimisation decision for the preceding period had been made against numbers that did not exist.

Once the data was corrected, a second problem appeared: campaigns were segmented by product type, not by margin. High-COGS drinkware was competing for the same budget as low-COGS bestsellers, so the account was systematically buying the least profitable growth available.

Decision

Fix measurement before touching spend. No budget decision was made until the baseline was accurate.

Then give each product range a commercial job rather than a shared blended target, so bidding follows margin instead of volume. Creative and landing page experience were already strong and were deliberately left alone.

What changed

Conversion tracking rebuilt and deduplicated, establishing an accurate baseline for the first time.

Account architecture restructured by product category and COGS tier to enable margin-aware bidding.

Targets moved from revenue and ROAS to contribution margin after COGS.

Product feed data improved to support Shopping and Performance Max across the drinkware range.

Result

Contribution margin increased 94% and revenue 86%, measured against the preceding equivalent period after correcting the double-counted conversions.

Because the baseline was restated first, the improvement is a genuine commercial gain rather than an artefact of better-looking reporting.

What we learned

An account cannot be optimised faster than its measurement is honest. Correcting the numbers looked like a step backwards for two weeks and was the reason everything after it worked.