Efficacy vs Efficiency: Is Your Large Language Model Putting Your Brand First?

Decades ago, the world’s first mass spectrometer hit the market, capable of providing the precise gram-per-litre composition of any tested material. One of the first materials tested was, of course, the closely-guarded secret formula for Coca-Cola. The spectrometer identified ingredients such as sodium and phosphoric acid, but other ingredients were completely unidentifiable because the mass spectrometer could not figure out what it had not been trained to recognise.

A mass spectrometer, a device that can identify the composition of materials

The story of the mass spectrometer might be arcane, but it contains a very important lesson for the large language models (LLM) that are the engine of generative artificial intelligence (gen AI).

Training Your LLM

To get an optimal gen AI model, the LLM needs to be trained. The more task-specific knowledge the LLM ingests, the better the AI algorithm performs at the generative task. It’s that simple.

The foundation LLM is trained for general tasks, but for knowledge-intensive tasks in particular – tasks that are outside its initial training data – the base LLM needs to be consistently fine-tuned and augmented.

LLMs are architecturally-designed to provide an estimate based on probabilities. So, the ability of the LLM to address a specific marketing requirement, such as the production of animatics or storyboards, is dependent on the quality of the data ingested by the LLM. In the decision-making or problem-solving process, crowds are collectively smarter than individuals. The same is true of LLMs, kind of like your mobile phone predicting your next word.

The quality of the prompt engineering and finetuning of the LLM used to produce the animatics is almost entirely responsible for the test outcome.

Gen AI is remarkably efficient. It works and produces outputs at lightning speed. But the efficacy of the resultant marketing communication is dependent on what specific knowledge the LLM has ingested. A lot of gen AI products are fast and cheap, but are they trained well-enough to provide efficacy specific to a brand’s needs? The answer is no.

Science vs. Science Fiction

In the pursuit of great marketing communications, when it comes to gen AI there is a giant chasm between science and science fiction. The ingestion of knowledge (be it in the prompt or the finetuning) must be science-based. The greatest opportunity for a marketing LLM and gen AI is to ensure that science and not science fiction is building the capability.

To ensure the LLM is fit for purpose, it should – at a minimum – pull in the category’s rational and emotional drivers of choice, the relative importance and performance of those category drivers, past creative test results, qualitative and quantitative insights into consumer behaviour, business and campaign objectives, and targeting and segmentation data. Yes, that much information at a minimum. That volume of information will help lead to a gen AI platform that creates efficacy for brands, as opposed to a surface-level platform built only for efficiency (or a low price).

Gen AI, a technology that is changing the marketing landscape

Gen AI should not be thought of as an invention but rather as a creative inventor. The inevitability of gen AI replacing elements of marketing creativity is undeniable, and it’s already happening. It might start off as a collaboration, but the dramatic efficiency, rapid-fire on-brief creativity and compliance of gen AI will see it eventually roll across the entire advertising industry. What will matter most is how well-trained and tested your LLM is.