Generative AI for Marketing
How to really use it, and what’s next
The hype is over. Now is a time to get real about AI and marketing. That means assessing use cases, cutting through sales hype, and admitting that AI is also starting to disappear back into software. We need to crack on with using these powerful tools for what they are - without a cloud of misunderstandings and wishful thinking pulling us along. In this article, Edmund Hardy outlines a practical view of generative AI for digital.
Start with what Generative AI can’t do
The first, most useful rubric for marketers to focus on is what AI tools cannot do. Otherwise, we won’t get good results or we’ll give up when reality kicks in.
Generative AI shouldn’t be seen as a database, or as a form of intelligence which can think for itself. It is built to predict the next token, such as the answer to a question or a likely plan - but at a vast, sophisticated scale.
This makes it good at more abstract tasks, but less good at implementing detail where accuracy and specific localisation is needed.
How does it work, then?
Central to the functioning of generative AI models are neural networks, modelled on the structure and functioning of the human brain. The network groups items according to similarity, but it does so in multiple dimensions and at a huge scale.
Neural networks consist of interconnected layers of "neurons" that process and group data and then extract meaningful patterns. These neurons apply numerical weights and biases and connections to other neurons - applying a function to an input and passing it on. They’re called transformer networks - the ‘T’ in ChatGPT. At the end, we get the output - what we hope is the useful ‘response’ to our input question. These networks can be trained to recognise intricate relationships between data points, making them ideal for tasks involving language, image, and even code generation.
ChatGPT Use Cases
The key point is that you should be using ChatGPT to think with, not to think for you. It offers a responsive, iterative, back and forth, not a clever robot on your shoulder who does your work for you.
The first use case is formulae writing and Excel analysis. This will take you beyond watching Excel YouTube videos, allowing you to do basic spreadsheet analysis directly in ChatGPT. Basic tasks? Consider them done. More complex ones? You need to check the workings. It can also do basic report commentary - highlighting a data table’s key points and trends, but again, it needs to be directed rather than left to describe what it sees.
Secondly, basic education. With access to the internet, you can learn about processes faster using ChatGPT to guide you than if you sifted through search results directly yourself. For concrete, limited applications, this is a time saver. For example, finding out quickly the answer to questions such as: what are tracking utms? A direct answer from ChatGPT sidesteps Google’s passive highlighting of answers within bigger reams of text, where the context isn’t always right. Again, try to learn something more complex via ChatGPT and hallucinations and errors will creep in, because the likely answer starts to trump the correct one.
Abstracting useful plans and summaries. A joke that the analyst Benedict Evans makes notwithstanding (that OpenAI’s ChatGPT strategy appears to have been drawn up by ChatGPT itself), Generative AI can indeed be used to assemble information and condense it into lists and plans; and likewise to generate creative copy. The distinction we’ve drawn between using it to think with not think for you is crucial here - use it like a template machine to give you a head start, not a producer of original strategies which are fully complete and ready to share with your colleagues and execute.
For efficient and useful application, we recommend using these steps to work with the limitations of ChatGPT: break the task up into steps; regenerate the answer; always ask follow-up questions; ask it to imitate a competent agent.
From Unbundling to ‘Powered by AI’
ChatGPT can be so open and powerful, the user doesn’t know where to start or what it’s best at. This leaves the field open for particular uses cases to be chipped off, packaged up, and sold back to users as ‘An AI app for meeting notes’ or ‘AI powered code generator’ and so on. We’re in the middle of an acceleration of this unbundling. Some of the tools are great - and others simply charge for something you could do on ChatGPT just as easily.
We’re not going to add to the listicle articles suggesting the ‘must have’ AI tools which will ‘transform your business’ here. You should start with the question: are there processes in my business which could be speeded up by a simple AI tool? Accounting, coding, note taking, product shot backgrounds, and simple data analysis are strong candidates.
Further to this, unbundling isn’t all about apps - larger organisations are also scrambling to develop their own proprietary generative AI tech. The aim is to leverage existing data these companies hold; build a revenue generator for the future; and build a moat which is harder and harder for new competitors to cross.
Beyond unbundling is the use of ‘AI in the background’ to simply enhance products and software, with no need for users to consciously engage with generative AI inputs and outputs directly at all. The prompt disappears, and generative AI becomes a part of what is expected of software. This holds true of marketing - Amazon have launched an AI powered product image enhancer; Google have AI powered videos in Demand Gen, and a new augmented reality tool for beauty brands; and on it goes.
There’s Always The Next Hype Cycle
Somewhere, a team is working on the next development which will push generative AI into a new phase. It could, for example, be a development of transformer networks, or it could be an incredible new use for them. It could be a true multimodal app - generating images, video, text, seamlessly.
Google wants the next phase to be ushered in for its Gemini launch - billed as a step beyond its current Bard competitor to ChatGPT. Meanwhile Google and Amazon have both made huge investments in Anthropic, trying to match Microsoft’s OpenAI partnership and investment.
For now, we need to move beyond disillusionment and onto a phase of practically using these tools to speed up and gain capabilities, embedding them into work flows so that whole teams and companies can really benefit and gain an edge, while it’s there to be gained.
More on AI from the Kinase Blog
