Early Proxies to Predict Cohort Success

Lucas Moscon, one of the most technically knowledgeable people in mobile attribution, breaks down how post-ATT measurement really works, why most marketers are using outdated mental models, and how to build a modern, resilient measurement stack. Lucas clarifies what’s deterministic vs probabilistic today, exposes where MMPs still add value (and where they absolutely don’t), and explains why IP-based fingerprinting quietly powers 90%+ of attribution today. He also walks through SKAN in plain English, conversion-value strategy, web-to-app pipelines, and why looking at blended ROI beats chasing ROAS illusions on iOS.

If you want to understand the actual mechanics behind click → install → revenue pipelines — and why Apple’s privacy tech is failing in practice — this episode is for you.

What you’ll learn:

• Why ATT didn’t “kill” attribution — it forced marketers to juggle deterministic, probabilistic, and blended layers
• How Meta/Google matching actually works (spoiler: 90%+ relies on IP, not magic AI)
• Why SKAN isn’t enough — and why relying on ROAS on iOS is the least trustworthy metric
• How to measure effectively without over-reacting to noisy campaign-level data
• When you truly need an MMP today — and why most apps don’t
• How to correctly design conversion values for SKAN without over-engineering
• Why retention determines how many conversion values you even receive
• How to triangulate data across store consoles, subscription platforms, MMPs, and ad networks
• Why focusing on payback windows (D60–D180) outperforms optimizing for short-term ROAS
• Why probabilistic fingerprinting is still powering the ad ecosystem — and why Apple hasn’t stopped it

Key Takeaways:

• iOS ROAS is the noisiest metric you can use. Without IDFA, everything is extrapolated. High-confidence decision-making must use blended revenue and cohort ROI, not ad-platform ROAS.

• Modern attribution = multiple layers. Post-ATT, performance requires triangulating data from SKAN, ad networks, subscription platforms, and product analytics — not trusting a single source of truth.

• Fingerprinting ≠ complex algorithms — it’s mostly IP. Internal tests showed that greater than 90% of probabilistic matches come from IP alone. All the “advanced modeling” narratives are overstated. 

• Most apps don’t need an MMP anymore. Exceptions: running AppLovin/Unity DSPs, React Native/Flutter SDK support gaps, or complex Web-to-App setups where Google requires certified links. Otherwise, MMPs mostly add cost, not clarity.

• Retention determines SKAN visibility. If users don’t reopen the app, conversion values won’t update — meaning SKAN under-reports trials/purchases unless retention is strong.

• Blend deterministic + probabilistic + aggregated signals. The goal isn’t precision — it’s directionally confident decisions across imperfect data. Marketers should work in ranges, not absolutes.

• Longer payback windows unlock scale. Teams willing to accept D60–D180 payback dramatically out-spend competitors optimizing for D7 ROAS — assuming they have strong early-day proxies to detect failing cohorts.

• MMPs don’t magically fix discrepancies. Even with one SDK, marketers still see mismatches across networks, stores, and internal analytics. The “one SDK solves it” narrative is outdated.

Links & Resources

• Appstack: https://www.appstack.tech/
• Appstack library of resources: https://appstack-library.notion.site/
• Lucas Moscon LinkedIn: https://www.linkedin.com/in/lucas-moscon/

00:00 Opening Hot Take: “Are You Really Saturating Meta?”
05:00 Early Indicators & Proxy Metrics (D3–D10)
09:00 Predicting Cohort Success from Day 3–10
11:00 How Click → Install Attribution Actually Works
14:00 Web-to-App Infrastructure (Fingerprinting + SDK Flow)
18:00 Meta/Google Matching: IDFA, AEM, SKAN
24:30 Fingerprinting Reality: Why IP = 90% of Matches
27:00 Apple’s Privacy Messaging vs Actual Enforcement
30:30 How Apple Ads Uses (or Ignores) SKAN
35:00 Should You Use an MMP in 2025?
46:00 SKAN Conversion Value Mapping: The 63/62 Strategy
49:00 Why Retention Determines SKAN Postbacks
54:00 App Stack Overview + Closing Thoughts