What you’ll learn

• The optimal 3-event conversion structure for Meta/Google (and why tracking more hurts performance)
• Why speed of event delivery is one of the strongest levers for match quality & cheaper CPAs
• How to incorporate value signals (trial filters, buckets, predicted value) without full LTV modeling
• Why using PII (hashed email/phone) dramatically improves attribution & optimization
• How to separate measurement vs. optimization data so each system actually does its job
• Lightweight ways to identify high-value users early and filter out low-quality trials
• Why Meta-reported ROAS doesn’t matter unless your business metrics move too
• How to diagnose whether you have a strategy problem or a measurement problem
• Why small apps should use holdouts & blended metrics instead of over-complicated attribution setups
• How fast event sending helps platforms reconnect the full click → browser → app → purchase chain

Key Takeaways

• Keep it to ~3 conversion events. Event tracking is “free,” but every extra event adds maintenance, confusion, and breakage. For ad platforms, you rarely need more than:
  1. a top-funnel/engagement event (e.g. survey completion),
  2. signup/registration (first PII),
  3. trial start (earliest strong revenue proxy).
• Design the event ladder from value, not vanity. Early events show intent; signup lets you pass PII; trial start is the closest thing to revenue that usually falls inside platform lookback windows.
• Fire events fast. The shorter the delay from click → event, the easier for Meta/others to probabilistically match user journeys. Even within a 24-hour window, “the faster, the better.”
• Include value data, but don’t over-engineer LTV. For subscription apps, the actual charge often happens after the lookback window. You don’t need a perfect 2-year LTV model—start by bucketing users (e.g. worth 0 / 5 / 10 / 20) based on early behavior and use that as a value signal.
• Predictive value is about ranking users, not forecasting to the penny. The goal is: out of 100 trials, which ~30 are most likely to convert? Use early feature usage (first 24–48 hours), plan views, return sessions, etc. to distinguish high- vs low-value users.
• If you don’t send value, platforms optimize for cheap installs. Without a quality or revenue proxy, bid models will chase the lowest-CPI users—often low-intent segments like teens—at the expense of payers.
• Deduplicate client + server events on purpose. If you send the same “signup” from multiple sources (SDK, MMP, CAPI), use a deduped “master” event for optimization and keep source-specific events for troubleshooting. Check that SDK_signup + CAPI_signup roughly add up to the unified event.
• Pass PII where you legally can. Emails, login IDs, names, location, and device info (when allowed) greatly improve matching and attribution—especially now that IDFA and deterministic links are limited. Always align with privacy law + platform policies.
• Separate optimization data from decision data. Events in Meta/Google exist primarily to help their algorithms optimize—not to give you perfect causal measurement. Use them for bidding & creative testing, but use incrementality tests and holistic metrics to decide budget allocation.
• Don’t mistake a strategy problem for a measurement problem. If you’re a small app running many channels with tiny budgets and can’t tell what works, the issue is fragmentation—not that you need fancier attribution.

Links & Resources

• Fix My Tracking: https://fixmytracking.com/
• 021 Newsletter: https://www.021newsletter.com/
• Barbara Galiza on LinkedIn: https://www.linkedin.com/in/barbara-galiza