Clients are anonymized. Most would rather not advertise who buys
their media, and one made it a contract term. Every figure below
states the window it was measured over. None of it is a promise that
your account does the same.
E-commerce / D2CDE, FR, NL
A European D2C skincare brand
The account had plateaued at roughly €180k a month in spend. Every
attempt to push past it moved blended CAC in the wrong direction,
and nobody could tell whether new campaigns were finding new
customers or re-buying the existing ones.
What we did
Collapsed 34 ad sets into 4 and let the auction segment the
audience.
Rebuilt the event pipeline server-side, which turned out to be
dropping about one purchase in six.
Ran a geo holdout for six weeks to get an incrementality read
that didn't depend on the platform grading its own
homework.
Result
Metric
Change
Window
Monthly spend
€180k → €540k
over 7 months
Blended CAC
−19%
same period
Incremental ROAS
2.6x
geo holdout, 6 weeks
Monthly media spend
€180kBefore
€540kAfter
The ceiling was never the audience. It was a tracking gap that
made half the working creative look like it was losing.
Install volume was fine. Funded accounts were not. The team was
optimizing toward a registration event that a third of users never
moved past, so the media was buying the wrong person efficiently.
What we did
Moved the optimization event down the funnel from registration
to first deposit, and ate the two weeks of learning-phase pain
that came with it.
Split brand and non-brand search, which exposed that brand was
absorbing credit for paid social demand.
Rebuilt creative around a single objection: whether the money
is safe.
Result
Metric
Change
Window
Cost per funded account
−34%
quarter over quarter
Funded rate from install
11% → 19%
90 days
Paid search waste
−£41k
annualized
Funded rate from install
11%Before
19%After
Optimize for the event that pays the bills, even when it makes the
dashboard look worse for a month.
Native and programmatic had been written off internally as a channel
that only produced junk trials. The blocklist hadn't been
touched in a year, and the landing page was the product homepage.
What we did
Pruned supply weekly and cut roughly 600 placements that had
never produced a qualified trial.
Built three advertorial landing pages, each arguing a
different use case rather than describing the product.
Passed lead quality back into the DSP so the bidding optimized
for sales-accepted, not for form fills.
Result
Metric
Change
Window
Sales-accepted leads
+2.1x
vs. previous 90 days
Cost per SAL
−28%
same window
Junk trial rate
41% → 12%
60 days after pruning
Junk trial rate
41%Before
12%After
Native doesn't have a traffic quality problem. It has a supply
hygiene problem and a landing page problem.