How Fragile Is Performance Marketing, Really?

A brand can show high ROAS, controlled CAC and growing sales — and still become weaker every month.

It only sounds contradictory if we equate efficiency with resilience.

ROAS shows how efficiently we bought a result within a specific system. It does not show what happens if the terms of that system change. If media costs rise. If the algorithm changes. If a competitor enters the same auction with a bigger budget. If targeting effectiveness declines. If product discovery moves from search results to the answers of an AI engine.

Performance marketing is not fragile in itself.

What is fragile is a growth model that depends almost entirely on it.

A simple stress test

Run a simple mental stress test:

If the company stopped buying exposure tomorrow, how much of the demand would remain?

Would customers keep searching for the brand by name? Would they come back directly? Would there be CRM, community, organic interest, referrals and repeat purchase? Or would demand vanish almost simultaneously with media spend?

If it is the latter, the business has not built a strong growth mechanism.

It has rented momentum.

Efficiency is not resilience

Performance marketing gave us something extremely valuable: the ability to connect investment to behaviour and behaviour to a commercial outcome.

The problem starts when the ability to measure becomes the sole management criterion.

The dashboard sees clicks, conversions, attributed revenue, cost per acquisition and conversion rates. It does not see as easily:

• how strong brand memory is,
• how dependent demand is on one platform,
• how easily the customer can replace the product,
• how often they buy only because of a discount,
• how much pricing power the business has,
• how much demand would exist without the campaign,
• how vulnerable the company is to a change in distribution terms.

This does not mean performance metrics are wrong.

It means they are local measurements of a larger system.

The problem is not measurement. It is the moment the measurable is confused with the important.

When a business optimises only what can be attributed quickly, it gradually removes everything that works over a longer horizon: brand memory, differentiation, creative risk, organic presence, relationships, content, category authority and customer experience.

In the report, this removal can look like efficiency.

In the system, it is a loss of alternatives.

And the more alternative sources of demand shrink, the more dependence grows.

• Short-term efficiency ↑
• Alternative sources of demand ↓
• Dependence ↑
• Fragility ↑

The most dangerous dependence is the one that looks like success.

Performance marketing does not only capture demand

A widespread view needs an important correction here.

Performance marketing is not condemned to do only demand capture.

Search mainly captures active intent. Paid social, video, creators and strong creative can also create new demand, shift perception and bring a brand into the consideration set.

The real problem is not the channel.

It is the objective function we give it.

If a campaign is judged solely through a short attribution window, the algorithm will do exactly what it is rewarded for: find the people most likely to buy now.

It will not necessarily protect future demand. It will not build strong differentiation on its own. It will not make sure the brand can raise prices. It will not invest in audiences that do not convert today but could become the biggest source of growth tomorrow.

These are strategic decisions.

They are not media optimisations.

Research by Les Binet and Peter Field has highlighted exactly this tension: short-term response activity and long-term brand building produce different effects over different time horizons, while using very short-term online metrics as the main performance criterion can have dangerous consequences for long-term success (source: ipa.co.uk).

This does not mean every company should mechanically apply a 60:40 rule. Their own later work stresses that the balance depends on category, business model, size, position and the brand's life stage (source: ipa.co.uk).

The point is not the percentage.

The point is not to judge two completely different functions by the same time criteria:

• creating future demand,
• and converting today's demand into revenue.

The four debts that don't show up in the performance report

An over-dependent performance model can create four forms of debt. They do not show up directly in the media dashboard, but at some point they get paid.

1. Demand Debt

The business captures more demand than it creates.

At first this seems logical. The bottom funnel performs better, so it gets more budget. High-intent audiences bring lower CAC, so the system concentrates there.

Over time, however, the pool does not refill at the same rate.

The company starts competing for the same people against more competitors. Costs rise. Incremental reach falls. Retargeting runs dry. Frequency climbs, but growth slows.

The dashboard says “market saturation”.

In reality, the business has stopped creating enough new reasons for someone to enter the market or to choose it.

2. Platform Debt

The more growth a channel produces, the more budget moves into it.

It sounds perfectly rational.

Until the channel becomes a single point of failure.

The company does not control the auction, the cost of access, the targeting infrastructure, the interface, the attribution model or algorithm changes. It can improve performance daily without reducing its strategic dependence at all.

A channel's success today can become tomorrow's risk concentration.

3. Meaning Debt

When conversion pressure is constant, the message often drifts towards whatever produces the fastest reaction:

Price. Offer. Urgency. Availability. Discount. Retargeting.

These can increase transactions.

They do not necessarily build a clear reason to choose.

Gradually, the brand trains the market to respond to an incentive, not to prefer the business. The customer buys because the price was right, because the ad appeared at the right moment or because the product was the easiest to reach.

Not because they would ask for it by name.

That is how pricing power is lost. Price elasticity increases. And a competitor willing to offer a bigger discount can shift demand very quickly.

4. Learning Debt

Platforms learn constantly.

The question is whether the business learns too.

An algorithm can find audiences, placements and combinations that produce results without the team truly understanding:

• why the message worked,
• which need it activated,
• which element built trust,
• which demand was incremental,
• which customers would have bought anyway,
• which creative principle can transfer to another channel.

Then the know-how stays inside the platform's black box.

If the platform, agency, account or technology changes, much of the “learning” is lost.

The campaign was optimised.

The organisation was not.

AI is not just another channel. It is a new layer of choice

Until recently, a brand competed mainly for a spot in the feed, in search results or on marketplaces.

Today it increasingly competes for a place inside the answer itself.

AI does not only change the search interface. It introduces a new intermediary between intent and choice: a system that synthesises information, compares alternatives and forms a limited set of recommendations.

Google describes AI Mode as suited to complex questions, comparisons and multi-step research, using “query fan-out” across different sources and, where needed, shopping data for billions of products (source: blog.google).

The strategic consequence is significant:

The brand no longer competes only for the click. It competes to be included in the answer set.

Here, though, caution is needed.

AI does not “understand brand meaning” the way a human does. It can retrieve and synthesise available signals: clear product information, a consistent category, content, structured data, reviews, third-party mentions, product feeds and proof of credibility.

Humans respond to meaning.

The machine needs accessible proof.

That is why the answer is not to turn GEO into yet another list of technical hacks. Google itself states that no special AI markup or new schema is required to appear in AI Overviews and AI Mode. The fundamentals remain: helpful and reliable content, technical accessibility, clear text, correct structured data and up-to-date information in Merchant Center and Business Profile (source: Google for Developers).

What is needed, in other words, is Evidence Architecture:

• clear positioning,
• a consistent product description,
• content that proves expertise,
• product and business data that can be retrieved,
• reviews and real social proof,
• mentions from credible external sources,
• a clear link between the promise and its proof.

Even so, AI visibility is not fully predictable. Early 2026 research on commercial recommendations from ChatGPT, Gemini and AI Overviews found significant variation in both the recommendations and the sources shown across different systems and repeated queries (source: arXiv).

This means AI discovery should not be measured with a single prompt and a screenshot.

It needs repeated tracking of a fixed set of commercial queries, across different engines and over time.

Brand is not decoration. It is a buffer.

In stable periods, brand can look like a slow and vague investment.

In volatile periods, it reveals itself as a protection mechanism.

A strong brand means that:

• more customers search for you by name,
• you need less persuasion in every transaction,
• your creative is recognised faster,
• the conversion path has less friction,
• price is not the only criterion,
• paid media amplifies existing preference instead of starting from zero every time.

Brand does not replace performance.

It makes performance more efficient and less dependent on the platform.

This is also why growth cannot be treated as a single-engine problem. The Brand Engine creates meaning, the Performance Engine turns intent into motion, the Experience Engine proves value and the Leadership Engine keeps the system aligned. A weakness in one engine limits the output of the others.

In Nassim Taleb's terms, a brand is:

• fragile, when an external change can dissolve its demand,
• robust, when it can absorb the change,
• antifragile, when it can use volatility to gain ground.

A truly antifragile growth system does not depend on everything staying the same.

It has brand memory, owned audiences, diverse sources of demand, a strong experience, organisational knowledge and calibration rhythms that let it adapt faster than competitors.

The Performance Fragility Test

Leadership can assess the fragility of its growth model through seven questions:

1. If we reduced paid spend in a controlled way, how much of total demand would remain?
2. What share of new customers depends on our strongest platform?
3. How much revenue comes from branded search, CRM, referrals, returning customers and organic discovery?
4. If we remove the discount, what specific reason remains for the customer to choose us?
5. Do we know the real incrementality of our campaigns, or only the attributed ROAS the platform reports?
6. Can we reach a customer again without paying for access from scratch?
7. When an AI engine compares the options in our category, what verifiable evidence do we have to be included in the answer?

The test is not meant to prove that performance marketing is problematic.

It is meant to reveal whether it has become the system's only lifeline.

From the efficiency dashboard to the resilience dashboard

The answer is not to reduce accountability or return to a vague marketing model where nothing connects to business outcomes.

The answer is to measure two things at once:

Performance efficiency

• Incremental CAC
• Contribution margin
• Conversion rate
• Payback period
• Customer lifetime value
• Funnel velocity
• Incremental revenue

Growth resilience

• Channel concentration
• Branded demand
• CRM and owned-audience contribution
• Referral and repeat-purchase revenue
• Promotional dependency
• Pricing power
• Organic and AI answer-set visibility
• Share of demand that survives when paid pressure drops

One scorecard shows how efficiently the engine runs today.

The second shows whether the business will still have an engine tomorrow.

Five moves for less fragile growth

The answer is not anti-performance.

It is a fuller growth system in which Clarity, Creative, Conversion, Channels and Calibration work as one loop, not as isolated specialisms.

1. Map the demand portfolio

Separate the demand you create from the demand you capture.

Measure how much growth depends on high-intent audiences, retargeting, branded search and promotions. Look for risk concentration, not just media efficiency.

2. Protect a horizon beyond the attribution window

You do not need an arbitrary brand-building percentage applied to every company.

You do need a protected budget for demand creation, creative experimentation, category education and community building that is not cancelled because it did not produce last-click revenue within a few days.

3. Turn attention into an owned relationship

Email, CRM, community, loyalty, thought leadership and the after-sales experience are not secondary activities.

They lower the cost of the next contact.

The business must be able to speak to the customer again without buying back the whole relationship from a platform.

4. Build Evidence Architecture for people and machines

Every important brand promise needs proof that both the customer and search systems can understand:

use cases, methodology, reviews, clear product data, external mentions, real expertise and a consistent category identity.

5. Make calibration organisational learning

Don't just record which ad won.

Record:

• which hypothesis was tested,
• why you believe it worked,
• what you learned about the customer,
• where the knowledge can be transferred,
• which result was truly incremental,
• what needs to change in the strategy.

The platform should optimise the campaign.

The business should optimise the system.

The real question

Performance marketing is an excellent engine.

It should not be the whole vehicle.

A mature business does not choose between brand and performance. It uses brand to create preference, performance to turn intent into revenue, experience to grow LTV and calibration to make every cycle smarter than the last.

It does not reject efficiency.

It refuses to let efficiency eliminate the assets that will make the next phase of growth possible.

That is why the most important question at the next budget review is not:

“How high is our ROAS?”

It is:

“How much of our demand will still be ours when we can no longer buy it on the same terms?”

That is not simply where marketing performance is measured.

It is where the resilience of the business itself is measured.