AI is the fastest-adopted technology ever. The speed and scope of the changes are awe-inspiring.
But as much as I hate to be that guy cracking the whip, I’m going to advocate for faster — or at least different — adoption here.
Why? Well for one, of all the variety of AI technologies at our disposal, most marketers still only use generative AI. When they do, it’s mostly for basic content creation or brainstorming. As far as use-cases go, we’re just scratching the surface of what’s possible.
And I don’t mean “possible at some point in the future”. I mean right now.
I don’t claim to be a completionist. When I go to the gym, I walk straight past the pulley-and-cable contraptions and head for the stair-stepper. But that’s because I’m not training to be an all-around excellent athlete. I just want to be able to summit big mountains on the weekend.
Marketers have access to a world-class “gym” of AI infrastructure today. But most only use one machine: the text-generation treadmill. But unlike me, marketers can’t afford to only train for one thing, and the real gains come when we work the full stack.
Every marketer on earth should explore beyond the LLM and consider what other types of AI are at their disposal, and how they can use them in concert to achieve outsized results.
It’s an exciting world when creative, bidding, measurement, and media planning come together in a neat, AI-powered package.
From One Tool to a Toolbox
Beyond the LLM, you have access to a full toolbox of AI, built from different model types, each with its own purpose. Marketers who learn to use the whole set will gain a massive advantage.
Here’s a tour of the types of AI and tools performance marketers can and should be using today.
1. Generative Language Models & Tools
Something that I’ve found is that while lots of marketers are already using tools like ChatGPT, Gemini, Claude and Perplexity, most people don’t yet know about how to use them well. Over time, I’ve learned to:
- Use prompt chaining: If you sense your LLM is underperforming in response to your tasks, break your task down into steps. Prompt it with the first and proceed with the second only when it’s successful. Make sure your second prompt refers to its successful response to the first, and so on.
- Try different prompt frameworks: There are several popular prompt framework acronyms: RISE, RTF, and CLEAR are probably the most popular. Take your pick and try them out for the same questions. It’ll help you get more familiar with how your LLM of choice “thinks”.
- Use few shot learning: This is a technical term that really just means “give your LLM examples of what you want it to deliver.” They’re great imitators. If you already know what you’d like the outputs of your prompt to look like, provide that to the LLM to improve its performance.
What do LLMs struggle with? Most things aren’t text. Let’s talk about a few other types of AI.
2. Diffusion Models
Diffusion models create images, video, and design assets from text prompts. And they’ve come a long way since their uncanny-valley days.
Similar to LLMs, there are a few different diffusion-based image and video generation AI models. What’s important for marketers to know is that there are now several options on the market that make use of strong models under the hood, but are also built specifically for the task of creating marketing materials.
Choose the option best for you, but know that AI-powered creative generation is possible now, with tools built specifically to help marketers create brand-safe images at scale, account for approval workflows, and make it easy to use the images you create in ads.
3. Reinforcement Learning
Reinforcement learning means that the model tries things and constantly learns from small successes and failures to constantly improve while it works for you.
Marketers are using reinforcement learning AI models to adjust bids, budgets, and creative placements as performance data rolls in.
The most commonly used models are Google’s Performance Max and Meta’s ASC. They’re great, but they’re black box systems by design, so you don’t get all the insights about what the models are doing that could inform your strategy. And obviously it goes without saying that Google’s PMax does nothing to help your Meta ads.
But there are third party tools available, and it’s now possible to achieve AI-powered campaign optimization without sacrificing insight, and with cross-channel line of sight.
4. Probabilistic & Causal Models
This is where strategic value lives. Probabilistic models, like Bayesian MMM and causal attribution engines, help marketers understand why something worked and what to do next.
This is an area that’s getting investment, even though it’s maybe not as flashy as LLMs because it’ll be adopted by far fewer actual users. The available tools are getting faster, cheaper, and more accessible all the time, though. For example, Lifesight now offers flat-rate MMM pricing, and Meta’s Robyn is open-source and community-supported.
This has led us to building models that help you see and act beyond the limits of each ad platform. So what you learn from a test on Google Ads can be automatically tested in Meta right away.
So What Should Marketers Do Now?
- Use the right model
Don’t ask a general-purpose LLM to make high-stakes marketing calls. ChatGPT’s models were trained on the entire internet. That makes them effective and fast for general requests, like copywriting tasks. So, they can polish a brief just fine, but shouldn’t be used to optimize your creative mix or targeting strategies.
Use models that can prove they were trained on data sets relevant to the task they’re intended to help with.
- Expand access to insights and actions
It’s annoying to download a csv and upload it to your LLM each time you want a new analysis. Or to have to re-upload brand guidelines or a creative brief for each prompt.
It’s also unnecessary. There are AI platforms now that — without the need for configuring integrations — can see and understand the data in all your disparate ad platforms, your crm, your ecommerce platform; whatever you like.
You can also create something called an “MCP Server” with Zapier that will allow you to use other LLMs to query your data without navigating through dashboard views or pulling reports manually.
The Takeaway
The future isn’t about more AI. It’s about more of the right AI used in more of the marketing workflow.
The iceberg is real. GenAI content is just the surface. Beneath that: creative scaling, real-time media optimization, and always-on measurement are capabilities offered by a variety of marketing technologies ready to drive impact today.
