LLMs vs SLMs: Why AI Doesn’t Need to Be Expensive to Be Effective

By Mateusz Rumiński, VP of Product, PrimeAudience

The advertising industry has always evolved rapidly; however, few could have predicted the scale of Generative AI adoption over the past 18 months. In a remarkably short time, AI has moved from the margins to the centre of media planning, activation, and optimisation.

In fact, our research suggests that 80% of marketers say generative AI now plays an important role in planning and activation, while three-quarters rely on it more than traditional methods. The value is clear: improved targeting accuracy, more effective personalisation and stronger data analysis.

This rapid adoption has also led to a common misconception: many brand decision-makers believe that the most powerful AI must also be the biggest, most complex and most expensive. To date, Large Language Models (LLMs) have dominated the conversation, trained on enormous datasets and supported by vast infrastructure. Without question, these LLMs have demonstrated impressive capabilities in content generation, analysis and automation. Despite this, the assumption that meaningful AI requires sizable budgets is now being challenged.

While many teams understand how to use AI tools, fewer fully understand how decisions are being made beneath the surface. When factors such as cost, transparency, and scalability are taken into account, the current focus on LLMs appears unsustainable.

LLMs and its strengths

With billions of parameters and training data drawn from across the web, LLMs excel at generalised language understanding. For global technology platforms, consultancies and hyperscalers, this breadth is a major advantage. LLMs can be applied across multiple use cases, languages and industries with minimal reconfiguration.

In marketing, LLMs have helped accelerate creative production, streamline insight generation and automate repetitive tasks at scale. They are particularly well-suited to organisations with the infrastructure, data volumes and budgets required to train, fine-tune and continuously update them. For big tech companies, these models become strategic assets that already operate at a massive scale.

However, using LLMs isn’t always the most efficient option, as training and maintenance can be expensive. Even accessing them via APIs can introduce unpredictable costs, latency issues and dependency on external platforms.

More generally, pricing is often hard to determine. With each model, the token price changes, and you can never fully predict how many tokens your query will require. In many cases, you also use additional software on top of the model, which introduces an additional pricing layer, for example, in the form of credits. The use of these credits can also be unpredictable, depending on the specific query.

As a result, for many advertisers, especially agencies and mid-sized brands, the question is no longer what AI can do, but how to apply it most efficiently.

Small is beautiful

This is where Small Language Models (SLMs) come into play. “Smaller” does not mean simplistic. It refers to models with fewer parameters, trained for specific tasks rather than generalised language mastery. Crucially, recent advances mean these leaner models can now deliver performance levels comparable to earlier generations of LLMs, without the same computational, financial or environmental overhead.

For marketers, the implications are significant. SLMs are much easier to retrain or fine-tune, enabling models to better account for the specific requirements or circumstances of a given brand in a given use case.

As a result, SLMs can be deployed closer to where decisions are made, such as within campaign workflows, audience engines, and optimisation layers. Because they are lighter and more focused, they run faster, cost less and are easier to adapt.

This evolution creates new opportunities. Campaign optimisation cycles shorten from weeks to days as models can test, learn and refine without heavy infrastructure. Scaling across markets becomes more straightforward because teams are not constrained by centralised IT costs. Budgets stretch further, allowing parallel strategies and broader experimentation, e.g. increased  A/B testing activity.

Crucially, SLMs lower the barrier to entry. Advanced AI is no longer accessible only to those with the deepest pockets. Agencies, mid-sized brands, and even niche players can now access automation and personalisation capabilities that were previously out of reach.

An untapped opportunity

The future of effective advertising will be defined by relevance, efficiency and accessibility. LLMs will continue to play an important role, particularly for broad, exploratory or content-heavy applications. For many practical marketing use cases, however, smaller, task-focused models may provide an alternative.

The key takeaway is that AI does not need to be expensive to be transformative, allowing for processes that were previously out of reach to become a reality. Using it effectively can drive high-quality reach without wasting impressions, making it a central part of the modern digital marketing landscape, regardless of the tool used.