For most of digital marketing’s history, the industry has been getting better at observing the customer.
A consumer searched, clicked, browsed, opened an email, viewed a product, abandoned a basket, returned through another channel and eventually bought. The journey was never as clean as the dashboards suggested, but enough of it was visible for marketers to build a plausible story about what influenced the sale.
AI is beginning to remove parts of that visibility.
A customer can now describe what they need to an assistant, compare several products, narrow the criteria, ask follow-up questions and form a preference before ever reaching a retailer or brand. By the time the business sees the customer, much of the persuasion may already have happened somewhere else.
The customer journey has not disappeared.
Our view of it has.
For marketing leaders, that is more consequential than another change in media format. Much of the modern marketing machine, from attribution to optimisation to budget allocation, was built on the assumption that meaningful customer intent leaves observable digital traces. When more of discovery happens inside interfaces the brand does not control, that assumption starts to weaken.
Attribution was always imperfect. AI makes the imperfection harder to ignore.
Digital marketing created an extraordinary amount of measurable behaviour, and with it came the temptation to equate what could be measured with what created value.
A click was visible. A search query was visible. An email open, product-page visit or paid-media conversion could be captured and assigned somewhere in the journey. That made digital channels appear more accountable than many traditional ones, but it also encouraged an uncomfortable habit: giving disproportionate credit to the activity closest to the transaction.
Anyone who has managed a serious marketing budget knows the limitations of that logic. Customers do not wake up as empty vessels waiting for the final ad before purchase. Brand familiarity, previous experience, word of mouth, product reputation, store exposure, cultural relevance and earlier consideration all contribute, even when the attribution system gives most of the credit to a search click several minutes before checkout.
AI discovery makes this tension much more obvious because the invisible part of the journey becomes larger.
Imagine a customer asking an AI assistant for a work bag suitable for frequent travel. They discuss size, durability, laptop capacity, style and price. The assistant compares several options, explains differences, eliminates two brands and recommends another. The customer then searches the recommended product by name and purchases it.
A conventional dashboard may celebrate branded search.
Commercially, that is an incomplete interpretation of what happened.
The search did not create the preference. It harvested it.
That distinction matters because businesses allocate money based on these interpretations. If we continually reward the channel that captures demand while underestimating the activity that creates it, budgets eventually become very efficient at funding the bottom of a shrinking funnel.
Product data is becoming part of marketing
There is another important change happening underneath AI discovery.
When customers increasingly ask machines to help them decide, the quality of the information machines can understand becomes part of the brand’s ability to compete.
Historically, product data was often treated as operational infrastructure. Titles, attributes, specifications, availability, taxonomy and descriptive content were necessary to power the website, search engine and feeds. Marketing sat somewhere else, focused on campaigns, creative, audiences and media.
That separation becomes harder to defend.
An AI assistant trying to recommend the right running shoe, hotel, skincare product or television needs structured, accurate and useful information. If one brand provides detailed product attributes, credible reviews, transparent policies, strong third-party references and clear differentiation while another offers vague marketing language and inconsistent data, the first is easier for both people and machines to understand.
That creates a new connection between merchandising, product content, data governance and marketing effectiveness.
The customer may never see the product-information architecture directly. They will experience its consequence when the assistant understands the product correctly, recommends it for the right use case and explains why it fits.
For retailers, this can feel uncomfortable because product data has historically been distributed across teams that do not necessarily think of themselves as demand-generating functions. Yet if AI becomes an important discovery interface, those teams begin influencing acquisition whether the organisation chart acknowledges it or not.
The implication is broader than “optimise content for AI.” That phrase risks sending companies down the same path they followed with search, trying to reverse-engineer every new interface.
The more durable strategy is to make the underlying proposition unusually easy to understand.
Good products, clear differentiation, reliable information and strong customer evidence travel well across interfaces.
Marketing will need to rely less on platform-reported truth
As visibility fragments, measurement needs to become more independent.
For years, marketers have lived with an awkward reality: many of the platforms selling advertising also provide the measurement demonstrating that the advertising worked. This does not make the data useless. It does mean sophisticated businesses need another layer of evidence.
That becomes more important when discovery spreads across AI assistants, social platforms, creators, marketplaces, retail media networks and environments where attribution may be partial or unavailable.
The answer is not to find one new universal attribution model. I am increasingly sceptical that such a model exists.
The better direction is to combine several imperfect methods that answer different questions.
Attribution can still help understand observable journeys. Marketing-mix modelling can identify broader relationships between investment and outcomes. Controlled experimentation can test incrementality. Customer cohorts can show whether acquired users create durable economic value. Geographic or audience holdouts can reveal whether activity generated demand or merely claimed credit for it.
None is perfect alone.
Together, they force the organisation to think beyond the conversion report.
This is a meaningful operating-model change because it alters the conversation between marketing, finance and commercial leadership. Instead of asking only which campaign produced the highest reported return, the business starts asking whether total customer demand changed, whether acquisition improved, whether repeat behaviour strengthened and whether incremental gross profit justified the spend.
Those are harder questions.
They are also much closer to the P&L.
Brand may become more important precisely when it becomes harder to attribute
One of the ironies of AI-mediated discovery is that brand could become more economically important while becoming less directly measurable.
If consumers increasingly delegate part of discovery to an assistant, strong brands carry useful signals into that interaction. Familiarity, reputation, customer reviews, authority and cultural relevance help the consumer and the machine reduce uncertainty.
The customer may ask for “the best option” rather than naming a brand. But the information available about brands, and the confidence attached to them, will still influence what appears in the consideration set.
This makes long-term brand investment difficult to evaluate using only transactional attribution.
A customer might encounter a company repeatedly over several years, understand roughly what it stands for and trust its products. An AI assistant later recommends that brand in response to a specific need. The final transaction may then appear as direct traffic.
Very little of the system that made the purchase likely will receive credit.
Commercial leaders therefore need to resist a dangerous conclusion: that anything difficult to attribute must be economically weak.
The opposite mistake is also possible. Brand should not become an excuse for avoiding accountability. The objective is not to measure less; it is to measure differently.
Strong brand investment should eventually reveal itself through outcomes such as higher direct demand, lower dependence on paid acquisition, stronger conversion, greater repeat, pricing resilience and improved customer economics.
Those signals take longer to emerge than a campaign dashboard.
That does not make them less real.
The next measurement model has to start with incrementality
When attribution becomes less complete, incrementality becomes more valuable.
The central commercial question is simple: what happened because we invested this money that would not have happened otherwise?
That question is much harder than asking where the sale was recorded.
A customer who searches a brand name after already deciding what to buy may have purchased regardless of the ad they clicked. Another customer who discovers a product through a creator, asks an AI assistant to compare it and later converts through organic traffic may have been genuinely influenced by activity the attribution system barely recognises.
Incrementality helps separate those two situations.
It also forces companies to confront uncomfortable parts of their media economics.
Some campaigns generate genuinely new customers. Some accelerate purchases. Some shift demand from one channel to another. Some simply place a measurable touchpoint immediately before a transaction that was already going to occur.
This distinction matters most when budgets are under pressure.
Cutting the activity with the weakest attribution may destroy demand creation. Protecting the activity with the strongest attribution may preserve spend that adds very little incremental value.
That is why the measurement function increasingly needs to sit closer to commercial planning rather than being treated as a reporting discipline inside marketing.
The purpose of measurement is not to produce a perfect explanation of the past.
It is to improve the next allocation decision.
AI will make customer economics more useful than channel economics
As discovery becomes harder to observe, the unit of analysis also needs to change.
Marketing organisations have traditionally been structured around channels because channels were where budgets were deployed and performance could be measured. Paid search had its economics. Social had its economics. CRM, affiliates, retail media and display each had separate dashboards, teams and optimisation logic.
The customer does not experience those boundaries.
AI makes that mismatch more visible because part of the journey may happen outside all of them.
A more durable measurement architecture starts with customer economics: the cost to acquire a customer, the margin they generate, how frequently they return, how their behaviour changes over time and whether the relationship becomes more or less dependent on paid intervention.
This does not make channel metrics irrelevant. A search manager still needs to know whether search is performing. A CRM team needs engagement measures. Media buyers still need tactical optimisation signals.
But those metrics should feed a larger commercial view rather than become the final definition of success.
A channel can look efficient while acquiring customers who never return. Another can look expensive while introducing customers who become highly profitable over several years. The first wins the dashboard; the second may win the business.
That distinction becomes more important when attribution gets noisier.
Customer economics gives the organisation something more stable to optimise around.
Measurement will require more judgment, not less
The marketing industry has spent years trying to reduce judgment through increasingly sophisticated measurement.
That ambition was understandable. Better data should improve decisions, and in many areas it has.
But there was always a risk that precision in the dashboard created more confidence than the underlying data deserved.
AI discovery is going to make that harder to sustain.
Leaders will have plenty of numbers. They will simply have fewer reasons to believe that any single system sees the whole customer journey.
That means measurement becomes more plural. Attribution, experimentation, modelling, customer cohorts, commercial outcomes and qualitative understanding all contribute different pieces of evidence.
The executive skill is not choosing one and declaring it truth.
It is interpreting them together.
There is an important difference between accepting uncertainty and abandoning accountability. Businesses still need to know whether marketing creates economic value. They need to become more sophisticated about how that value is inferred.
The temptation will be to solve the new problem the same way we solved the previous one: build another dashboard, create another attribution model and add another layer of technology that promises to restore visibility.
Some of that will help.
But we should probably accept that the customer journey is becoming less observable by design.
People will discover products through interfaces brands do not own, conversations marketers cannot see and recommendation systems whose reasoning will never fit neatly inside a conversion path.
Marketing measurement therefore has to mature from tracking everything to understanding enough.
The journey still exists.
The advantage will belong to organisations that can make good commercial decisions even when they can no longer see all of it.