Mobile App Funnel Bottlenecks: The 7 Conversion Points Between an Ad Dollar and a Retained Payer

The short answer: seven gates, checked from the top down

Mobile app funnel bottlenecks are almost never in one place, and they are almost never where the team is looking. Between a marketing dollar and a paying, retained user there are seven gates: marketing to store view, store view to download, download to first open, first open to onboarding done, onboarding to trial start, trial to paid, and then retention. Each gate has one number to watch, one dominant failure mode, and a fix that is usually not the fix for the gate below it.

Check them top down. A leak at gate 1 or 2 burns spend before it shows up in any downstream metric, so plugging it multiplies every dollar already committed to the channel. A leak at gates 5 to 7 means the opposite problem: you're paying to acquire users the product can't convert or can't keep. Same dashboard, opposite work.

One item on the list is not a bottleneck at all. Apple and Google take a cut of every transaction, and refunds land after that. That's a haircut between billing and revenue, not a conversion problem, and it belongs in the math before you compare a channel's CAC to LTV. More budget never fixes a broken stage. It buys more of the same leak, faster.

Gates 1 to 3: from the ad impression to the first app open

Gate 1 is marketing to store view. Watch tap-through, or click-through on paid. A weak rate here means the hook is wrong or the audience is. On Apple Search Ads specifically it's usually a keyword-relevance problem and not a bid problem, so raising the bid buys more of the wrong query. The fix is creative volume before scale. One structural constraint worth knowing: a Meta ad set needs roughly 50 optimization events in a 7-day window to exit the learning phase, and below that threshold CPAs run 20 to 50% above the post-learning average. Optimize for trial starts rather than installs, because the delivery algorithm returns whatever user type you asked it to find.

Before you declare gate 1 dead, check that you can measure it. iOS ad performance is reconstructed from three systems that disagree with each other. With about 27% global ATT opt-in, Meta-reported ROAS understates true performance by 20 to 40%, and TikTok's last-click attribution has been reported to undervalue by up to 10.7x (figures as reported, no underlying dataset cited). A channel that looks unprofitable in-platform is sometimes just badly attributed.

Gate 2 is store view to download, and it's the highest-leverage fix on this whole page, because it sits above every dollar already spent to get the impression. Three levers, in order of how often they're neglected. The icon is the tap gate: bold color, one clear subject, readable at 60x60 points, and checked for differentiation against the top 20 results for your target keyword (practitioner heuristic from @seraleev, no dataset attached). The first three screenshots have to tell a story a scrolling user can follow, not showcase features one at a time. And the autoplay app preview video is the most underused lever of the three, with install-rate lift reported up to +25% on product-page visits versus screenshots only (single practitioner note, no dataset size given). Custom Product Pages are the newer unlock: they began appearing in organic search results in July 2025, you get up to 70 per app, and CVR lifts +5.9% when the creative tightly matches keyword intent. A fitness app can run separate organic store pages for running, strength training, and yoga.

Gate 3 is download to first open, and it's the one nobody instruments. The brain has no benchmark for it, so be honest about that and go measure your own: App Store Connect reports downloads, your analytics reports first opens, and the gap between them is the leak. The usual causes are an impulse install that never gets returned to, a slow cold start, or a crash on launch. Check crash-free rate and cold start time before you touch anything creative. Then check the promise: if screenshot 1 sells an outcome that screen 1 doesn't deliver, the store page is writing a check the first run bounces.

Gate 4: first open to onboarding done, where Day 0 decides almost everything

Watch completion rate per step, not the aggregate. The aggregate tells you 40% of users finish onboarding. It does not tell you that 22 of those 60 lost points leave on one permissions screen. Instrument every screen as a funnel step with an entry event and an exit event, the way onboarding is already measured, because the same leak-finding logic works between any two screens in the app.

The reason this gate carries so much weight is timing. Across 115,000 apps and more than a billion transactions, 78 to 90% of trial starts happen on Day 0, and about 50% of all paid conversions happen on Day 0 (Productivity hits 71.9%). Adapty's 2026 dataset of 20,000+ apps lands in the same place from a different angle: 90% of conversions happen within 24 hours of install. You are not buying users. You are buying 24-hour windows, and the whole of the first-session window is spent inside onboarding.

The most consistent leak at this gate is a hard gate placed before the user has reached the value moment. It converts your highest-intent users worst, because they never got to feel the thing they downloaded the app for. SuperChinese gates free access after 2 to 3 lessons, but its actual aha moment needs a completed speech-recognition exercise to reach, and the SuperChinese teardown estimates roughly $800K left on the table from that one decision on an app doing $70K a month. The counter-case is Deepstash, whose 14-screen onboarding is built entirely around delivering the aha before any commercial ask, at $200K monthly revenue on 90K downloads. The fix is never a better paywall. It's moving the gate.

The practitioner shorthand for this gate is that onboarding is 70% of the app (assertion from @Jahjiren, no dataset behind the number). Treat the 70% as a priority signal rather than a measurement. What's measurable is that a user who does not feel value in the first session almost never comes back to find it, which is what makes the aha moment the thing to engineer here and screen count a downstream detail.

Gate 5: onboarding to paywall view to trial start, the rate that predicts revenue

Watch trial-start rate, and watch install-to-paywall-view above it. Adapty's read on its 20,000+ app dataset is that install-to-paywall-view is the single metric that predicts revenue best, with revenue moving almost linearly against it. Double that rate and revenue roughly doubles. It's also the rate most teams never look at, because it sits between two screens nobody owns.

Benchmark trial starts against your category, not against a global number. Download-to-trial at D30 runs 9.1% for Business, 6.9% for Health & Fitness, 6.5% for Education and Utilities, 4.4% for Gaming, and 4.0% for Media & Entertainment. Geography moves it as much as category does: North America 7.1%, APAC 5.7%, Western Europe 5.0%, and 3.0 to 3.7% for the rest. Price tier moves it in the direction most founders don't expect, with high-priced apps at 8.9% against 4.4% for low-priced ones (RevenueCat SOSA 2026, medians).

Those medians matter for one decision in particular. The operator gate for turning on paid ads is at least a 10% download-to-trial rate, below which you're paying to fill a leak (Mau Baron, Prayer Lock, founder playbook). Put that next to the table above and the demand is clear: 10% sits above the median of every single category. Paid math needs an above-median funnel, not an average one, which is the whole argument in when to start paid ads.

The failure mode at this gate is a paywall with wrong timing, unclear pricing, or no reason to act now. Test placement and price framing as separate variables rather than shipping a redesign that changes both. On the access model, the dataset is blunt but the variance is the real story: D35 download-to-paid is 10.7% for hard paywalls against 2.1% for freemium, and the hard-paywall range runs from 4.2% to 38.7%. The label is not the lever. A hard paywall shown after enough onboarding to make the user feel the product reads as a conclusion, and the same wall shown cold is just a wall.

Gate 6: trial to paid, usually lost on Day 0 and not at expiry

Watch trial-to-paid conversion, and watch when the cancellations happen. This is where counting trial starts as a win goes wrong. Trial-to-paid by category runs 43.5% for Travel, 37.7% for Health & Fitness, 25.0% for Gaming, and 22.2% for Photo & Video. By geography it's 34.2% in North America, 29.7% in Western Europe, and 15.2% in India and Southeast Asia. A single global median hides a 2 to 3x spread, so a number that looks broken against the median can be fine for your segment, and the reverse is just as common.

Trial length is the lever with the cleanest data attached: trials of 4 days or less convert at 25.5%, 5 to 9 days at 37.4%, and 17 to 32 days at 42.5%. Longer converts better, and the honest counterweight is that shorter trials compound experiment velocity and shorten the cash cycle, which is why plenty of good teams shorten anyway. The trade is optimization goal against learning speed, not right against wrong. Free trial versus no trial is the prior decision.

Now the timing detail that changes what you build. 55.4% of 3-day-trial cancellations happen on Day 0, and 84% happen by Day 1. The user starts the trial, and cancels in the same session, usually right after the purchase screen. A reminder before the trial ends is table stakes and it arrives days after the decision was already made. The higher-leverage work is making the value obvious on day one, then daily, rather than at signup. Hard paywalls do get a second spike at D4 to D7 (25.7% of conversions, trial expirations), while freemium keeps a long tail where 23% convert six or more weeks out, so the reporting window you choose changes the answer you get. The trial-to-paid benchmark breaks the full table down.

Not a bottleneck, a haircut: the store cut and the billing nobody watches

Apple and Google take a cut of every transaction. It's commonly 30%, dropping to 15% under the small-business and post-year-one subscription programs. That's store policy rather than a benchmark, so treat the exact rate as something to check against your own payouts, not something to model from a table. The brain carries the 30% figure as the thing web funnels exist to route around: 41% of the highest-revenue apps run web revenue against 1.3% of hobby apps, a 31x adoption gap, and the web-to-app funnel is the playbook.

This is not a conversion problem and there's no fix to ship. It's arithmetic between billing and revenue, and the mistake is leaving it out of the comparison. Net it out before you grade a channel's CAC against LTV, or you're measuring cost against money you never actually collect.

The quieter version of the same haircut is involuntary churn. About 31% of Google Play subscription cancellations are involuntary billing failures, against about 14% on the App Store. Nearly a third of your Android churn is an expired card, not a decision, and recovering it needs grace periods, retry logic, account hold, and update-payment prompts rather than a product change. Refunds and chargebacks are the sibling problem: they hide the downstream cost of a paywall experiment that looked like a win, which is why winning cohorts are worth rechecking at 3 and 6 months (Sara Grana, Yousician).

Gate 7: retention, which sets the LTV you just compared CAC against

Watch D1, D7, D30, and renewal rate. The bottleneck here is a product that doesn't earn a second visit, or an ad that promised something daily use doesn't deliver. The fix is a habit loop with a reason to return, and the reason has to be the core output being right. In priority order the drivers are accuracy of the core output, a frictionless core loop, and a bug-free experience, in that order and not the order of a feature roadmap.

For a reference point on D7, one tracked small app sits at 13% at 5 months old and $2.8K MRR (self-reported, one app, directional rather than a category benchmark). Expect it to vary heavily by category, and note that the 300-installs-a-day threshold applies before your own D7 is statistically reliable at all.

One dataset finding settles an argument that comes up at this gate. Year-1 retention is nearly identical between hard-paywall and freemium apps. The access model buys you conversion, not durability. Whatever you do at gate 5 does not save you here, and retention stays a product problem.

Then the scale check, because it decides which benchmark you should even be chasing. Among new apps in their first two years, 17.3% reach $1K MRR, 4.6% reach $10K, and 1.7% reach $25K. Median monthly revenue one year after launch is around $72, while the top 10% clear $2,574. The distribution is a power law, so benchmarking against the median tells you how to be typical. Benchmark the upper quartile of your category instead.

How to run this diagnostic on your own app

One honest disagreement sits underneath all of this, and it's worth knowing which side you're on before you start. Steven Cravotta's order is paywall first, then onboarding, then marketing, on the logic that scaling spend into an unoptimized funnel just scales a leak. Mau Baron's order is the reverse: don't touch the app at all until the distribution engine is running near 300 daily downloads, because there is no point optimizing a funnel nobody enters. Both are single-operator playbooks, and both operators got where they were going.

They agree on more than they disagree. Neither will scale paid spend into a funnel converting below 10% of downloads to trials. The deciding variable is your app's phase and whether the traffic is organic or paid. A brand-new app posts and ships until roughly 300 installs a day, because below that a 2-week A/B test on your paywall returns noise instead of signal, while at 300 or more tests reach significance in 5 to 7 days. A live app in steady state dispatches on one number: today's downloads against the trailing 7-day average. Below it, the constraint is attention, so go market. At or above it, the constraint is conversion, so go work the gates.

• Write down all seven rates before forming an opinion about any of them. Missing instrumentation at a gate is itself a finding
• Pull your category's benchmark for each rate, not the global median. Every step of this funnel varies 2 to 5x by category, geography, and price tier
• Fix top down. A gate 1 or 2 leak multiplies every dollar already committed downstream, and a gate 5 to 7 leak means you are buying users the product cannot keep
• Instrument every screen with an entry and an exit event. An aggregate completion rate hides which single step is doing the bleeding
• Split your Day-0 numbers out. 78 to 90% of trial starts and 55.4% of 3-day-trial cancellations happen on Day 0, so a monthly average tells you almost nothing about either
• Net the store cut, refunds, and involuntary churn out before comparing any channel's CAC to LTV
• Don't A/B test the funnel below 300 installs a day. Go get traffic first, then the tests will return signal
• Benchmark the upper quartile of your category, not the median. 4.6% of new apps reach $10K MRR in two years, so typical is not the target

The point of the checklist

The seven gates are a checklist, not a strategy. What they buy you is the ability to name which one is actually broken before you spend another month on the wrong one. If you want the benchmark for a specific gate while your AI coding agent is building the screen, tasu is the MCP it connects to for sourced onboarding, paywall, pricing, and retention claims. It's at /mcp.