Why Most Apps Hit a Revenue Ceiling (and How to Plan for It) — Patrick Falzon, The App Shop

Screen each app or category on three lenses: (1) size: the product must plausibly reach tens of millions of dollars in revenue, avoiding a long tail of zombie apps doing a few hundred thousand each; (2) competition: a 'Goldilocks' market where you can take real share, e.g. RoboKiller was the leading independent spam blocker with carriers not invested and no big adjacent entrants; skip VPNs (everyone spends heavily on paid, so unit economics fail) and personal finance/budgeting (banks bid on the same keywords with far higher LTV); (3) existing demand vs. new category: existing demand allows ASO and performance marketing to pay back right away, while a new category means running at a loss to build demand.

Accept that most apps cap out when top-3 keyword rankings max out organic and CAC hits the (usually sub-$100) subscriber LTV, after which only single-digit growth comes from optimization. Grow by adding products instead: e.g. three $30M apps rather than one $90M app, with diversification as a bonus. A portfolio with a consolidated tech stack is also what makes a business sellable to private equity.

Before building, talk to users about what they want. Then ask whether the idea is a feature or a product: on a 6-inch screen, eight features confuse users, so prefer two or three slim apps with two or three features each and build a net-new product when there is demand. Every feature that does get built must be labeled up front as either an ARPU feature (e.g. an upsell) or a retention/engagement feature, because they are built and marketed differently.

Integrated a data broker removal service into RoboKiller as an upsell: sell users into the base spam-blocking subscription first, then offer the add-on, since it solves a related privacy problem.

Early in a product's life, run a range of LTV scenarios (e.g. plus or minus 20% in 5% steps), assign probabilities, pick a best estimate and act on it. Re-check every 1-2 months as renewal data comes in, dropping scenarios (six to four to two) until one precise estimate remains, which takes multiple years. At Mosaic's scale an in-house model could predict the next ~20 renewals from the first one or two, since most churn happens in the first renewals.

Always cap LTV at a set number of years (e.g. start at 2 years, extend to 3-4 as the company matures) because you must get paid back and future dollars are worth less. Early-stage: payback period is the primary filter (make money back in month 1 or within 3 months), with LTV:CAC as a secondary check; a high LTV:CAC under a tight payback signals that finding liquidity would unlock growth. Later-stage: manage to an LTV:CAC threshold (e.g. 2-3x) and treat payback (possibly 12+ months) as a sanity check.

Annual retention isn't known when a price test ends, so use the monthly plan's retention change as a proxy, or past price rises ('last time I raised annual 10%...'), and trial-to-paid as a leading indicator. Then sensitize: would the winner still win if retention comes in 10%, 20% or 50% worse? If only a ~5% buffer, don't roll out or roll out cautiously; if it survives a 50% hit, ship it. The proxy is weakest when one plan dominates (e.g. a 70/30 annual/monthly split).

Across Mosaic apps, price increases lowered retention less than one-to-one: roughly a 10% price rise cost a couple percent of retention, so LTV usually rose if conversion held. Sensitivity differs by product; apps with a thin layer of value have a lower price ceiling than deep-content products like language learning.

Tested early cancellation (turning off auto-renew) as a predictor of churn for LTV modeling. Trial-to-paid conversion and usage of the one key value action proved better leading indicators.

Pick one or two key metrics per product (more creates noise) that show the product delivering its value. For RoboKiller it was passive: is the user getting spam calls and are we blocking them, plus 'did we get one wrong?' For TapeACall it was an action: did they record a call. A misclassified call became the trigger for proactive retention outreach.

Apple limits on user-level call data and the company's own privacy layer made personal stats hard, so RoboKiller showed stats across the whole user base (millions of calls blocked this month) to prove the background app was working.

As the largest US spam blocker, RoboKiller had unique data on spam calls and texts. It published regular reports (how many spam calls Americans get per month, how it is growing), worked with reporters to help them cover the issue, and shared data with the FCC and FTC, which lacked good data. No single article drove installs directly, but accumulated coverage built brand and organic lift.

Since Google and Meta are black boxes where you only set bid and budget, creative is the main lever. Always keep a share of creatives in each campaign experimental/unproven. Because every designer has a personal style that leads to 5-10% tweaks, rotate creative team members across products every couple of months (within brand guidelines) and occasionally bring in a new designer told to 'go wild', then double down on styles that win.

Don't launch Google, Meta and TikTok together: launch one, let it reach steady state, then add the next against a new baseline. Once live, change one bid or budget, wait about a month, then the next. Ramp new channels conservatively because the first users are the cheapest and retention usually comes in worse than predicted. Measure several ways and look at them together: MMP channel reports at face value, blended CAC, paid vs. organic, and channel-specific CAC; pick one primary cost metric matching the model (CPI for ad-funded, cost per trial if using trials, cost per subscriber otherwise). Ground everything in blended performance, since that is the P&L.

Apalon, acquired by IAC during the paid-download era, moved its apps to in-app subscriptions early, then used that know-how to build a portfolio.