AI is no longer relegated to the IT Department. Over the last few years, AI ownership has begun shifting to marketing and brand teams that are in charge of creative, content, commerce, and analytics. Marketers now have a wide range of AI tools at their disposal. While marketing AI has often been used for performance testing or sentiment analysis, CMOs and their teams have gained access to tools allowing them to generate copy, images, video, translations, summarizations, synthetic media, and more.
This type of AI is called Generative AI. The speed at which these AI tools are coming to market and how we use them at work can be overwhelming. In this article, we’ll explain what we mean by Generative AI, why CMOs need to understand it, and how to get started building your Generative AI tech stack.
What is Generative AI?
Generative AI refers to artificial intelligence that can generate novel content, rather than simply analyzing or acting on existing data. Generative AI can create, manipulate, and synthesize content from images to video, copy, music, and data. You’ve heard of ChatGPT, a text-based Generative AI tool, but it’s just one in a quickly evolving landscape.
Generative AI is an abundance technology. If used correctly, a company can use these technologies to scale what they do and create better products and experiences.
Why CMOs need to know Generative AI
As a CMO, you’ve been using artificial intelligence for years. The 8th Salesforce State of Marketing reported that 62% of marketers use AI tools to capture and unify data. From analyzing customer segments, predicting buying patterns, and generating banner ads, AI is nothing new. Marketing teams have been using machine learning to fine-tune their media mix and accelerate revenue growth.
So, you may be wondering, “If my team is already using AI and testing with Generative AI tools, why do I need an AI tech stack?”
AI analytic tools don’t necessarily fit into the same bucket as Generative AI. Your tech stack needs to expand to use generative and analytics artificial intelligence together to create the best possible content. By combining analytical tools with Generative AI tools, your teams will be able to get the best of both AI disciplines, analysis and production.
How to get started
We’re not quite to the point of a Generative AI marketing department. Right now, if you fed an AI-generated ad into a predictive tool, it would fail due to AI hallucinations and wonky design outputs. That’s why it’s important to develop a Generative AI tech stack that involves not only the tech, but the people, agencies, and IT departments to make Generative AI a working tool for your brand.
We suggest you involve all these groups to assess how, when, and why AI might become a strategic part of your approach. Heinz, Coke, Mint Mobile, and Spotify are all examples of brands that have found ways for AI to complement or influence their creative campaigns.
AI subscriptions: Why there cannot be dozens of them
There are hundreds, if not thousands, of artificial intelligence platforms and solutions available to marketing and digital teams right now. ChatGPT plugins, Microsoft pouring money into Open AI, Google’s Bard, Baidu rushing a demo of their GPT, and VCs investing in every AI startup at $1B valuations — these are all examples of the tumultuous AI landscape. Teams will need a thoughtful approach to how to build their toolkit to avoid being buried under a pile of AI subscriptions.
Marketing and creative teams will require a big-picture consideration of what their AI tech stack is, what tools are in that stack, and how it will integrate into their already existing marketing technologies. When it comes to AI, there are layers of considerations, such as data privacy, customer protections, inclusivity, and authenticity to the brand content outputs.
While artificial intelligence has been around for a long time, regulations, usage, and how it will impact humanity is still up for debate. Still, we believe AI is an “abundance tech,” meaning it will aid our creativity and work. So, we caution signing up for AI subscriptions before analyzing your company’s needs, marketing efforts, and types of content that work for your brands.
The essential AI apps
Instead of giving you a list of specific Generative AI apps, we are going to focus on the fundamental types of apps you should have in your stack.
Public trained AI
Creating a Generative AI tech stack starts with conversations about what publicly-trained AI is and how your organization could leverage it. When we reference publicly trained AI, we are referencing large language models and AI tools like Open AI’s GPT, Anthropic’s Claude, Google’s Bard, or the dozens of lesser known tools that have their own foundational models. Publicly trained AI is open and trained on public data. We don’t suggest using raw, public AI apps for your brand. However, it’s important to understand publicly-trained AI as you develop your tech stack. Take public info and retrain it based on your brand and how your brand engages with the world.
Private Trained AI
Next is privately-trained AI. You will want to train these Language Models and other AI tools on your own data and content, to assure they perform through your brand’s lens. This can happen through prompt engineering of publicly trained platforms, fine tuning these AI models with your data, or a series of other methods like Vector Data Embeddings, etc. Prompt engineering is a lighter lift effort, where you simply give guidance to the AI tools, possibly with some uploaded data. Fine tuning is a much more involved, but controllable process. During this fine tuning phase, the public model adapts to the specific language patterns, terminology, and style of your brand’s dataset, with new specific performance goals trained. Leveraging vector embeddings allows your data to be captured via a more simple database and then reformatted by the LLM/GPT. The goal is to improve the model’s performance on a particular set of tasks to better suit the manner in which your teams will use it.
Tech Bridges
It may require a series of Generative AI tools and subscriptions for your team to achieve your desired production and marketing outcomes. What will you do with a series of tools, some being off-shelf public tools and others fine-tuned and personalized to your brand? You’ll need to ensure that they connect to your already-established technologies, APIs, CRM, CMS, etc. Tech bridges are the connections to your other tools. Depending on which platforms you select, there may be simple connections. In other cases, your teams or partners will need to build middleware to connect the dots.
Confirmation Training
Confirmation training is the process of training your AI to make sure it stays on brand, doesn’t talk about things you don’t want it to, and handles bad actors who may want to trick it. Filter the AI through your systems so that it understands your brand’s values and language.
Every time you design an AI process with a client, make sure to factor for the human checkpoint. The human editor, the human manager, the prompt engineer, and the AI QA director — all of these roles will make sure your generative outputs are quality. They will also be keeping an eye out for these slippery feedback loops.
Output Guidelines
You’ve strategized on use cases, selected your tools, fine tuned the AI models to better understand your brand, built tech bridges to ensure they work together, and performed confirmation training. Before your AI tools can represent your brand, you’ll need to ensure that production outputs fit your guidelines. Similar to how you would brief an agency, or guide your marketing team, output guidelines give direction to your AI tools on exactly how you want content to be produced, templatized, formatted, and outputted as final assets.
Tackle your AI tech stack
Start with the possible AI platforms that meet your needs, likely publicly-trained, and figure out how you want to leverage them. Assess and take action on what type of private training or fine-tuning your teams may need to do to assure the AI is an expert on your brand and data. Discuss what tech bridges you need to build and determine your blind spots, and decide how you will train and moderate the AI to make sure the output guidelines fit.
This is your chance to decide how your organization wants to meet the world.