The Ugly Ads
Doctrine

The reasoning is that outsized returns in paid social come from discontinuities, not from grinding out a few percentage points. One creative, one concept, one angle produces a day where purchases triple. If that has never happened to you, you have not yet found the thing that works — you have found something that merely does not lose money.

This has a direct implication for how you report. Weekly average CPA hides exactly the signal you are hunting for. A week containing one spectacular day and four mediocre days averages out to look identical to a week of five flat days, but those two weeks mean completely opposite things.

The canonical example discussed is a BetterMe illustration of belly fat — visually crude, borderline unpleasant, and so effective that the host still vividly recalled it roughly a decade later. That is the entire argument in one artifact: memorability and attention capture are the metrics, and they are frequently inversely correlated with aesthetic quality.

Her warning even names the tool, and it is worth quoting directly: “Don’t listen to Claude when they tell you which creatives to create and then they give you a fifteen-second creative.” The instinct of any general-purpose AI tool is to produce something that looks professional, which is exactly the thing that gets scrolled past. (Noted with some irony, since this page was prepared with Claude. The point stands: use AI for volume, not for taste calibration.)

Her position is deliberately provocative: unless you are Coca-Cola, your brand does not exist and nobody cares about it except you, your family, and some of your employees. Nike sells prettiness; a subscription app sells function. Brand marketing will never outperform performance marketing on short-term ROI, so the honest sequence is to earn revenue with performance first and buy brand later.

But she also volunteers the counter-example, and it is important. At Solid Starts she assumed brand did not matter and ran a fruit-baby creative concept. The feedback was users asking whether a company that made that ad could be trusted to advise them on feeding their infant. In trust-critical categories, ugly creative can attack the core value proposition itself.

Her worked example is her husband’s aquarium app idea. It is a real passion, there is genuine enthusiast content on social platforms, and the adjacent spend per user is high. Her position is that a small amount of paid testing would have revealed the audience is too small and the acquisition cost too high for the monetization to work — before any development began.

The host pushed back reasonably: niche passion categories with expensive adjacent purchases are often exactly where UGC and affiliate models work. She partially conceded, and the resolution is the useful part. The idea might work as a $20k/month independent business and simultaneously fail as a portfolio app, because the LTV:CAC required to justify studio resourcing is much higher than the bar for a solo operator.

She also names a social failure mode worth flagging: friends and family validate ideas because they want to be supportive, not because the idea is good. She frames it as a cultural contrast, with American feedback culture being more encouraging and Eastern European feedback more bluntly skeptical. The practical lesson is that enthusiasm from people who like you is not data, and should never be logged as evidence.

It is not only my assessment. Asked in the closing lightning round for the most impactful change of her past year, she answered localization — “the way to open markets that you have never took money from” — and said she would bet on it over everything else.

This is the mechanism that makes Doctrine I and Doctrine II financially survivable. High-volume creative testing is unaffordable on US CPMs, which is why most teams test twenty creatives a week instead of hundreds, which is why they rarely find step changes. Moving the testing layer to low-CPM markets — her passing example is Indonesia — breaks that constraint. Winners are then ported into expensive markets with the creative risk already retired.

The critical qualifier is similar conversion behavior. A test market is only valid if creative that wins there also wins in the target market. That correlation must be verified empirically, not assumed — the whole system silently fails if the test market’s audience responds to different mechanisms.

The second play is about revenue, not testing. Her argument: teams default to the US because it has the largest cash value, but ignore the markets they actually understand or that are quietly converting well. Babbel and Blinkist built enormous businesses starting from Germany. Ewa started in its own local language before expanding. With AI translation, adding a language is now a matter of hours rather than a project.

Her Japan bet at Replika is the cautionary case. They committed, hired a Japanese speaker, and did genuine localization — and the cultural adaptation still did not work. Her conclusion is a deliberate paradox: don’t go hard on countries you cannot understand, but also try many more countries than you currently do.

The resolution is a tiering model. Do not treat “add a language” and “enter a market” as the same decision.

She adds a genuinely useful reframe on cultural research: look for similarities, not only differences. Her example was recognizing shared generational experience with a Korean friend and with people from Central America, despite very different surface cultures. For app positioning this means the underlying pain is often far more universal than the surface presentation, and a working angle frequently ports across markets that look unrelated.

Both Replika and Solid Starts had enormous organic traffic. Money arrives without daily ad spend, the business is not hostage to CAC, and it feels like the ideal state. Her argument is that this is a trap with a delayed detonation, and she has watched it play out more than once.

Replika compounds the lesson with an unforecastable shock: ChatGPT commoditized the entire AI-companion category. An organic-dependent business has no dial to turn when its category collapses.

She attaches a story that makes the no-steering point sharper. The early ChatGPT-wrapper apps caught a purely organic influx and solidified top App Store positions — and when other teams copied their paywalls and monetization, nothing transferred, because the money was never coming from the paywall in the first place. In her words it was all just dumb luck: the owners could not repeat it, were not really monetizing it, and could not even sell it. Copying the visible parts of an organic winner teaches you nothing, because the part that mattered is invisible.

She is blunt that organic strategy only applies if the product is genuinely unique. If you are building a time tracker or a storage cleaner, organic will not happen, and neither will brand. Uniqueness is the precondition, not the outcome. The host’s example is Cal AI: it went viral because the mechanic was novel and visually demonstrable in a single screenshot. He also notes the decay — once every calorie app has the same feature, that same creative stops working.

ASO gets flagged as the one organic channel that keeps delivering for years, with the caveat that algorithms appear to favor established apps, making a top position harder to win now than previously. Both speakers treat serious ASO as a specialist discipline requiring dedicated daily keyword work, not something to do casually.

The specific trap she names is seductive and familiar: “we already have the developer on salary, so this new idea costs nothing to build.” Development capacity is not the binding constraint. Attention is. Marginal engineering cost being near zero says nothing about the marginal cost of the leadership focus, creative iteration and market learning that the new product will consume.

Her closing lightning-round answer: growth would be easier in blue markets rather than red ones. She describes re-entering the mental health category and finding it brutal — “it’s mental out there” — and says she does not know how people are surviving without a genuinely unique product.

The host supplies the counterweight: Cal AI entered a category already crowded with calorie counters and won, and he admits he made the same misjudgment about Ladder, assuming a new fitness app was walking into an impossibly brutal market. She answers with her own history. In 2017 everyone assumed there was no room for another fitness app, and companies bigger than BetterMe have been built from that market since. Her conclusion: when a market is big enough, there is room for everyone.

The Higgsfield section describes a team working seventeen-hour days, six-day weeks, sleeping in the office. She is careful not to prescribe it universally — she says she would not want to work that way now — but observes it produced a genuine market lead, and that afterward she pushed harder in subsequent roles.

The host’s reframe is the more useful takeaway and the one worth adopting: intensity works when it comes from genuine passion and is chosen, not when it is imposed. Both speakers agree that cracking the whip on employees to work ten-hour days does not produce the same result. The management implication is to hire for intrinsic motivation in the categories you operate in, and not to attempt to manufacture Higgsfield-style intensity through pressure. It does not transfer.

The final substantive discussion covers deceptive subscription funnels: quiz flows that obscure pricing, post-purchase upsell chains that exploit sunk-cost momentum, per-day pricing that hides the real charge. Her account of how it happened is the important part, because it is not a story about bad actors.

The host’s account is concrete: his wife completed a long quiz funnel, authorized a $1 charge via Apple Pay, and was then walked through a chain of upsells she clicked through while trying to reach the plan she believed she had bought. The real charge was $350–400. Support refused a full refund citing a 30-day term. He notes a chargeback is the appropriate response, and that the pattern was almost certainly extremely profitable while it lasted.