I've reviewed accounts where installs were up 40% and the founder was still worried. They were right to be.
The MMP chart looked great. Trial starts were flat. Payback period had blown out. ASA was buying brand terms that would have converted organically. Meta was optimising for installs, not purchases.
Vanity metrics feel good in weekly standups. They don't tell you if marketing is building a business.
Here's the stack we use to measure app marketing success - and the traps that catch teams who stop at install volume.
Layer 1: Store health (before you blame channels)
Before interpreting campaign data, confirm the listing isn't leaking:
| Metric | Where to find it | What "good" looks like |
|---|---|---|
| Product page conversion rate | App Store Connect / Play Console | Often 25–40%+ for strong categories; below 20% = creative or positioning problem |
| Impressions → page views | Store analytics | Low page views despite impressions = metadata / ranking issue |
| CPP conversion by source | App Store Connect (2026 analytics) | Organic vs ASA vs web referrer split |
| Rating velocity and sentiment | Store + review tools | Stable 4.5+ with recent reviews |
If conversion is broken, every paid channel looks expensive. Fix ASO first or in parallel with paid - not after six months of scaling.
Layer 2: MMP truth (one source, deduped)
Use one mobile measurement partner - AppsFlyer, Adjust, Singular, Branch - and make it the single source of truth for:
- Installs by source / campaign / creative
- Cost per install (CPI) and cost per action (CPA)
- In-app events: registration, trial start, purchase, key activation milestones
- Retention cohorts (D1, D7, D30) by channel
- SKAdNetwork and Aggregated Event Measurement reconciliation on iOS
Rules we enforce:
- Same event names across iOS and Android
- Revenue events firing with correct currency and net/gross consistency
- Organic vs paid split trusted before scaling budget
- Re-engagement campaigns tagged separately from acquisition
If ASA and Meta both claim credit for the same install, your ROAS math is fiction.
Layer 3: Unit economics (the metrics that matter)
Installs are a means. These are the ends:
For subscription and consumer apps:
- CPI / CPA vs target by channel
- Install-to-trial rate and trial-to-paid rate
- ROAS or LTV:CAC at D30, D90, D180 (whatever matches your payback model)
- Retention curves by acquisition source - Meta installs that churn D3 aren't worth the same as ASA installs that retain D30
For B2B and high-ACV apps:
- Cost per qualified signup or cost per sales-assist lead
- Activation rate (completed setup, first value action)
- Pipeline influenced - connect CRM (HubSpot, Mailchimp, Salesforce) to app events
- Account expansion - seats added, not just first install
See SaaS B2B vs consumer app marketing for how targets differ.
Red flags:
- CPI down but payback period lengthening
- Install volume up, revenue flat
- One channel at 90% of spend with no incrementality test run in 12 months
Layer 4: Channel-specific benchmarks
Don't compare ASA CPI to Meta CPI directly - different intent.
Apple Search Ads: Track CPT, CPI, impression share on brand vs category vs competitor campaigns. Brand campaigns should be efficient; discovery campaigns validate new keywords.
Google App Campaigns: Separate Android UAC, iOS UAC, and YouTube-only where creative strategy differs. Watch modelled vs MMP-reported iOS installs.
Meta / TikTok: Optimise toward downstream events (purchase, trial), not link clicks. Creative fatigue shows up as CPI creep before frequency caps alert you.
Our comparison pages - Apple Search Ads vs Google App Campaigns, TikTok vs Meta for app installs - cover when each channel deserves budget; measurement should follow that same split logic.
How do you measure AEO and GEO for app marketing?
Keyword rank and organic clicks still matter - especially for Google AI Overviews, where a large share of citations come from pages that already rank well. But they no longer tell the full story. AI discovery needs a presence-and-influence layer alongside store and MMP data.
| Metric | What it measures | Rough benchmark | Cadence |
|---|---|---|---|
| Share of Model (SoM) | % of core buyer-intent prompts where your app is named | 20–40% = competitive; 40%+ = strong | Weekly |
| Citation frequency | How often assistants cite your site or name your brand | Trend up month-on-month; compare vs top 3 competitors | Weekly |
| Citation accuracy | Whether AI describes your category, pricing, and geo correctly | 85%+ of mentions accurate | Monthly |
| Sentiment / framing | Recommended as a leader vs "another option" | Qualitative scorecard on 10 core prompts | Monthly |
| Prompt coverage | Breadth of intent clusters where you appear (not just head terms) | Expanding quarter on quarter | Quarterly |
| AI referral traffic | GA4 sessions from chatgpt.com, perplexity.ai, etc. | Small but rising; watch trend not absolute volume | Weekly |
| AI referral CVR | Conversion rate of AI-referred sessions vs organic average | Often higher intent when volume is meaningful | Monthly |
| Branded search lift | Branded query volume in Search Console | Proxy when AI tools under-report; rising = growing mindshare | Monthly |
GA4 setup: custom channel group or exploration filtering AI referral sources; cross-reference landing pages (homepage, pricing, blog) and downstream events (trial start, apply, purchase).
Tooling: run the AI App Visibility Check quarterly for a 125-prompt baseline across five assistants, or track manually with a spreadsheet of 30–50 core prompts. Track by platform - ChatGPT and Perplexity often cite completely different sources for the same question.
Rankings and traffic aren't dead. They're inputs. The decision metric for 2026 is whether you're in the AI consideration set when someone asks for the best app in your category.
Full audit framework: How to check if AI is recommending your app. Content and structure guidance: App discovery in 2026.
Layer 6: Incrementality and holdouts
Sophisticated teams ask: would this install have happened anyway?
Tactics:
- Geo holdouts - pause spend in one region, compare organic + paid mix
- Brand ASA tests - reduce brand bidding, measure organic brand search impact
- Platform pauses - short, controlled off periods on a channel (not during launch spikes)
You don't need weekly incrementality studies. Run one disciplined test per quarter on your largest spend line item.
The one-page dashboard we'd actually use
If you forced me to pick six numbers for a fortnightly review:
- Blended CPA to primary conversion event (trial or purchase)
- Payback period or ROAS at target day by top two channels
- Store conversion rate (default + top CPP if iOS)
- D7 retention for last month's paid cohort
- Organic install share (trend, not single week)
- AI visibility spot check - 5 core prompts, yes/no on appearance
Everything else is drill-down.
Common measurement mistakes
Optimising for installs because they're easy. Platforms default to install optimisation because they have the data. You have to wire purchase and LTV events and force the algorithm to use them.
Ignoring organic. Paid spikes that don't lift organic installs or branded search often indicate you're buying clicks that would have happened anyway.
Weekly panic on CPI. Seasonality, creative fatigue, and store featuring move CPI. Look at 14-day rolling averages and cohort outcomes.
No event taxonomy document. When marketing calls it "signup" and product calls it "registration_complete," dashboards lie.
When numbers look wrong
If metrics conflict - MMP says ROAS is fine, finance says it isn't - audit in this order:
- Event mapping and revenue validation
- Store conversion rate change (did a release hurt listing?)
- Channel mix shift (did low-quality inventory enter?)
- Attribution window mismatch (7-day click vs 1-day view)
Want a measurement review?
If you're spending on ASA, Meta, Google AC, or ASO and can't answer "what's our payback by channel?" in one sentence, apply for a strategy call.
We can review your stack, tell you what's trustworthy, and what to fix before you scale another dollar.