AIERA Frontiers — the direct continuation of the article “The Closed Circuit”.
1. Behind the interface facade
To see the mechanism, you have to go where outsiders usually do not look: the ad account — the interface in which a business buys impressions from a platform (Meta, Google, TikTok, Yandex).
Every morning the team responsible for paid traffic opens the account and looks at three figures: how much an impression costs, how many clicks were obtained, how much a conversion cost. The day is built around these figures: raise the bid here, turn off the ad set there, refresh the headline by lunch. These figures have an unobtrusive property that organizes the whole working day: they are updated every few minutes, and there is an action for each of them. A metric that updates creates work; a metric that updates once a quarter creates waiting.
From the platform’s point of view all the companies that come to it are built the same way. The platform looks at business blind on the main thing: it is indifferent whether the product is good, whether the brand is honest, and whether the company will make money. It sees three things:
| Element | What it is for the business | What it is for the auction |
|---|---|---|
| Idea | Product, margin, meaning | Blind spot: events are seen, the economics behind them are not — margin, returns, repeat purchases |
| Form | Cover, headline, presentation format | Metric: predictability of response (CTR) |
| Budget | The company’s money | Direct rent: the payment for participation |
The table explains why a conversation about strategy inside the account is impossible: strategy has only the currency that the algorithm sees — and it sees Form and Budget. The Idea participates in the deal only insofar as it has managed to become Form.
In the buyer’s language: the algorithm does not see you. Not your life, not the quality of what you bought, not whether the goods will be useful in a month. Only whether you will click.
The priority of an impression is computed by a simple formula:
Impression priority ∝ Budget (bid) × Interest metric (predicted response)
A simplification. Real auctions also take into account the quality of the ad, the behaviour of similar audiences and the conversion history; for the argument this does not change.
The economic mechanics of such auctions was described long ago [1]; what matters for what follows is not the formulas but the consequences.
First consequence: a template is cheaper than the original. A worn-out Form has accumulated click statistics — the algorithm sees a predictable response and gives a discount on the impression. An original has no statistics: for the algorithm this is a risk, compensated by a higher Budget. Imagine two teams with the same product and the same budget: one spends a month on filming and text, the other reassembles the framework from five winning ads. The first pays more for the impression — not because it is worse, but because it is unpredictable. Ranking turned differentiation into a surcharge, and similarity into a discount.
Second consequence: companies do not delay drawing conclusions. It is faster and cheaper to adapt the Form than the Idea — and thousands of teams simultaneously reduce their headlines, covers and edits to what the auction considers predictable. So the language of brands is brought to a single denominator — an effect that Kyle Chayka called Filterworld [2], and observers of the media industry the “age of average” [3].
Who pays for the auction
There is an ultimate payer for the rising cost of contact, and it is not the advertiser. The cost of acquiring a customer is an item of the cost structure just like materials and logistics: the company bakes it into the price of the product. When the auction makes contact more expensive, the shelf gets more expensive in its wake — the buyer pays for both the goods and the fight for his own attention.
The second expense line is the impulse itself. A templated ad sells not the product but the resolve: a promise with a number and a deadline with the buzzworks, the goods are bought, and a month later it turns out the knives are the same. Returns and discounts to retain the disappointed buyer are also expenses of the auction, and they too are spread into the price for everyone.
The account closes: brands overpay for contact, buyers overpay for goods, and both overpayments feed the same party — the owner of the auction. “Make us overpay” in the subtitle is not a figure of speech, it is the design of the market.
The only participant in the chain that bears none of the listed expenses is the platform itself: the costs of returns and retention are created by its own auction, and someone pays them, but not it. Its income arises at the moment of selling the impression and does not depend on whether the promise in the headline came true. This does not imply that advertising should be stopped: in a category where competitors buy contact, silence hands the shelf over to them. The problem is not in participating in the auction, but in who determines the result of that participation.
The platform does not need your success. It needs the myth of it
From the design of this auction it follows something deeper than is usually thought. The platform is structurally not interested in performance advertisers becoming profitable.
There is a temptation to say: “platforms are interested in advertisers not recouping.” That is wrong. And here is why.
In 2017–2018 P&G and Unilever publicly raised the question of brand safety before the largest platforms: P&G pulled its ads from YouTube, Unilever threatened to leave Facebook and Google. This was a real blow to revenue. In response Meta introduced inventory filters, third-party verification, and spent engineering resources [4, 5]. So large advertisers are not indifferent to the platform.
But here is what matters: the platform does not need your success. It needs the myth of your success, packaged into a case study six months old to attract the next participants of the funnel. These are different things. The first is structural. The second is image.
The difference is visible if you compare two business models.
Shopify earns a percentage of the merchant’s sales volume. If the merchant grows — Shopify’s revenue grows. That is why its tools are tuned to the seller’s success: analytics, logistics, customer retention. The client’s success is the platform’s income. A direct link.
An ad platform earns on the intensity of the bid: the number of advertisers multiplied by the volume of their bids. Not on the advertisers’ profit. Not on a percentage of their margin. If ten companies pay a thousand euros a month and none recoups — the platform receives the same ten thousand. If one company pays ten thousand and earns — the platform receives the same ten thousand.
From this arithmetic follows an unpleasant conclusion. A sustainable, profitable performance business of the advertiser is the ceiling of his budget. He no longer needs to grow the bids. He saturates the niche, exits into organic, and lowers his dependence on the auction. And the continuous flow of new players who come with burning money, burn out and are replaced by the next — that is the basis of the revenue.
Caveat: a burned-out advertiser is also not to the platform’s advantage — the cost of acquiring a replacement has to be paid again, so the system gravitates to an equilibrium in the middle: recoupability slightly above zero — enough that you do not leave, not enough to saturate the niche.
Scope. Everything said works in the performance segment and in SMB — the larger part of the auction. Mature brands with brand objectives live by other rules: they buy frequency and mind share, not the funnel result, — for them the auction is a media channel. Hybrid models (TikTok Shop, Instagram Checkout), where the platform shares with the seller both the risk and the success, are still small — and compete on the same terms as purely auction ones.
This is not a conspiracy. This is the inversion of the subscription model. In SaaS a successful client pays more. In the auction a successful client leaves.
The platform does not optimize against your revenue. But it does not optimize for it either. It optimizes for its own metric: engagement, probability of a click, the intensity of the auction. And this creates the situation in which the platform can simultaneously be counted as successful for you — because recoupability is green — and be unprofitable for you — because the client’s returning behaviour is broken.
The platform profits from your unit economics remaining structurally fragile. Not because it wants your failure. But because the fragile client keeps buying. The stable one stops.
2. The infection of the product
The real takeover begins when the logic of the auction penetrates inside the product itself.
In practice this is expressed in the formula that governs all major ad auctions:
eCPM = bid × pCTR × pCVR
pCTR — the predicted probability of a click, pCVR — the predicted probability of a conversion. For the further reading what matters is only one thing: the more accurately the algorithm predicts a specific user’s response, the lower the effective cost of the impression. Confidence — a discount. Uncertainty — a surcharge. So everything the algorithm has already “seen” costs the advertiser less than what it sees for the first time.
From this property a concrete consequence follows. The algorithm does not reward a “bad” audience — it rewards a familiar one: those users whose behaviour is already like the patterns it has seen before. That is exactly why the auction does not care who it brings: it delivers clicks, not customers. In its loss function there is no variable for “the user’s correctness for the product” — there is only the bid, pCTR and pCVR.
The product team receives the result of this optimization at the input. Together with the target audience, a huge mass of random passers-by comes into the product — people the product did not choose and was not built for. This is the algorithm’s “familiar” audience, not the one the product needs.
The mechanism is not limited to applications. A bank that buys traffic brings people who came for the cashback — and a year later the first screen of the online bank sells credit cards, not manages the account. A restaurant that lives on an aggregator cooks to the photo in the establishment’s card — a dish that survives delivery, not a dinner. Every industry knows its own version of the fourth screen of settings.
Then the standard cycle: new users do not stay, retention falls, the product team wants to fix it. Retention is a legitimate metric, and the desire to fix it is natural. But you have to retain the passer-by, not the resident, and the product begins to restructure for the passer-by: complex tools move deeper into the menu, the first screen gets entertainment features and discounts, push notifications teach returning for a reward, not for value.
Here is the answer to the symptoms from the start of the article. The interface that “got worse” actually got more convenient — not for you, but for the random passer-by that the auction brought in.
The cost of the passer-by is not free. A click costs money, and the passer-by returns almost nothing: registers for the discount, takes the bonus, and leaves at the first price increase. A permanent user brings margin for years — but he has no chart on the dashboard, whereas the passer-by does. In the first seven days, inside the attribution window — the short period over which the platform records the sale to itself — they look the same: both stand in the report as “conversion.” The passer-by makes more expensive the things not visible in the account at all: the questions a long-time user answers to himself he brings to support, and the service expenses are spread into the same price.
The leakage of the core is not visible at once. Subscriptions are cancelled at the end of the paid period, the habit of recommending the product does not zero out in a day, so for several quarters the dashboard shows growth: conversions from passers-by outweigh the residents who are leaving. By the moment the total cash box falls, the product is already something else. The reverse restructuring does not run symmetrically: to return the depth means, for a while, to worsen the passer-by’s metrics, and the auction does not give a pause for rehabilitation and compares the current week with the previous one. A product, once restructured for the passer-by, falls into a trap: any attempt to return the complexity at first looks like a drop.
Then the loop closes itself:
The auction brings random traffic
│
▼
The product simplifies for passers-by
│
▼
The permanent core gets disappointed and leaves
│
▼
Organic sales fall
│
▼
The company buys more traffic ──► (back to the top of the cycle)The permanent user loses the value of the service and leaves; to cover the leakage the company buys more traffic; the new traffic accelerates the simplification. Marketing becomes the only department of the company that converts its own destruction into its own budget.
The mutation of the language
Alongside the product, the language mutates in synchrony. Ten years ago brands spoke in different languages: the knife maker — about steel and edge, the airline — about punctuality, the local bakery — about the morning oven. Today their ads speak one dialect with a recognizable grammar: a promise with a number, a fear with a deadline, a secret with an entry threshold, a question instead of a statement in the first line.
The mechanics are visible on “before — after” pairs. The knife maker: before — “Steel X50CrMoV15 holds its edge for ten years,” after — “Why your knives go dull (and how to fix it before Friday).” The airline: before — “94.7% of flights on time last year,” after — “Stop overpaying for tickets while the airlines stay quiet about it.” The product in both cases did not change — the framework changed, because only the framework is what the auction sees and only the framework is what it assesses.
Then a chain of six links works: a creative that showed a high probability of a click gets more impressions → its authors and competitors see the result and copy the structure → original forms do not accumulate statistics → their prediction stays uncertain → the cost of their impression is higher → they lose the auction again. Each next campaign descends from the previous winners, as a breed descends from breeding individuals. After several iterations only the forms remain that the algorithm has already “learned.”
The result is not one template but a whole class of advertising structures in which over the last five years all the niches leveled out: edtech, fintech, DTC, food delivery. Put five random ad creatives from one category side by side and remove the logos, and a professional marketer will not tell them apart.
For a brand that spent years building its own intonation, the mutation has a separate cost: the recognizability of the voice. This asset does not show up in any account, and so it is written off unnoticed — like everything that has no counter.
The language mutates not because someone decided so — but because through the ranking only passes that which the ranking rewards.
3. The two-stroke trap of the platforms
Why is this situation profitable to the platforms? Because they earn on it twice.
Stroke 1 — creating the noise. Thousands of teams with identical stimuli raise the bids and bury the feeds in templated content; each next contact with the audience costs more than the previous one. The platform’s income grows with the intensity of the competition and does not depend on whether the advertising recouped for its participants [1].
Stroke 2 — selling the silence. When the noise becomes a problem for the advertisers themselves, the platform sells them protection from it:
- Meta — a three-level inventory filter (Expanded / Moderate / Limited), introduced in 2023 [4]; third-party checking of the feed is carried out by the platform’s partners [5];
- Google — the same three levels in content suitability, renamed in May 2026 to Maximum / Moderate / Limited; accounts without an explicitly selected mode have the default switched to the widest level [6]; plus the announced in November 2025 excluded thematic blocks for YouTube feeds [7];
- TikTok — carried the mechanism to a separate product: Pulse places ads next to the top 4% of trending content and declares 99.9% brand safety confirmed by third-party services [8]; the core of the product, Pulse Core, works through the strictest filter level.
Silence is bought not for the best of reasons. In a feed where a scam site with the same framework can stand next to a serious brand’s ad, the company pays not for impressions — for the distance to the scam. It is hard to refuse the purchase on one’s own: while competitors pay for safe placement, the neighbourhood with the scam remains your problem. Each rational response strengthens the system for everyone — the same logic as in the first stroke.
First the platform creates the noise in which any advertising drowns, and then it sells the business paid access to silence. This is not diversion — it is a product.
The asymmetry of the hints
This same logic works at the level of the interface. In the ad account there is no “Fix your unit economics” button. There is no hint “Your margin does not cover the cost of acquisition.” There is no warning “This channel brings off-target traffic that is destroying retention.” A “Increase the budget by thirty percent” button — there is. A “Decrease” button — there is not.
Single-campaign diagnostics are formulated to optimize it within the given budget, not to put the budget itself in question.
This is not a design accident. The platform optimizes its own revenue, not the advertiser’s profit — so in its loss function the price of error exists in only one direction: when it under-receives its percentage. The loss of an advertiser is not an error for it but a normal rotation of clients. If the algorithm says “increase the budget” and you lose money — that is your risk. If the algorithm is silent about the fact that your economics does not balance — that is also your risk. The platform does not pay a penalty for either of these errors. The whole financial risk is shifted onto the advertiser.
From this property three concrete mechanics follow that are visible to anyone who has worked in the account longer than a month:
Onboarding is aimed at the speed to the first results, not at their sustainability. Attracting a new advertiser is more expensive than retaining an old one, so the whole onboarding is built to bring him to the first conversion as fast as possible — even at the price of sub-optimal settings. The hints “launch faster,” “expand the audience,” “raise the budget at the start” are not malice. This is the pressure of CAC economics — the cost of customer acquisition. After the advertiser has “hooked on” and sees the first results, his churn falls sharply. But the price of that hooking is — often a wrongly built unit economics that no one reexamines afterwards.
Success cases age out in six months. An advertiser that grew by four hundred percent in January is often closed by December. But the case stayed in the platform’s blog, because it works for the inflow of the next. The myth of your LTV — the total profit from the customer over his whole life with the product — matters more than your LTV.
Planning tools hint “spend more” before “check whether it balances.” This is not malice. This is the consequence that the system has no stimulus to diagnose the problem on the client’s side. The diagnosis “you spend little” increases the platform’s revenue. The diagnosis “you are losing money” — does not.
4. The war of time horizons
The question remains: why does company management agree to this system for years? The asymmetry of the hints answers half of it: the account does not warn that the economics does not balance. The other half is the difference in the speed of feedback.
The ad account updates its figures every few minutes. Repeat purchases, margin and reputation appear over months. The update speed is also a metric, and it chose the winner long before any meeting.
The transmission link: the growth team
The takeover goes not through power and not through a conspiracy. It goes through the speed of the data.
Finance reports to the board once a month. Product — once a quarter. Marketing shows the figures every day. The board of directors believes the department that brings reports most often — and this is why: frequent updating creates the illusion of control over reality. That which changes every hour is felt as a fact. That which changes once a quarter is felt as a model. So all else equal, the one wins whose figures update faster.
And it is passed not only the budget but also the right to define what is even counted as a result.
At the board meeting this looks routine. The financial director brings a table of twelve columns — one per month; they discuss it for twenty minutes. The growth team brings a chart that updated this morning; they discuss it first, because it is “fresh.” Freshness is also a currency, and its holder gets the microphone.
The right to define the result is rarely formalized by order. It is written into the details: in the template of the weekly report for the board, in the order of the questions at the meeting, in which tab opens first on the general director’s tablet. The concrete link is the growth team and its head. None of them took the power — the schedule brought it. But when the person who speaks with the board most often defines the result as the conversion in the attribution window, that definition becomes the definition of the whole company.
This is the takeover in its pure form: the loss function is no longer written inside the company. What is to be counted a loss is decided by the system that optimizes its own revenue — it updates faster than anyone who could dispute that definition.
The growth team is not the villain. It is the channel. Optimizing the fast metrics, it transmits the logic of the auction into the company and gradually restructures the product to it. The takeover happens not through an order from above but through the rhythm of reporting from below. Nothing described requires a conspiracy: every link is rational in isolation, so the sum of rational decisions cannot be reversed by firing the “guilty ones” — only by changing the schedule.
In January 2026 this was shown plainly. On 12 January the platform Meta removed from the reporting API the long view attribution windows [9]: the campaigns stayed the same, the budgets the same, the sales did not change — and the reported conversions dropped 15–40% overnight [9]. No one learned to sell less — the window through which sales were looked at changed. The dashboard that was considered a measure of the business turned out to be a measure of the auction.
The difference between companies appeared not in the products but in the presence of a second page. Where, besides the account, there was cohort reporting, it was possible to reconcile the cash box with the CRM and see: the sales did not change. Where the account was the only source of truth, the source changed the formula — and the company had no way to check this. January did not change a single business; it showed in whom the result is counted inside the company and in whom inside the platform.
| Metric | Horizon of appearance | Where it is seen | What is happening in reality |
|---|---|---|---|
| Click, CTR | Minutes | Account | Displaces everything else by speed |
| Conversion, ROAS | 1–7 days | Attribution window | Sees the sale, not the returns |
| Margin, cash box | Quarter | Accounting, CRM | Ignored while the account asks for budget |
| LTV, reputation | Years | Off the dashboard | Destroyed for the fast metrics |
The default attribution window was long seven days by click and a day by view [9]; before the forced enabling of Apple ATT it was 28 days and had already shrunk fourfold as far back as 2021 [10]. Long metrics cannot simply be “added to the dashboard”: a screen that updates once a minute has no visual place for a variable that updates once a year. Not because management is stupid — but because the screen does not hold that which does not update at its rhythm. A metric that has become the goal stops measuring that for which it was set — Goodhart’s law has been known for half a century [11]; the new thing is that it for the first time has a counter that updates every minute.
The first thing that comes to mind is to look at the account less often. That does not work for the same reason that the takeover happened: while competitors optimize daily, a pause turns into the loss of position in the continuous bid. Refusal of frequency — again a loss by speed; a win is possible only by the ground, that is in changing who and what counts as the result.
The auction sees a week. Physics counts a year.
5. How to take back control
Marketing itself is not evil — it is a tool with two modes.
Marketing-as-exoskeleton transmits outward the ready value of a strong product: it amplifies what is already decided. Marketing-as-prosthesis reverses the direction: the company begins to restructure the product and the meaning for what is cheap to scale by the algorithm.
You can check the mode without consultants and formulas:
The test of subjectivity. Is the company ready to switch off an ad channel that is profitable in the dashboard, if the physical accounting shows it brings off-target audience and destroys retention?
If yes — the account is still a tool, and marketing remains an exoskeleton. If no — the company’s goal has already coincided with what is convenient to scale for a foreign algorithm.
Who already passes this test
In 2018 Uber stopped Meta advertising on passenger acquisition in the US and Canada for three months — not a cutback, but a pause to see the incrementality of the channel: what happens to the trips when the auction is switched off. Nothing measurable happened. About 35 million dollars of annual spend turned out to be unnecessary [12].
In 2020 Airbnb cut its performance budget by 541 million dollars — the total marketing budget fell by more than half, while traffic held at 95% of the 2019 level. Brian Chesky called the shift permanent: the previous share of performance marketing does not come back [12].
Neither company left the auction: Uber tested the channel by a pause, Airbnb — by a budget cut, and both reconciled the result with the cash box, not with the auction’s report. Such checks are called incrementality tests; for mature teams they enter the quarterly rhythm on par with cohort reporting.
Such checks have a personal version too. Would you have done that action — the click, the purchase, the install — if there had been no popping-up banner? Refusal of the automatic click is your own incrementality test.
The cohort cut
Cohort reporting is simpler to build than it sounds. A cohort is a group of customers acquired in one period and through one channel. Instead of the question “what is our ROAS this month?” a different one appears: what happened to the people acquired in January, after 30, 90 and 180 days — and how do they differ from the February cohort from another channel.
In the attribution window the passer-by and the permanent customer stand in the report in one line — “conversion.” After three months the cohorts diverge:
- the Meta-prospecting cohort often shows a high first ROAS — and an LTV that falls after it: the auction brings those who are easy to promise to;
- the cohort of brand search and organic is more expensive at the entrance — and holds the margin for years;
- the cohort brought by the discount lives until the first price increase.
Without the cohort cut the company optimizes what is visible in the account and destroys what the account does not see. With it, decisions appear that the account counts as an error. The working form — an empty template for your data:
| Channel of acquisition | CAC | LTV 30 days | LTV 90 days | LTV 180 days | Repeat purchases | Cohort margin |
|---|---|---|---|---|---|---|
| Meta Prospecting | ||||||
| Google Non-brand | ||||||
| Organic / Direct | ||||||
| Referral |
CAC in such a table is computed from the advertising invoice and the size of the cohort in the CRM — not from the account’s attribution.
The same cut works on personal purchases too. An impulsive cohort — the pleasure on the day of the click, by the ninetieth day the thing dusts in the wardrobe. A conscious one — the difficult choice at the entrance and the benefit over distance. Look at purchases six months out, not through the seven-day window, and half of the impulsive purchases fall away on their own.
Taking back control does not require refusing advertising. It requires the regular checking of reality: once a quarter to collect a reporting page on which there is not a single metric from the ad account. The structure of such a page looks like this:
| Indicator | Where it is taken from | Who signs it |
|---|---|---|
| Cohort margin | CRM + accounting | Financial director |
| Repeat purchases over 90 days | CRM | Product manager |
| Core churn | Client base | Product manager |
| Cash gap | Treasury | Financial director |
If such a page is collected and decisions are made on it, not on the per-minute charts, the company keeps its product, its customer and its strategy. The page must have a signatory — a person who is responsible for every number on it. This is not a bureaucratic detail, it is the centre of the construction: the page works exactly as long as someone by name is responsible for it.
The first meeting on such a page is noticeably quieter than the customary one. There are no charts that can be scrolled and no positions that can be switched off by Friday; the one who signed the numbers reports, and he answers the questions. The freshness of the report for the first time does not decide whose word is the main one.
The second regular check concerns the very mechanism of the takeover. Once a quarter it is worth reexamining the list of delegations: which decisions the account makes itself — the threshold for switching off the channel, the bid limit, the criterion of a “successful” campaign, the right to replace the target audience with similar ones. While the list is short and signed, marketing remains an exoskeleton; when the list has grown unnoticed, there is nothing left to check — the auction has long since made the decisions.
Why there will be no universal recipe
Any instruction for taking back control is instantly packaged into a new playbook and sold back. The system has already done this with meditation, minimalism and downsizing: the practices of returning to oneself turned into products. The playbook is dangerous not through an error but through a substitution — it offers to execute a foreign judgment without making your own. The only thing that remains unpackageable is the ability of the company to set its own goals and, on their basis, to choose justified risks — before the account manages to hint. This ability cannot be bought for a percentage of the advertising budget.
Two objections
First: conversions do have to be counted. They do. Optimization is a legitimate part of the work: lower the price of the lead, test the landing. The argument is not about the mathematics, but about who sets the objective function: you can optimize only that which has already been defined as the result. When the department that optimizes begins to simultaneously define what is to be optimized, the tool took a right that the tool did not have.
Second: long-term branding exists and is measured. It exists — and it is precisely it that loses budget first, because its horizon does not coincide with any reporting window. The quarterly page is not a substitute for brand metrics, but the place where they finally get a voice: recognizability, repeat purchases and the cash box are seen there at once, and do not compete for one screen with CTR.
The decision that cannot be automated
Marketing has no strategy — and it should not have one. The strategy is with the company, and it lives not in the dashboard but in the signature: one person named what counts as the result and answers for that name.
The algorithm optimizes any metric that you set for it: bids, texts, impressions, the distribution of the budget. It cannot do two things — to choose which metric will count as the result, and to answer for that choice. Responsibility works only personally, so the automation of management hits not on the technology but on the readiness of a concrete person to sign the definition of the result with his name.
The order can safely be handed to the machine: the machine cannot be made an author. This is the decision that cannot be automated — not to switch off the dashboards and not to forbid the auction, but to return the signature under the definition of the result to the one who is ready to answer for it. Everything else — bids, texts, impressions, the daily reporting on clicks — is the execution of the signed. Execution is automated excellently.
A company that has done this changes almost nothing outwardly: the advertising runs, the account is open, the reports arrive every day. One thing changes — the order of the signatures. This unobtrusive difference is the one because of which some companies, ten years on, still sell what they were once chosen for, and others sell what the auction is able to sell.
The finale
In the first article the runner was returned his own map of the trap: the knowledge of how the race is built did not lead out of it, but it returned subjectivity. Here the map is completed by a criterion.
Marketing has no strategy — but that does not mean a strategy is impossible. It appears where the company itself determines which risks are justified: it builds its own bids from its own goal, not from the KPI of a foreign optimization system. This is not a refusal of the tool. This is the restoration of the sequence: first the goal, then the tool, then the metric. Not the other way round.
Strategy begins where the goal is set before the choice of the tool. When it is the other way round, the tool becomes the strategy.
An application that you installed for one function can again become convenient for you — if someone signs the definition of the result with his name.
Sources
- Varian H. R. Online Ad Auctions // American Economic Review. 2009. Vol. 99, No. 2. P. 430–434.
- Chayka K. Filterworld: How Algorithms Flattened Culture. New York: Doubleday, 2024.
- Murrell A. The Age of Average // alexmurrell.co.uk, 20.03.2023.
- Meta Business Help Center. Use Inventory Filter (Brand Safety and Suitability); Social Media Today, 30.03.2023.
- IAB Europe. Meta Launches Brand Suitability Controls and Third-Party Verification, 04.04.2023.
- Google Ads Help. Google Ads renames inventory types and changes the default to Maximum from May 2026 (support.google.com); PPC News Feed, 20.04.2026 — Expanded → Maximum, Standard → Moderate, Limited unchanged.
- Google Ads Help / PPC Land, 17.11.2025 — announcement of Excluded content themes for YouTube feeds (Demand Gen).
- TikTok Ads Help. About TikTok Pulse suite — top 4% of content; brand safety 99,9% (IAS, DoubleVerify, Zefr); Pulse Core — Limited Tier.
- Meta Business Help Center. About attribution settings; Meta for Developers, 12.01.2026 — removal of the 7-day and 28-day view windows; Dataslayer, 19.01.2026; Conversios, 26.03.2026 — the drop in reported conversions by 15–40%.
- Apple. App Tracking Transparency — forced application from iOS 14.5, April 2021.
- Goodhart C. A. E. Problems of Monetary Management: The UK Experience, 1975; the running formulation — Strathern M., 1997.
- Humans of Martech, interview with Sundar Swaminathan (Uber), 21.01.2025; Financial Express, 27.01.2025 — Uber, 2018: full Meta pause for three months (US and Canada), about $35 million a year, no measurable effect. Airbnb. Q4 2020 Shareholder Letter (SEC, EX-99.1), 25.02.2021; Campaign Asia, 02.03.2021; Adriaan Dekker, 2025 — Airbnb, 2020: performance marketing −$541 million, total marketing −58%, the shift called permanent, traffic held at 95% of the 2019 level.