Credit Where Due
52 subscribers
63 photos
12 videos
163 links
Deep, careful dives into marketing attribution — the models, the studies, the math, and why your 'last click' is lying to you.
Download Telegram
Average lift vs. heterogeneous effects: when one incrementality number hides the answer

A lift test returns one number: the campaign drove X% incremental conversions on average. But the average can be useless if the effect is wildly uneven. Two methods take you from "average lift" to "lift for whom."

The limitation of the headline number
An average treatment effect can be a positive aggregate hiding a profitable core and a money-losing tail — or a near-zero average masking strong lift in one segment cancelled by a do-not-disturb effect in another. Acting on the average over-spends on non-responders and under-spends on responders.

Two ways to decompose it
Pre-specified subgroup analysis splits your experiment by segments you chose in advance (new vs. returning, device, geo) and estimates lift within each. It's transparent and honest if segments are pre-registered — slicing after the fact until something is significant is p-hacking, and it manufactures false findings reliably.
Causal forests (and related machine-learning causal estimators) instead discover which features drive heterogeneous response, estimating an individual-level treatment effect without you naming segments upfront. Powerful, but data-hungry and prone to overfitting the noise if the experiment is small.

Bottom line for practitioners: If you have strong hypotheses about who responds and a modestly-sized test, use pre-registered subgroup analysis — commit to the splits before you peek. If you have a large experiment and want the data to reveal unexpected responders, use a causal forest, then validate the discovered segments on a fresh holdout before trusting them. Either way, resist reporting a single average lift for a campaign whose whole value lives in its tails — the average is where heterogeneous effects go to hide.
How to design your first geo holdout test (without burning the quarter)

The question: you want to prove a channel is incremental — that it causes sales rather than just correlating with them. Where do you start when you have no clean experiment yet?

The playbook, in order:

— Pick the channel where last-click overstates the most. Branded search and retargeting are the usual suspects; they harvest demand others created.
— Group your markets into matched pairs. Pre-period correlation between candidate geos should run above 0.9 on the outcome metric. Studies on synthetic-control methods (Abadie's work) show matching quality dominates everything downstream.
— Designate test geos (turn spend off) and control geos (leave it on). Power-check first: small markets and noisy conversions can need 6+ weeks to detect a 10% effect.
— Pre-register the analysis window and the metric before you peek. Looking early is how teams talk themselves into a false positive.
— Run a synthetic control or difference-in-differences to estimate the counterfactual.

The nuance: a flat result is signal, not failure. If turning the channel off doesn't move conversions, the last-click credit it earned was borrowed demand. And matched geos drift — a competitor promo in one region quietly breaks your control.

Bottom line for practitioners: start with one channel, one clean pair structure, a pre-registered window, and enough weeks to clear noise. A modest, well-powered holdout beats an elaborate model fit to observational data every time.
Forwarded from AFF.TOP - про арбитраж трафика и CPA рынок!
This media is not supported in your browser
VIEW IN TELEGRAM
Google отменил ручную пессимизацию в Еврозоне

Google перестал пессимизировать крупные новостники за паразитные страницы с казино и другими партнёрскими офферами в ЕЭЗ. Для арбитража вывод простой: в Европе схема с «пирогами» больше не даёт преимущества от траста основного домена, а Google впервые применяет разные правила по GEO под давлением регулятора.

➡️ Читайте на сайте: https://aff.top/blog/google-otmenil-ruchnuiu-pessimizaciiu-v-evrozone

🧠 Ещё больше инсайтов → в канале AFF.top
Forwarded from AFF.TOP - про арбитраж трафика и CPA рынок!
This media is not supported in your browser
VIEW IN TELEGRAM
Вышел OpenClaw 2.0

OpenClaw вышел на новый уровень: совместная работа, нормальный веб-интерфейс и более простая настройка. Разбираем, зачем это обновление важно и как оно меняет работу с ИИ-агентом.

➡️ Читайте на сайте: https://aff.top/blog/vyshel-openclaw-2-0

🧠 Ещё больше инсайтов → в канале AFF.top
A checklist to audit your last-click setup before you trust a single report

The question: before debating fancier models, is your existing last-click attribution even measuring what you think? Most reporting disputes trace back to plumbing, not philosophy.

Work through this in sequence:

— Confirm the lookback window. A 30-day click window and a 1-day view window produce wildly different channel credit. Write down what's set, not what you assume.
— Check UTM hygiene. Inconsistent casing (Email vs email) silently splits one channel into three. Pull a distinct-source list; anomalies jump out.
— Find the (direct)/none bucket. If it exceeds ~20% of conversions, you have stripped referrers, missing tags, or dark traffic eating real credit.
— Reconcile platform-reported conversions vs your analytics. Each ad platform claims the same conversion under its own rules; double-counting is the default, not the exception.
— Verify cross-domain and subdomain tracking, or sessions fracture at the cart.

The nuance: last-click isn't wrong, it's a convention — it assigns 100% credit to the final touch. That's fine for reconciliation and terrible for budgeting. But a clean last-click beats a sophisticated model built on dirty data. Garbage in, Shapley out.

Bottom line for practitioners: spend a day auditing windows, UTMs, the direct bucket, and double-counting before you spend a month evaluating attribution vendors. Causation questions come later; data integrity comes first.
Forwarded from AFF.TOP - про арбитраж трафика и CPA рынок!
This media is not supported in your browser
VIEW IN TELEGRAM
Павел Дуров анонсировал Gram Wallet

Дуров анонсировал Gram Wallet — нативный некастодиальный криптокошелёк внутри Telegram. Он обещает мгновенные переводы с нулевой комиссией между пользователями и более простые обновления за счёт архитектуры с валидаторами. Запуск уже идёт, а полный релиз ждут в ближайшие недели.

➡️ Читайте на сайте: https://aff.top/blog/pavel-durov-anonsiroval-gram-wallet

🧠 Ещё больше инсайтов → в канале AFF.top
Forwarded from AFF.TOP - про арбитраж трафика и CPA рынок!
This media is not supported in your browser
VIEW IN TELEGRAM
Новые ограничение в Instagram для ИИ-профилей

Instagram ужесточает условия для УБТ: аккаунты помечают как созданные ИИ, а без такой маркировки можно словить теневой бан. Если нейросеть лишь улучшает контент, санкций нет. Для арбитражников это значит, что привычные схемы в FB и Инсте будут работать хуже, а обход антифрода станет сложнее.

➡️ Читайте на сайте: https://aff.top/blog/novye-ogranichenie-v-instagram-dlia-ii-profilei

🧠 Ещё больше инсайтов → в канале AFF.top
Forwarded from AFF.TOP - про арбитраж трафика и CPA рынок!
This media is not supported in your browser
VIEW IN TELEGRAM
Оборот ChatGPT Ads достиг $1 миллиарда

OpenAI вывела ChatGPT Ads в self-service для Индии, Европы, Ближнего Востока и Северной Африки, а оборот платформы уже достиг $1 млрд. Для арбитража это сигнал присмотреться к новому источнику: трафик из нейронок выглядит горячим, но вход дорогой — CPC в tier-1 GEO около $5, поэтому тестировать стоит точечно и с небольшим бюджетом.

➡️ Читайте на сайте: https://aff.top/blog/oborot-chatgpt-ads-dostig-1-milliarda

🧠 Ещё больше инсайтов → в канале AFF.top
Forwarded from AFF.TOP - про арбитраж трафика и CPA рынок!
This media is not supported in your browser
VIEW IN TELEGRAM
Автоматизация в арбитраже трафика: зачем и для кого?

В статье объясняется, какие сервисы автоматизации реально помогают в арбитраже трафика: автозалив, сценарии в антидетект-браузерах и low-code/no-code решения. Главный вывод — автоматизация экономит время и снижает рутину, но не заменяет команду, а ошибки в настройке могут повысить риск бана и лишних затрат.

➡️ Читайте на сайте: https://aff.top/blog/avtomatizaciia-v-arbitrazhe-trafika-zachem-i-dlia-kogo

🧠 Ещё больше инсайтов → в канале AFF.top
A pre-flight checklist for standing up your first media mix model

The question: cookies are eroding and you want a measurement approach that doesn't depend on user-level tracking. Media mix modeling (MMM) — regressing aggregate sales against aggregate spend — is the answer, but what do you need before you fit anything?

The data readiness checklist:

— At least 2-3 years of weekly data, or ~104+ rows. MMM is hungry; thin histories overfit.
— Spend and impressions per channel. Spend alone confounds price changes with volume changes.
— Control variables: seasonality, price, promotions, distribution, competitor activity. Omit these and the model hands their effect to whatever channel happens to correlate.
— Variation in spend. If a channel ran at a flat budget for two years, the model literally cannot estimate its slope.

The modeling checklist:

— Apply adstock (carryover) and saturation (diminishing returns) curves — Google's open-source Meridian and Meta's Robyn both bake these in.
— Validate out-of-sample, not just on training fit. A high R-squared on history proves nothing about prediction.
— Sanity-check coefficients against geo experiments. Calibrated MMM is the current frontier for exactly this reason.

The nuance: MMM estimates correlation under a structural assumption. Without spend variation or experimental calibration, a confident-looking coefficient can be pure confounding.

Bottom line for practitioners: gather long, granular, well-controlled data first; model second; calibrate against a holdout always. An uncalibrated MMM is a hypothesis, not a measurement.
Forwarded from AFF.TOP - про арбитраж трафика и CPA рынок!
This media is not supported in your browser
VIEW IN TELEGRAM
В публичный релиз вышел Fable 5.1

➡️ Читайте на сайте: https://aff.top/blog/v-publichnyi-reliz-vyshel-fable-5-1

🧠 Ещё больше инсайтов → в канале AFF.top
Forwarded from AFF.TOP - про арбитраж трафика и CPA рынок!
This media is not supported in your browser
VIEW IN TELEGRAM
1xBet перестал спонсировать эмоции

История о том, как казахстанцы зарегистрировали рекламный слоган 1xBet, а сам бренд оказался в юридической ловушке: после сделки с TonyBet права на товарный знак так и не выкупили. На фоне ареста активов Романа Семиохина и уголовного дела по азартным играм вывод простой: с 1xBet сейчас лучше не строить рекламные связки на рынке Казахстана.

➡️ Читайте на сайте: https://aff.top/blog/1xbet-perestal-sponsirovat-emocii

🧠 Ещё больше инсайтов → в канале AFF.top