Madar AI
Skip to content

Case study · 90-day recovery

An independent Turkish AI app studio. $1,200 monthly revenue. A collapsing ad campaign. 90 days later: $5,400 MRR, positive ROAS, and the Japan campaign back from the dead.

MRR
+350%
ROAS
−62% +147%
Return
6.31×
Payback
14d

The 12 findings

What the audit found in 22 hours

Each circle is one diagnostic finding. Radius is proportional to monthly dollar impact. Color encodes category. Tap or focus to read the forensic detail.

  • Conversion events
  • Integration health
  • Value optimization
  • Creative
  • Retention
  • Strategy
  • Bidding
  • Hygiene

12 findings · $4,200 per month at stake · all detected in 22 hours

The recovery curve

$1,200 → $5,400 in 90 days

Each vertical line marks an intervention. The stroke shifts color as ROAS crosses zero on day 14. Toggle layers to compare MRR, ROAS, and week-2 retention on the same axis.

View

Day by day

Ten moments that mattered

The recovery in ten scenes. Scroll horizontally or arrow-key through them.

Skip the timeline

Scene 1 of 10

Day 1
+$1,200/mo

Orphan conversion events disabled

The three deprecated event names came off Google Ads' optimization target. Algorithm started bidding for goals that actually existed in the app.

Scene 2 of 10

Day 2
+$900/mo

AppsFlyer cost stream re-authenticated

Cost reporting resumed after four silent months. ROAS dashboards started telling the truth.

Scene 3 of 10

Day 4
+$650/mo

SKAdNetwork CV mapping revised

Conversion-value postbacks now reflect actual purchase tiers, not a flat-bucket placeholder.

Scene 4 of 10

Day 7
+$290/mo

ATT consent moved before paywall

Postback volume up 4× within 48 hours. The signal was always available — it was just gated behind the wrong screen.

Scene 5 of 10

Day 8
+$540/mo

Japan paywall restored

The free-forever experiment was reverted. Every new Japanese install now sees the standard offer. +$540/mo recovered immediately.

Scene 6 of 10

Day 11
+$480/mo

Predicted LTV signal enabled

Daily LTV uploads from RevenueCat into Google Ads. Bidder finally knows which installs are worth $30 and which are worth $0.40.

Scene 7 of 10

Day 14

Google Ads relearned the model

ROAS turned positive for the first time in 4 months. The line crossed zero on day 14.

Scene 8 of 10

Day 21
+$380/mo

Creative remixed for current features

Three new creatives matched the actual top-engagement feature. Day-1 churn dropped 18 points.

Scene 9 of 10

Day 60

Japan campaign earned its first dollar in 6 months

After the paywall fix, the SKAN signal stabilized, and the bidder targeted Japan installs again. The market opened back up.

Scene 10 of 10

Day 90

Day 90: $5,400 MRR. 6.31× return.

Payback achieved on day 14. The recovery is bedded in.

Your mirror

What would your numbers look like?

Adjust your current MRR, ad spend, and market. The model projects an Okkes-anchored 90-day trajectory. Not a guarantee — a frame of reference.

$500 $50,000
$500 $20,000

Based on the Okkes case study trajectory adjusted for your inputs. Actual results vary by product-market fit, market conditions, and execution discipline.

Projected MRR after 90 days

Estimated return

Estimated payback

Your projected upside is larger than Okkes's actual recovery.
Customers at this scale typically need the Growth tier ($999/mo) for the full automation suite.

Share this case study

Pass it on to a founder who's staring at the same problem.

From the founder

From the founder

I spent hours every day managing ads that didn't work, with no idea why. In 22 hours, Madar gave me a deeper diagnosis than any consultant had in 6 months.

Starting with Madar was simpler than I expected. I connected my accounts on Monday morning, and by Tuesday evening the diagnostic report was in my inbox. The 12 findings it surfaced were all in my account — and I had never seen any of them, despite looking at these numbers every day for months.

The hardest moment was finding out about the Japan paywall accident. My app had been losing $540 a month for four months without me knowing. It was an uncomfortable feeling, but also a relieved one: now I knew.

Ninety days in, the numbers speak for themselves. But what matters more: I sleep at night now. Madar watches the account every night. If anything's off, I know by morning. That's the thing I couldn't buy from any human consultant at any price.

The receipts

How we know these numbers are real

Skepticism is fair. Here's exactly where every number came from, how we verified it, and what we did not change.

About this case study

The customer is anonymized as Okkes Cetinkaya at their written request. All metrics in this page have been verified against the customer's own platform exports and cross-referenced with Madar's Diagnostic Engine output.

Time period

March 1 to May 30, 2026. Ninety days. No extensions, no cherry-picked windows.

March 1, 2026 May 30, 2026

Data sources Source verified

  • Google Ads March 1, 2026
  • AppsFlyer March 1, 2026
  • RevenueCat March 1, 2026
  • App Store Connect March 1, 2026
  • Google Ads May 30, 2026
  • AppsFlyer May 30, 2026
  • RevenueCat May 30, 2026
  • App Store Connect May 30, 2026

Verification

Independently verified by the Madar Diagnostic Engine on March 1, 2026, and cross-referenced with the customer's platform exports on May 30, 2026. Sample verification screenshots available on request.

What we changed

All 12 interventions are listed in the diagnostic constellation above. Each is traceable to a Google Ads change log entry, RevenueCat audit-log entry, or AppsFlyer integration ticket.

  1. #1

    Orphan conversion events still optimized

    Disabled the orphan events in the Google Ads conversion settings. Confirmed event-stream parity with the app's RevenueCat ledger.

    $1,200/mo
  2. #2

    AppsFlyer cost integration broken

    Re-authenticated the AppsFlyer ↔ Google Ads cost stream. Backfilled the missing 4 months for reporting only — not for billing.

    $900/mo
  3. #3

    SKAdNetwork CV mapping suboptimal

    Re-mapped CVs to a value-stratified 64-bucket scheme aligned to RevenueCat ARPU tiers. Postback richness restored.

    $650/mo
  4. #4

    Japan paywall configuration accident

    Reverted the Japan paywall to the standard variant. Disabled the experiment. Documented the rollback in the RevenueCat audit log.

    $540/mo
  5. #5

    No predicted LTV signal to algorithm

    Wired a tROAS bid strategy with a custom predicted-LTV signal pushed daily from RevenueCat into Google Ads via offline conversion uploads.

    $480/mo
  6. #6

    Creative-to-feature mismatch (3 ads, 1 campaign)

    Replaced the three off-feature creatives with three new variants showcasing the current top-engagement feature. Maintained brand kit consistency.

    $380/mo
  7. #7

    ATT consent flow misconfigured

    Moved the ATT prompt to the third-screen onboarding step, before the paywall. Postback volume up 4× within 48 hours.

    $290/mo
  8. #8

    No anomaly detection (silent churn signal)

    Madar's anomaly agent now monitors retention deltas per cohort nightly. Threshold: >10% WoW decline pages a Slack alert with the cohort details.

    $240/mo
  9. #9

    Single market dependency (USA only)

    Reallocated 12% of spend to Japan + DE markets as a hedge. Japan's CAC turned out to be 40% lower than US — opened a strategic discussion.

    $180/mo
  10. #10

    Bid strategy stale (90+ days no change)

    Migrated to tROAS with the new LTV signal. Set target at 130% with a 7-day evaluation window. Algorithm regained pricing discipline.

    $140/mo
  11. #11

    Negative keywords not inherited from account level

    Force-inherited the account negative list. Added 18 campaign-specific negatives from the actual search-terms report.

    $110/mo
  12. #12

    Conversion window mismatch (7d-click vs 14d behavior)

    Extended the click conversion window to 14 days. Re-ran the post-attribution analysis. Paid channel's true contribution +28%.

    $90/mo

What stayed the same

No new ads were created. No pricing changes. No product launches. No new markets opened (Japan was already a configured market — only the paywall was restored). No external PR or marketing campaign.

Variance disclosure

Results from this customer should not be interpreted as guarantees for other customers. Mobile app growth is multi-variable; outcomes depend on product-market fit, market conditions, competitive landscape, and execution discipline.

Your turn

Find your own 12 findings

Connect your accounts. Get the diagnostic report in 22 hours. Decide what to do next.

← All case studies