Writing

I cofounded Subtext, an app that reads the tone of a message before you send it. What it does, what it will not do, and why a growth person built it.

One Meta account, $1.2M over 22 weeks, 722 pairs of ad weeks. ROAS predicted the next week best once an ad had 50 or more purchases, at a correlation of 0.44, which is still only moderate. How to find the threshold in your own account.

A subscription app on Meta, more than $1.2M in spend, 917 localized ads in 10 languages. Language explained 4.8% of the variation in ROAS. The creative concept and its variant explained 25.2%, more than five times as much. Test in English, localize the winners, validate each market.

From an account with more than $1.2M in Meta spend, and consistent with many others I have managed: ROAS bidding tends to win for weekly subscriptions, games with in app purchases and apps that sell credits. Purchase bidding can hold up for a single annual plan.

Two Google Ads tests on app campaigns, $73K across 6 campaign types and $27K across 6 bid strategies. Max ROAS beat tROAS on both platforms and every market. iOS led on Max Conversion Value while Android led on tD0ROAS and MaxAction. Search took 58% of budget and returned 25%, Display took 32% and returned 43%.

A tROAS campaign kept scaling five ads that never reached the 30% D0 target. We duplicated the campaign without them and ran both for 30 days at comparable budget. The original top five hit target 0 out of 5 times, the test campaign 3 out of 5, and the algorithm surfaced six new creatives above target.

A casual game opened at $4 CPI and 27% D0 ROAS, then reached the same D0 again six weeks later with completely different users. When D0 halved from 22% to 11%, CPI was responsible for 11% of the drop. AOV decline accounted for 62%. How to decompose D0 ROAS into conversion, AOV, and CPI before blaming acquisition.

The case for keeping a trial and the case for cutting it, including the channel question nobody asks: whether a source with high intent like Apple Search Ads should be treated the same as a discovery source like Meta or TikTok. Plus the Meta constraint that decides it, since you cannot cleanly optimize for trial starts and purchases at once.

Six criteria that separate a strong subscription app consultant from a weak one, and how I score against them. I run paid UA, signal engineering, creative and funnel audits as one lead strategist for apps spending $100K a month or more. Named results: EatBetter to $180K MRR, Videa AI past $200K monthly spend at 43% D0 ROAS.

Two reports that catch iOS install misattribution: App Store Connect search installs moving with spend, and organic installs from the MMP plotted against paid. 103 days on one account: iOS slope 0.277, about 28 of every 100 paid iOS installs reappear as organic. Android slope 0.059, the normal halo of a paid campaign.

A health app account, 209 days, 346,317 installs. Sunday and Monday installs convert best, Wednesday worst. Meta puts the biggest share of weekly spend on Wednesday and the smallest on Sunday and Monday. p < 0.00001, holds on both halves of the data and both platforms. The manual fix, and what it is worth.

One US account, 8 campaigns. Four never cleared 50 optimisation events a week, and produced 70 events between them: one campaign's worth of signal spread across four. Those four ran on $16,843, 23% of budget, and returned 54% ROAS at day 28. The other four ran on $58,012 and returned 80%. Holding launch age constant, the split still holds at 75% against 54%.

I ranked the same 38 Meta app campaigns, $606,462 of spend, two ways: by day 0 ROAS and by cost per purchase. Rank correlation with day 28 ROAS came out at +0.96 for day 0 value and -0.81 for CPA. Back the best 13 and the gap is 96.6% against 85.9%.

Real D28 ROAS from one subscription app across four paid UA channels: Meta 1.61x, Apple Search Ads 1.30x, Google Ads 1.09x, TikTok below parity. What each channel is good for, and why the effort spent chasing a new channel usually earns more back on Meta.

A subscription app removed its free trial for paid Meta traffic. Conversion went from 7.1% to 12% and cost per result fell from $58 to $40. The full test, week by week.

An audit of $383K in Meta tROAS spend, country by country. 7% had been failing the exact metric the algorithm optimizes, and no dashboard flagged it. The scoring framework that catches it in thirty minutes a month.

A fitness app with organic traction and positive unit economics, $1,500 a day, 40+ creatives, and a $30 cost per trial that would not move. Why we paused instead of pushing, and the five lessons from the decision.

Google Search for app installs is broken by design: no keyword control, no search term visibility, and iOS attribution that dies at the App Store redirect. The five workarounds and their tradeoffs. Written with Burak Arslan.

Andromeda changed how Meta processes signals and finds audiences. The full playbook: channel order, Tier 1 West testing, IPM benchmarks, the campaign structure in three phases, signal choice, scaling, and subscription cash flow.