02Case File 02 · Real-Money Casino

Accounts Are Consumable.
Deposit Learning Is Not.

A fresh advertising account could die within two hours. That was never the expensive part. The expensive part was that everything the delivery system had learned died with it.

VerticalReal-money casino
ChannelMeta
Window9 weeks after rebuild
Spend$94,800
Baseline2 weeks, $21,600

What happened

Quick answer

Over nine weeks and $94,800, registration to deposit moved from 2.4% to 8.8%. The change was not creative or targeting. It was separating what is consumable from what must survive.

The 2.4% starting figure is not a benchmark we are proud of. It is a snapshot of a broken process — duplicated postbacks, review-driven buying and learning that reset weekly.

The numbers

Before And After,
Same Budget

MeasureBeforeAfterNote
Ad spend$21,600$94,800―Baseline vs rebuild window
Registrations1,9404,760▲Volume grew with budget
First-time deposits47418▲789% more
Registration to deposit2.4%8.8%▲3.7x
Recovery after a disableDaysHours▲Structure, not luck
Learning after a disableFrom zeroResumes▲The single biggest change

Diagnosis

What Was Broken

ISSUE 01Postbacks were leaking and duplicatingSome deposits never reported. Others reported twice. The account was being optimised on a signal that was partly fiction.
ISSUE 02Every disabled account reset the learningAccounts were treated as the asset. They are the consumable. What had to survive was the conversion history, and it did not.
ISSUE 03Buying was driven by what passed reviewCreative was selected for its chance of approval rather than its chance of producing a deposit. Approval is a precondition, not a result.
ISSUE 04Registration was the reported KPISo registration is what the account produced, in quantity, at a rate nobody could convert.

The work

What We Changed

CHANGE 01Fixed the signal before touching the campaignsDeduplicated postbacks and confirmed every deposit reported exactly once. Nothing else matters while the input is wrong.
CHANGE 02Made the deposit the optimisation eventRegistration stayed visible as diagnostics. It stopped being the thing the budget was paid to produce.
CHANGE 03Detached learning from the working accountsA disabled account became an operational inconvenience of hours rather than a project of days.
CHANGE 04Prepared replacements before they were neededVerified and warm, sitting ready. The gap after a stop is almost never the account itself — it is having nothing to put in its place.

The uncomfortable part

What Did Not Work

Published because a case file without this section is an advertisement. These cost time and money before they were abandoned.

⚠Abandoned after testing

Trying to keep accounts alive longerWeeks went into hardening accounts that were always going to stop. The gain was marginal. The same effort spent on faster recovery returned far more.
Optimising toward deposit in week oneDeposit volume was too low at the start to leave the learning phase. We ran a middle event for eleven days until volume supported the deeper one, then moved.
A high-frequency creative rotationRefreshing creative every 48 hours restarted learning as often as the bans did. Slowing the rotation improved stability more than any individual asset did.
Buying prepared accounts from more suppliers to spread riskReasonable in theory. In practice the failure rate rose, because supply quality varies far more than supply volume and we had no way to inspect it before use. Two suppliers with a known history outperformed six with an unknown one. Diversifying a supply chain only reduces risk when you can tell the good part from the bad part.

What it means

The Takeaway

→The advertising account is a consumable and should be treated as one. What cannot be consumable is the conversion history, the prepared replacement and the creative library. Build those and a disable costs hours.

Related

How this is structured, in fullOpen → iGaming, market by marketOpen → The same rebuild in forexOpen → Price a year of stopped deliveryOpen →
Discuss your account → All documented files