AI and Advertising: the future of creativity and campaign optimisation

of Alberto Narenti

In the new digital advertising landscape, dominated by machine learning, privacy-first and automation, a new role for advertisers is emerging: less executors, more strategists. In this article, we explore how Artificial Intelligence is changing the work of advertisers, between liquid creativity, predictive optimisation and new attribution models such as Marketing Mix Modelling. A journey inspired by Patrick Gilbert’s Join or Die, an essential book for anyone who wants to survive (and succeed) in modern advertising.

What we will cover in this article:

Welcome to the age of algorithms
The automation of automation: the AlphaZero case
From manual CPC to strategic guidance
Liquidity and Micro-Moments: when AI really works
Confidence vs Accuracy: avoiding illusions
Attribution: all models are wrong (but some are useful)
MMM and AI: allies, not alternatives
True human value: creativity, strategy, critical thinking
Join or Die

Welcome to the age of algorithms

Once upon a time, the advertiser’s job was a craft. Campaigns were built piece by piece, CPC was optimised manually, and ad groups were segmented like a chessboard.

Today, however, we have entered a new era: that of “Automation of Automation”.

As Patrick Gilbert explains in Join or Die, the advertiser’s role is no longer that of someone who manually operates levers. The modern advertiser is an algorithm tamer, a strategic coach who works to create the ideal conditions for artificial intelligence to learn, explore and optimise.

“It’s not about who has the best data anymore. It’s about how you use that data.”

The automation of automation: the AlphaZero case

In 2017, AlphaZero became the symbol of machine learning. In just four hours, it learned to play chess from scratch and defeated Stockfish 8, the most powerful chess engine in existence at the time.

How? By playing against itself, without anyone teaching it the rules.

This paradigm is now also applicable to advertising: it is no longer necessary to tell the algorithm what to do, but where and how to learn. Our role becomes that of building the right environment and feeding the model with meaningful data.

From manual CPC to strategic guidance

In the book, Gilbert describes two profiles:

Traditional advertiser
Prefers manual control
Relies on granular structures, rigid targeting and split testing
Measures success based on “vanity” metrics and thinks in silos

Modern advertiser
Trusts the algorithm
Builds fluid, flexible environments
Focuses on broad audiences, dynamic creativity, predictive models

The future is obviously the latter. Effective advertisers are those who create the conditions for the most promising patterns to emerge, leaving room for AI to explore.

But be careful: trusting the algorithm does not mean abandoning strategy. On the contrary, modern advertisers must develop an even deeper understanding of the context in which their campaigns operate. This includes:

Understanding direct and indirect competitors
Analysing the value proposition of their brand in relation to the market
Monitoring economic trends and cultural changes
Reading emerging needs and weak signals in user behaviour

Only those who know the playing field well can guide the algorithm in making effective decisions. Because, like any good coach, the advertiser does not play in place of the machine, but maximises its potential with vision, context and strategic direction.

Liquidity and Micro-Moments: when AI really works

The concept of liquidity is central. It means removing unnecessary constraints and letting budgets flow towards the most valuable impressions.

The four forms of liquidity:

Placement liquidity: this means abandoning the logic of fixed or preferential placements, allowing the platform to independently identify the most effective placements based on real-time performance. This includes display, video, search and cross-device environments.
Audience liquidity: avoid forcing the platform into overly narrow or predefined segments. It is better to provide strong signals (e.g. conversions, qualified leads) and let the algorithm automatically expand similar audiences.
Budget liquidity: instead of fragmenting your budget across multiple campaigns or assets, it is more effective to aggregate it into a few smart containers. This allows the platform to dynamically allocate resources where it has the most margin for return.
Creative liquidity: provide a varied set of assets (images, headlines, videos, formats) so that the platform can independently test, combine and optimise creatives based on context.

This “controlled fluidity” allows you to make the most of micro-moments, those key moments when the user is mentally ready to act: ‘I want to know,’ ‘I want to go,’ ‘I want to do,’ ‘I want to buy.’ But for the platform to truly capture these moments, it is essential that advertisers understand the context in which they operate: who their competitors are, what unique value proposition they can offer, how purchasing habits are evolving, and what economic trends may influence spending behaviour.

Only by combining this awareness with the potential of automation can a truly responsive, fluid and high-performing ecosystem be built.

Confidence vs Accuracy: avoiding illusions

One of the most enlightening passages in the book concerns the difference between:

Confidence: how sure the algorithm is of a prediction.
Accuracy: how correct that prediction really is.

Too many campaigns are optimised on incorrect but very confident data (see: spurious conversions, non-causal post-click events). This dissonance often stems from blind reliance on surface-level signals that do not reflect a real impact on sales or business.

To solve this problem, robust models are needed that can distinguish between correlation and causation. This is where two major families of statistical approaches come into play: frequentist and Bayesian.

The frequentist method is based on the analysis of representative samples and validation through hypothesis testing. It is useful for obtaining reliable results when large volumes of stable data are available over time.

The Bayesian method, on the other hand, continuously updates probabilities based on new evidence. This makes it extremely effective in dynamic environments such as advertising, where data changes rapidly and adaptability is an advantage.

In both cases, the key point is to be willing to reset the model, remove incorrect assumptions and put the algorithm back into learning mode. Not out of weakness, but out of strength: the strength of those who accept that every prediction is a hypothesis to be validated, not an absolute truth. A/B testing on Google and Meta is a true act of algorithmic humility and, at the same time, professional awareness.

In this regard, it is also essential to learn how to critically read the results of A/B tests conducted on platforms such as Google and Meta. Recent academic studies have highlighted how the way these platforms divide users between control and test groups can generate skewed delivery, i.e. a distorted distribution that compromises the reliability of the experiment itself (Ali et al., 2023).

Furthermore, algorithms often tend to optimise too early based on initial signals, leading to an overestimation of the campaign’s effectiveness. This phenomenon, combined with incorrect statistical interpretation, can lead marketers to draw hasty or even misleading conclusions.

For this reason, it is important not to be afraid to retrain the algorithm, especially after tests that show ambiguous or inconsistent results. Machine learning only works if it is continuously fed with valid feedback. This requires a more rigorous culture of experimentation, less influenced by the pressure to find “winning results” at all costs.

Attribution: all models are wrong (but some are useful)

Gilbert makes it clear: perfect attribution does not exist.

Google’s data-driven model has great potential, but it is opaque. We do not know what really matters. We only know that it is based on patterns within your data.

This is why alternative tools such as Marketing Mix Modelling (MMM) are making a comeback. Because they offer:

A holistic view of channels, not just digital ones
The ability to distinguish between causality and correlation
The integration of exogenous factors (seasonality, macro trends, competition)

MMM and AI: allies, not alternatives

MMM is often seen as an old-school method. In reality, it is increasingly based on machine learning models.

As illustrated in the document “The Privacy-First Mobile Measurement Method”, a good MMM model today:

Uses Bayesian and/or frequentist techniques, including linear and non-linear regression
Takes into account adstock and saturation
Is fed with aggregated, privacy-compliant data

In a world where user tracking data is disappearing (see: ATT, Privacy Sandbox), MMM becomes an alternative compass for continuing to optimise advertising spend with reliable models.

True human value: creativity, strategy, critical thinking

No artificial intelligence, however advanced, can replace the human ability to read between the lines, connect the dots and build strategic visions in uncertain contexts.

In the chaos of KPIs, formats and platforms, what really sets an effective advertiser apart is the ability to clearly define the campaign’s objective: to generate demand? To consolidate brand equity? To encourage trial? It is a choice that requires a deep understanding of the business context, the stages of the customer journey and the available attention signals (e.g. Share of Search, SoV, etc.).

Algorithms can help scale a message, but they cannot create meaning. Only humans can define a relevant, distinctive value proposition that is consistent with the positioning and expectations of an ever-changing market. As evidence from the IPA SoS ThinkTank also shows, brands with an eSoS (excess Share of Search) higher than their market share tend to grow faster.

A good advertiser does not just look at their own data, but interprets weak signals in consumer behaviour, competitor movements, and cultural or technological paradigm shifts. Knowing how to anticipate change is essential for effectively guiding AI. Continuous analysis of the context and market, including through indicators such as Share of Voice, Share of Search and media attention, allows for the development of more intelligent strategies.

Access to data is not a competitive advantage. It only becomes one when you are able to truly read it, going beyond appearances. It means knowing how to identify hidden patterns, distinguishing between what works superficially and what generates value in depth, linking numbers to psychological, cultural and economic dynamics. Insight does not come from raw data, but from the combination of analytical skills and strategic intuition. This is what guides decisions in the long term.

That is why no artificial intelligence will ever be able to completely replace human thought. Only a conscious mind can lucidly choose the most suitable goal for a given context, formulate a value proposition that is truly relevant to the public, read and interpret a constantly evolving market, and generate insights capable of driving innovation.

As Gilbert reminds us, our work is not finished. It is only changing. Human value is shifting upwards: we leave the execution to automation, but we remain the true architects of strategy.

Join or Die

The message is as simple as it is powerful: either you evolve, or you fall behind.

But it’s not just a technical issue. It’s a cultural revolution: a change of mindset.

Let go of obsessive control. Accept uncertainty. Develop new skills. Build ecosystems, not campaigns. Train AI with human intelligence.

Only then can we truly combine creativity and performance, becoming the advertisers of the future.

What SAY can do

At SAY, we believe that AI is a lever, not a substitute.

Here’s how we help our clients:

Liquid and smart campaign architectures, without bottlenecks
AI-friendly setup: dynamic creativity, shared budgets, broad audiences
Incremental testing to validate algorithm learning
Marketing mix modelling to read the big picture and optimise ROI
Critical reading dashboards to decode system signals and gain a holistic, unbiased view

Want to take your advertising strategy into a new era? Contact SAY S.P.A. for a personalised consultation: we can help you build an intelligent, creative, fluid media ecosystem.

AI or no AI: what will you choose?

Join

Die