Alex Rybakov

The performance team looks at spend, cost per acquisition, return on ad spend and whether an ad can scale; the creative team looks at the angle, hook, offer, format and execution.

• Both read the same result differently: one ad wins and gets scaled, another loses and gets switched off, while the harder question goes unanswered — what did the test teach that should change the next creative decision?
• They press on attribution directly: was the outcome driven by the angle, the offer or the format, or was the test too noisy to draw a conclusion?
• Connecting performance results back to the creative hypothesis remains work in progress at DTC Performance Lab, with the aim of getting better at finding winners over time, not just identifying them once.

An ad that wins and gets scaled sets a budget decision but leaves the lesson unstated. The same post asks what the team should do differently in the next brief as a result.

Rybakov frames the feedback loop at DTC Performance Lab as work in progress, not a finished method: a direction of travel, not a working system.

Read original

In one business, roughly 100 product ideas went in and about one launched.

• Each idea got a landing page, creatives and around $200 of test traffic.
• A near miss got about a month and a small budget to fix; a bad miss prompted three diagnostic questions rather than more ads: did they misread the price customers expected, fail to show the main benefit, or simply choose the wrong product?
• The process replaced gut feel with three limits: no more than a month, no more than roughly $500, and a target conversion rate at every stage.
• Rybakov notes that the right conversion rate depended on category, traffic source and season, and that acquisition cost at around 25–30% of revenue left little room for error at the top of the funnel.
• They ask: how do you decide when to stop testing an idea?

Mitrichenko's example: a 9% target against a 2% actual, followed by three questions about price expectation, benefit communication and product choice rather than another round of ads.

Read original

They start with an Alex Hormozi line they keep returning to, "match the format that's rewarded them before," and extend it beyond content formats to familiar mental models: what premium looks like, what scientific credibility looks like, what a trustworthy founder looks like, what a product demo looks like, what a transformation story looks like.

• The practical move is to put new value in a familiar container: use a format people already know how to process, then place the novelty in the product, mechanism, insight, proof or point of view.
• They reframe the question from "how do we make this unlike anything the customer has ever seen" to "what should feel instantly familiar so the truly new part gets understood."

A product demo. The format is familiar and thousands exist, but if the product does something genuinely surprising, the familiar frame makes the novelty easier to see.

Read original

Roughly a decade later, Davie Fogarty launched The Oodie around the same fundamental desire — sitting on a couch and being ridiculously comfortable — but with a different machine around it: an oversized hood, sherpa lining, a premium price, DTC distribution, social media, licensing and collaborations.

• They cite The Oodie's own claim of more than 9 million Originals sold.
• Rybakov distinguishes copying the output from copying the logic underneath, brings in Oded Shenkar's framing of imitation as a search for cause and effect, and uses Michael Porter to describe the ceiling: an activity can be copied without the system that made it work.
• Their practical signals are a competitor running the same ad for six months, or five successful companies converging on the same offer.
• The danger is convergence: everyone sees the same winners, copies the same visible layer and ends up looking the same.
• Rybakov proposes the sequence copy, observe, understand, modify, invent, and shifts the question from what to copy to why it worked in the first place — separating what was essential from what was contextual and what was an accident attached to a winner.

Snuggie's infomercial and ~$20 offer turned an odd product into a pop-culture phenomenon; The Oodie kept the underlying desire and rebuilt the price, distribution, product and partnership system around it. Rybakov cites the company's figure of more than 9 million Originals sold.

Rybakov states the change directly: "I used to think 'just copy what works' was lazy business advice.

• I'm starting to think I was wrong."
• They add that originality may be overrated at the start and underrated once you understand what you are changing and why.

Read original

They cite three reported examples to argue that AI makes strong people absurdly leveraged — a shift they are betting on for consumer brands and building an AI-native creative business around.

• The sharpest version is the ecommerce argument: generating far more creative with the same team does not make ad number 1,001 the hard part.
• The hard part is knowing which ten deserve spend, what is being tested, what was learned and what to make next.
• Making things is becoming abundant; knowing what deserves to be made is becoming more valuable.

Medvi, described as two people and $401 million in sales built in about two months with $20,000 and AI tools, with the numbers attributed to a New York Times review; Base44, a side project acquired by Wix for roughly $80 million about six months later; and a reported figure that around one in four companies in Y Combinator's W25 batch had codebases that were 95% AI-generated. Rybakov frames the last figure as strong engineers multiplying output rather than beginners.

Read original

Start here for the testing argument the rest of the work builds on.

Rybakov opens with a split inside the same result: the performance team looks at spend, cost per acquisition, return on ad spend, and whether an ad can scale, while the creative team looks at the angle, hook, offer, format, and execution.

• One ad wins and gets scaled; another loses and gets turned off.
• The question left unanswered is what the test actually taught the team that should change the next creative decision.
• They press on attribution directly, asking whether the outcome came from the angle, the offer, the format, or a test too noisy to support a conclusion.
• The post closes with a self-description rather than a result: connecting performance data back to the creative hypothesis is work in progress at DTC Performance Lab, with the aim of making the system better at finding winners over time, not just identifying them once.

For brands planning a campaign

This is the clearest statement of the problem a creative-testing workshop would tackle: agreeing what to record when a test ends so the result changes the next campaign plan. A practical exercise would take one recent win and one recent loss and state the hypothesis and conclusion in a form the next campaign plan can use.

Useful if you want a concrete stopping rule rather than a framework name.

A debrief after a 45-minute conversation with Konstantin Mitrichenko, described as a founder, CEO, and CFO across ecommerce, marketplaces, and consumer products.

• Rybakov reports the numbers as Konstantin's description of one business: roughly 100 product ideas going in, about one launched, each idea getting a landing page, creatives, and around $200 of test traffic; a near miss getting a month and a small budget to fix; and one example where a 9% target came in at 2%.
• Rather than making more ads, the team returned to three questions: whether the expected price was misread, whether the main benefit was shown, or whether the product was simply wrong.
• They trace how the call used to be made—one person's gut feeling that a tweak would work—and the three limits that replaced it: no more than a month, no more than roughly $500, and a target conversion at every stage.
• They note that there was no single good conversion rate, since it depended on category, traffic source, and season, and that acquisition cost at roughly 25-30% of revenue left little room to be wrong at the top of the funnel.
• The post ends with Rybakov's open question: how to decide when to stop testing an idea.
• The figures are examples from one business, not benchmarks.

For brands planning a campaign

A workshop can use this as a case for writing the stop rule before spending starts, including the budget ceiling and conversion target at each stage. The explicit note that these are one company's numbers, not benchmarks, is also a useful habit to copy when a client wants to reuse another brand's thresholds.

It reads as a counterweight to the testing posts: judgment about what to make, applied at the level of format.

Rybakov argues that most DTC brands try to be different in too many places at once, listing a new category, language, visual world, creative format, and product story.

• To customers, they can feel like work rather than differentiation.
• They build the argument from a quoted Alex Hormozi line about matching the format that rewarded the audience before, then widen it beyond content formats to the mental models people already hold: what premium looks like, what scientific credibility looks like, what a trustworthy founder looks like, what a product demo looks like, and what a transformation story looks like.
• The practical idea is expressed as familiar container, new value, with novelty placed in the product, mechanism, insight, proof, or point of view.
• The demo example shows why: the format is not innovative—thousands exist—but a product that genuinely surprises is easier to understand in a format the viewer already knows how to process.
• They reframe the differentiation question from how to be unlike anything the customer has seen to what should feel instantly familiar so the new part gets understood.

For brands planning a campaign

For a brand deciding how far to depart from category convention, this offers a way to spend novelty deliberately: keep the recognisable format and put the difference in the mechanism or proof. It also sets up a testable question: is an unfamiliar format doing real work or adding comprehension cost?

Read it to understand the bet behind the testing argument: if output is abundant, the scarce input is judgment about what deserves spend.

Rybakov stacks three reported examples.

• Medvi, described as two people and $401 million in sales, was built in about two months with $20,000 and AI tools; the numbers are attributed to a New York Times review.
• Base44 was a side project acquired by Wix for roughly $80 million about six months later.
• And around one in four companies in Y Combinator's W25 batch reportedly had codebases that were 95% AI-generated, framed as strong engineers multiplying their output rather than beginners.
• They conclude that AI makes expertise leveraged rather than removing it, and say they are building an AI-native creative business on that basis.
• The sharpest part is the ecommerce argument: generating far more creative with the same team does not make ad number 1,001 the hard part.
• The hard part is knowing which ten deserve spend, what is actually being tested, what was learned, and what to make next.
• They land on a distinction between making things, which is becoming abundant, and knowing what deserves to be made, which is becoming more valuable.

For brands planning a campaign

Lead with this argument if a client is asking whether more AI output solves a creative problem. It reframes the workshop around selection and learning rather than volume, and asks what process decides which few assets get budget.

Start here for one fully worked example rather than an argument in the abstract.

Rybakov uses the wearable blanket as a worked case: the idea predates Snuggie; Snuggie arrived in 2008 with a memorable infomercial and an offer around $20, turning an odd product into a pop-culture phenomenon.

• A decade later, Davie Fogarty launched The Oodie around the same underlying desire to sit on a couch and be ridiculously comfortable.
• What changed was the machine around it: an oversized hood, sherpa lining, a premium price, DTC distribution, social media, licensing, and collaborations.
• Rybakov cites the Oodie's own figure of more than 9 million Originals sold.
• They distinguish copying the output from copying the logic underneath, bring in Oded Shenkar's framing of imitation as a search for cause and effect, and use Michael Porter to describe the ceiling: an activity can be copied without the system that made it work.
• The useful warning is convergence: when everyone watches the same winners and copies the same visible layer, they end up looking the same.
• They propose a sequence—copy, observe, understand, modify, invent—and reframe the question from what should we copy to why did this work in the first place, separating what is essential from what is contextual and what is an accident attached to a winner.

For brands planning a campaign

Gives a structured way to mine a competitor's winning creative without cloning it, ending with named changes rather than a lookalike. In a workshop, use this exercise to take a competitor's best-performing asset and write down what is essential, what is contextual, and what to change.

In one post, they split the problem in two: the performance team sees spend, CPA, ROAS and whether an ad can scale, while the creative team sees the angle, hook, offer, format and execution.

• Both look at the same result from different angles, so one ad wins and gets scaled, another loses and gets switched off, and the harder question goes unanswered: what did this test teach that should change the next creative decision?
• They list the possible drivers — the angle, the offer, the format — and add a case they think is usually ignored: the test was too noisy to conclude much at all.
• They describe this as work in progress at DTC Performance Lab, connecting performance results back to the creative hypothesis so each test gives the next one more context.

Frames testing as a hypothesis-to-result feedback system rather than a method for picking winners.

For brands planning a campaign

For a workshop on creative testing frameworks, start with a shared way to record what each test was meant to prove and what it actually changed. A DTC agency could bring recent tests and score how clearly each result traces back to a stated hypothesis.

Examples: Source 1

In one of those businesses, roughly 100 product ideas went in and about one launched.

• Each idea got a landing page, creatives and around $200 of test traffic.
• If an idea came close to its target, the team got about a month and a small budget to fix it; if it missed badly — his example is a 9% target against a 2% actual — the response was not more ads but three diagnostic questions: did we misread the price customers expected, did we fail to show the main benefit, or is this simply the wrong product?
• The decision rules that replaced gut feel were three limits: no more than a month, no more than roughly $500, and a target conversion rate at every stage.
• They note there was no single good conversion rate, because it depended on category, traffic source and season, and that with acquisition cost at around 25–30% of revenue there was little room to be wrong at the top of the funnel.

Interview-based case detail, with the operator's numbers and decision rules quoted directly and the closing question put to readers: how do you decide when to stop testing an idea?

Rybakov says the testing figures are examples from Konstantin's description of one business, not ecommerce benchmarks. Ask for evidence from your own category before using them as thresholds.

For brands planning a campaign

These three limits — a time box, a spend cap and a conversion target per stage — are ready-made workshop material. An agency could ask each team to write equivalent kill criteria for its own funnel and compare where the definitions disagree.

Examples: Source 1

It reads as differentiation but can feel like work to the customer.

• Their reference point is an Alex Hormozi line they keep returning to — match the format that has been rewarded before — and they extend it beyond content formats.
• Customers already have mental models for what premium looks like, what scientific credibility looks like, what a trustworthy founder looks like, what a product demo looks like and what a transformation story looks like; they already know how to process those patterns.
• The practical move they propose is a familiar container carrying new value: use a format people understand, then put the novelty in the product, the mechanism, the insight, the proof or the point of view.
• A product demo is their example — the format itself is not innovative, but if the product does something genuinely surprising, the familiar frame makes the novelty easier to see rather than less original.
• They reframe the differentiation question from 'how do we make this unlike anything the customer has ever seen' to 'what should feel instantly familiar so the truly new part gets understood'.

Starts with a named outside source, then extends the idea into a rule and tests it against common creative formats.

For brands planning a campaign

A workshop can start with a clean separation exercise: label each element of current creative as container or value. Then ask which elements are novelty spent where the customer is not looking.

Examples: Source 1

They point to Medvi — two people, $401 million in sales, built in about two months with $20k and AI tools, with The New York Times reviewing the numbers — and to Base44, a side project acquired by Wix for roughly $80M about six months later.

• From YC they cite a figure that around 1 in 4 companies in W25 reportedly had codebases that were 95% AI-generated, and stress that these were strong engineers using AI to multiply output, not beginners.
• Their reading is that AI is not removing expertise but making it absurdly leveraged.
• They say they are building an AI-native creative business on the same bet for consumer brands, and draw the line they return to most often: making things is getting cheap, knowing what to make is not.
• If a team can generate 1,000 ads, ad 1,001 is not the hard part; the hard part is which 10 deserve spend, what is actually being tested, what was learned and what to make next.

External cases and reported figures support a prediction, which is then translated into the operating questions that follow from it.

The Medvi and Base44 figures and the YC proportion are reported by others, and Rybakov presents the YC number as a report rather than a verified dataset. Ask for source documentation before using them as benchmarks.

For brands planning a campaign

A workshop should answer this volume question before adding production capacity: if creative output is abundant, the work is selection, test design and learning capture. Ask what the team would do differently if it could make ten times as many ads tomorrow.

Examples: Source 1

Snuggie appeared in 2008 with an absurdly memorable infomercial and a roughly $20 offer that turned a strange product into a pop-culture phenomenon; about a decade later, Davie Fogarty launched The Oodie around the same fundamental desire, sitting on a couch and being ridiculously comfortable, but with a different machine around it: a giant hood, sherpa lining, a premium price, DTC distribution, social media, licensing and collaborations.

• They distinguish copying the output from copying the logic underneath it, and keep the two ways of using imitation separate rather than dismissing either.
• They quote Oded Shenkar on imitation as a search for cause and effect, and Michael Porter on the ceiling of copying a single activity without the system that made it work.
• Their practical signals: a competitor running the same ad for six months, or five successful companies converging on the same offer, are worth attention.
• They end with a sequence — copy, observe, understand, modify, invent — and the argument that originality is overrated at the start and underrated once you understand what you are changing and why.

Two named outside authors frame both sides of the argument, with a company case as the running example.

The Oodie sales figure is the company's own claim, quoted in the post. Ask for independently audited figures before treating it as a benchmark.

For brands planning a campaign

For a DTC agency, this is a copy-with-a-reason workshop frame: take a competitor ad that keeps running, then separate what is essential, what is contextual and what may just have been attached to the winner by accident.

Examples: Source 1

The Mitrichenko post is a worked example: background, numbers, decision rules and a closing question to readers, with the figures explicitly labelled as one operator's examples rather than benchmarks. Several posts end with a direct question to the audience, and they tie the writing to what they are building and testing rather than treating it as commentary.

Self-revision in public, external citations, interview-based case detail and questions that invite the audience to compare their own practice.

For brands planning a campaign

A collaborative workshop should run on cases and stated rules, not opinions alone. Bring the agency's own test data and framework drafts for the same treatment Rybakov's interview subjects get.

Examples: Source 1 · Source 2 · Source 3 · Source 4 · Source 5

The post does not state whether this is ownership, employment, or a client engagement, so the relationship is left as stated rather than classified further.

View example

Shakoor's post on a winning ad becoming the account's biggest risk makes the same case Rybakov makes from the other direction: build a pipeline of tested hooks and angles so the current winner is not the only thing holding the account up, and track what is being tested next.

Shakoor stays close to account-level execution — spend ladders, CPA and ROAS movement, when to start testing the next angle — and writes in a short, direct practitioner-diary style that also covers career reflection. Rybakov steps back to decision rules and frameworks, cites named outside authors, and ends with an open question for readers.

Read original · Original 2

Malik's landing-page audit and client case posts argue that speaking to one specific reader with one specific problem beats a general headline, which is the same instinct behind Rybakov's point that novelty should sit where it changes a buyer's choice.

Malik works at the level of copy and landing pages — headlines, hooks, email sequences, ad copy — and presents client results as before-and-after numbers with a services pitch attached. Rybakov writes about test design and creative strategy, reports figures as examples from someone else's business rather than their own results, and rarely discusses hands-on copywriting.

Read original · Original 2

Williams describes brands with plenty of capacity to make ads but no fresh angle, and lays out a research-insight-strategy-creative-testing sequence; Rybakov makes the same argument: making ads is getting cheap while knowing which ones deserve spend is not.

Williams packages the work as a named system with staged deliverables and a commercial offer attached, aimed at wellness and supplement brands. Rybakov writes more analytically and reflectively, quoting other operators and authors, and presents their positions as directions they are still working out rather than a productised method.

Read original · Original 2

Williams publishes a four-step method, INSIGHTS™: find the opportunity by mining customer conversations for untapped angle territories; build the creative into hooks, statics, scripts and briefs across personas and awareness levels; define what each concept is testing and read it against spend, return on ad spend, cost per acquisition, click-through, hook rate and hold rate; then kill, iterate or scale and feed the learning into the next cycle.

• The published claim is three to five fresh angles and forty-plus testable concepts a month with no extra headcount, and a 10–20% cost-per-acquisition reduction.
• Customer language and empathy lead the method — credited to direct-response mentor Alex Cattoni — so message mining comes first, not last.
• An agency that wants something it can pick up and run on day one, and is willing to adopt someone else's structure, gets more from this than from Rybakov's open diagnostic.

The framework and its numbers arrive as a pitch rather than a method audit.

• Throughput and cost-per-acquisition figures appear without the accounts, test design or period behind them.
• Before adopting the system, the agency should ask to see those details.
• The post closes with an offer of a free angle-gap audit.
• By contrast, Rybakov labels the numbers they cite as one operator's examples from a single business, not benchmarks, and offers a habit of separating angle, offer, format and noise before drawing conclusions.
• Choosing Williams buys a packaged answer; choosing Rybakov buys a diagnostic the agency builds with its own numbers.
• If owning the framework matters, Williams' system may crowd that out.

Examples: Source 1 · Source 2 · Source 3 · Source 4

Malik publishes audits and outcomes: twenty-three DTC landing pages reviewed with nineteen making the same fault of writing to everyone; a headline rewrite reported at 2.1% to 8.4% conversion on the same product and traffic; and three client cases — an email sequence at 55.6% average open and 8.23% average click-through with 22 calls booked in four weeks, a sales page at 9.7% conversion up from 2.1%, and ad creative at 53% hook rate, 4.33% hold rate and 3.87x return on ad spend.

• The stated common thread is deep audience research, clear positioning, human-sounding copy and metric-driven iteration.
• If the agency's real problem is the message and the page, and it wants cases with reported numbers to argue over, Malik is the closer fit.

The posts are results-forward and promotional: each ends by inviting a direct message for process, availability and pricing.

• The outcomes appear as one-line figures, with no test design, sample size or timeframe in the post text.
• An agency using them as teaching cases would be working from headlines; it should ask for those details before relying on the results.
• Rybakov supplies rules the room can disagree with — about a hundred ideas in and one launched, roughly $200 of traffic per idea, a month and about $500 as ceilings, and three diagnostic questions after a 9% target came in at 2% — and names the counter-positions they are drawing on.
• Malik's focus is further downstream: the page and the message, rather than what a finished test should change about the next creative decision.

Examples: Source 1 · Source 2 · Source 3

Shakoor frames the central risk as building an account around one creative, one hook, one angle and one message, so that when it tires the whole account feels it.

• The answer is to keep asking what is being tested next across different hooks, angles, customer problems and formats.
• The stated aim is a pipeline of winners rather than a single winner.
• The posts also carry five years of continuous work, three of them inside DTC brands and Meta advertising, so the session would be a practitioner describing live account conditions.
• If the agency wants someone to talk through the day-to-day of managing and replacing winners, this is closer than a framework conversation.

The posts are short.

• One is a career reflection; the other states the pipeline principle without the thresholds, stop rules or worked numbers another team would need to apply it.
• To assess whether the method is usable, the agency should ask for those specifics.
• Rybakov's posts include a per-idea test budget, time box, spend ceiling and staged conversion targets.
• They work an argument through to a decision question and name the sources of the ideas they borrow; Shakoor's posts assert the principle and stop, leaving more of the session to what the room draws out.

Examples: Source 1 · Source 2 · Source 3 · Source 4