Короткий домен Telegram перестал работать

Telegram лишился домена t.me: он разделегирован и больше не работает на уровне регистратора. Платформа срочно переезжает на telegram.me, а владельцам крупных каналов стоит обновить публичные ссылки. Сроки восстановления неизвестны, и есть риск, что t.me не вернётся вовсе на фоне давления на Telegram.

➡️ Читайте на сайте: https://aff.top/blog/korotkii-domen-telegram-perestal-rabotat

🧠 Ещё больше инсайтов → в канале AFF.top

Youtube тестирует поиск с AI

YouTube начал тестировать Ask YouTube — поиск с ИИ, где можно задавать вопросы обычным языком и получать не список ссылок, а готовую подборку видео и фрагментов.

Фича уже доступна в США и работает на сложные запросы: если нужно, ИИ уточняет вопрос и подсказывает следующий шаг.

Что это значит для поиска на YouTube и когда новинка дойдёт до других…

➡️ Читайте на сайте: https://aff.top/blog/youtube-testiruet-poisk-s-ai

🧠 Ещё больше инсайтов → в канале AFF.top

Is the 'naive' linear attribution model actually the most honest one?

The question: linear attribution splits credit equally across every touch in the path. It's dismissed as simplistic — the model you use when you've given up. But there's a contrarian case that it's the most intellectually honest default. Is it?

The argument for it: every weighted model — U-shaped, time-decay, even data-driven — encodes a claim about which touches mattered more. Linear makes the opposite, deliberately agnostic claim: we don't know, so we won't pretend. In the absence of a causal experiment, any non-uniform weighting is a hypothesis dressed as a result. Linear refuses to invent precision it doesn't have.

What the analysis shows: studies comparing models find linear is rarely the best fit to conversion data — but it's also rarely catastrophically wrong, and crucially it has no systematic directional bias toward openers or closers. Time-decay over-credits the bottom; first-touch over-credits the top; linear sits in the middle by construction. Its errors are diffuse rather than skewed.

The nuance: 'no bias' isn't 'accurate.' Linear is provably wrong whenever some touches genuinely matter more (they usually do). Its honesty is the honesty of a shrug — fine as a baseline, weak as a decision tool.

What to actually do: use linear as your reference point. When a fancier model disagrees with linear, ask whether the difference is evidence-backed or just the model's built-in prior talking.

Bottom line for practitioners: linear attribution is the null hypothesis of crediting. Don't ship it — but make every other model beat it on evidence, not on vibes.

Z.ai анонсировала новую GLM-5.5

Z.ai готовит релиз флагманской GLM-5.5: модель обещают показать в августе 2026 года.

Главная интрига — рост до 1 трлн параметров при том же контекстном окне в 1 млн токенов. Новинка снова будет заточена под код и агентные задачи.

Почему версия сразу 5.5, без 5.3 и 5.4, и что это может означать для рынка — в блоге.

➡️ Читайте на сайте: https://aff.top/blog/z-ai-anonsirovala-novuiu-glm-5-5

🧠 Ещё больше инсайтов → в канале AFF.top

Telegram запустил собственный сервер для ботов

Telegram запустил собственный сервер для ботов и мани-приложений: теперь backend можно размещать прямо внутри инфраструктуры мессенджера.

Сервер работает на JavaScript/TypeScript, через вебхуки, и позволяет подключать SQL-базу для сбора контактов без посредников.

Пока неясны цена и ограничения — что именно уже можно тестировать, а где скрыт подв…

➡️ Читайте на сайте: https://aff.top/blog/telegram-zapustil-sobstvennyi-server-dlia-botov

🧠 Ещё больше инсайтов → в канале AFF.top

Is your incrementality win real, or did you just test during a novelty spike?

The question: you run a clean lift test on a new channel or creative, see strong incremental conversions, and scale. Three months later the lift has evaporated. Was the experiment wrong, or did the world change?

What the literature describes: this is the novelty effect (and its cousin, ad fatigue) — well-documented in both UX and advertising research. Early exposure to something new generates a temporary response that decays as the audience habituates. A short incrementality test run at launch captures the peak, not the steady state. The estimate isn't biased by bad design; it's biased by timing — you measured a transient.

The deeper problem: this isn't correlation-vs-causation. The causal effect was real during the test. The error is assuming a point-in-time causal estimate is stationary — that what lifted conversions in week one will lift them identically in month six. Most attribution and incrementality work implicitly assumes stationarity, and audiences are anything but stationary.

The nuance: the inverse also happens — some effects build over time (brand, retargeting saturation), so a too-early test understates them. The direction depends on the channel.

What to actually do: run lift tests over a duration long enough to see the decay curve flatten, not just clear significance. Re-test scaled channels periodically — incrementality is a depreciating asset, not a constant.

Bottom line for practitioners: a single incrementality number is a snapshot of a moving target. Measure the curve, not the peak, before you bet the budget.

Google картинки станут конкурентом Pinterest

Google Картинки начали превращать в полноценную платформу с персональной лентой по прошлым запросам — по сути, в аналог Pinterest.

Во вкладке For you уже тестируют подборки, а ещё обещают коллекции и генерацию изображений во встроенной Nano Banana.

Как это будет работать и когда новинка дойдёт до других стран — в блоге.

➡️ Читайте на сайте: https://aff.top/blog/google-kartinki-stanut-konkurentom-pinterest

🧠 Ещё больше инсайтов → в канале AFF.top

Are your newest campaigns being unfairly penalized by math?

The question: you compare last week's campaign to last quarter's and the new one looks worse. But conversions take time to land. Could the comparison itself be structurally rigged against anything recent?

What's happening: yes — this is conversion lag bias (sometimes called the delayed-feedback or attribution-maturation problem). Conversions arrive on a distribution: some same-day, many over the following days and weeks. Any campaign you measure 'now' has only collected its early conversions; its long-tail hasn't matured. Compared against a fully-matured older campaign, the recent one looks artificially weak — not because it performed worse, but because it isn't done converting.

What the research formalizes: delayed-feedback models in computational advertising treat each conversion's delay as a random variable and model the survival curve, so a campaign's expected final performance can be estimated before all conversions land. Without that correction, optimization systems that judge campaigns too early will systematically kill long-cycle channels prematurely — the channels whose conversions arrive slowest get cut before they can prove themselves.

The nuance: the bias is worst for considered purchases and B2B, negligible for impulse e-commerce. It also interacts nastily with the attribution window — a short window plus lag bias double-penalizes long cycles.

What to actually do: never compare a fresh campaign's raw conversions to a matured one's. Either wait for maturation or model the lag curve and compare projected finals. Build a 'cohort maturity' check into reporting.

Bottom line for practitioners: recent campaigns aren't underperforming, they're under-aged. Compare like-aged cohorts or you'll execute your slowest-converting channels for the crime of being measured too soon.

Россияне не смогут покупать стейблкоины

Россиянам могут закрыть доступ к покупке стейблкоинов: в новой версии закона их приравняли к иностранным активам.

Купить такие токены смогут только квалифицированные инвесторы — например, с активами от 24 млн рублей или доходом от 12 млн в год.

Что это значит для обычных пользователей и когда правило заработает — в блоге.

➡️ Читайте на сайте: https://aff.top/blog/rossiiane-ne-smogut-pokupat-steiblkoiny

🧠 Ещё больше инсайтов → в канале AFF.top

Can a channel improve in every segment yet look worse overall?

The question: a channel's conversion rate rose in mobile, rose in desktop, rose in tablet — but its blended rate fell. Your reporting says the channel got worse. Every segment says it got better. Who's lying?

What's at work: nobody is. This is Simpson's paradox — when a trend present in subgroups reverses or vanishes after aggregation, because the mix of subgroups shifted. If the channel's traffic moved toward a lower-converting segment (say, more cold prospecting, less warm retargeting), the blended average can drop even as every component rises. The aggregate is a weighted average, and the weights moved.

Why it matters for attribution specifically: attribution reports live and die on aggregated rates — conversions per channel, ROI per source. Any time the underlying composition of a channel's traffic changes (new geos, new placements, new audience mix), the headline number can move in the opposite direction from reality. Acting on the aggregate without checking the mix is how teams cut channels that are genuinely improving.

The nuance: the paradox cuts both ways — a channel can look like it's improving in aggregate purely because its mix shifted toward easy converters, masking a real decline in every segment.

What to actually do: never read a channel's blended rate without a composition check. Decompose into stable segments and look at within-segment trends and the mix shift separately. If they disagree, trust the segments and investigate the mix.

Bottom line for practitioners: aggregated attribution numbers can flip the truth. When the total contradicts every segment, the total is an artifact of the mix — not a finding.