Seeing Through The Hype: The Difference Between AI And Machine Learning In Marketing | AdExchanger

Wednesday, February 26th, 2025 – 1:00 am

Artificial intelligence is inescapable nowadays.

There’s generative AI to create an ad and AI platforms to manage campaigns. Your refrigerator and maybe even your toothbrush have AI embedded in them. Or at least that’s what it says on the box.

There are many open questions for marketers who want to implement AI-driven ad tech – and for customers considering the new AI-powered Oral-B. What does it even mean when a vendor touts its AI capabilities? Does it matter if they feature AI or not?

Are you a marketer, agency exec or ad tech developer who wants to integrate AI and/or machine learning (ML) solutions for your business? Read on for a handy guide of what to look for (and look out for) from the tech.

The terms

AI and ML are often conflated.

One reason is that “AI” has become a buzzier, grabbier term, and companies with products that do ML often refer to their offerings as AI.

ML is the practice of training custom algorithms that process information and identify patterns at a vast scale. AI is the broader field of creating software that mimics complex intelligence – or, at least, that can devise ideas and iterate on its own, rather than as a result of human coding.

AI products of all sorts are being built atop large language models (LLMs), which include OpenAI’s ChatGPT, Google Gemini or Claude by Anthropic. These models incorporate a huge amount of content – text, images and videos – to recognize human prompts or inputs and return a comprehensible answer.

There is a lot of jargon being thrown around in the nascent AI field: deep learning, neural networks and virtualization are just a few examples. And it can get a little confusing. A generative AI chatbot like ChatGPT, for instance, is also a neural network, which is an ML model that allows software to identify unexpected patterns or connections across a breadth of data.

The complexity of neural networks – and of the connections identified by a computer between so much data – is why it is often difficult to satisfactorily explain how an ML product works. In other words, it’s unclear in some cases exactly why a generative AI chatbot like ChatGPT returned a certain response.

For a practical example, consider an AI-based ad product like Performance Max.

PMax doesn’t just retarget audiences or create logical lookalikes; it uses a neural network that can identify someone as a likely target to convert even if there are no obvious data signals that a human ad buyer would consider targetable.

But you can’t ask PMax how it identified a potential target. Even Google developers working on the product sometimes don’t know why it makes the decisions it does.

AI or ML?

The difference between AI and ML, therefore, can be boiled down to how much each requires human involvement in their respective process.

For instance, once an AI-based system is off and running, it’s not tinkered with or controlled by an individual, whereas an ML-based system is more customizable.

It’s likely that an AI- or ML-based vendor is itself built on another AI company, like OpenAI, for its AI processing. A vendor with human services and a customizable model is probably doing ML, whereas a product that comes prebuilt and is accessible by API is the truer AI product.

But AI and ML vendors are both very different from vendors that only use algorithmic modeling and familiar data sets to optimize campaigns – even though these latter companies often frame themselves as AI solutions for marketing purposes.

You might be wondering whether marketers evaluating AI or ML technology need to care about this level of detail.

It probably isn’t worth scrapping over some of these distinctions. AI and ML may be different, but both are a major step up from preexisting algorithmic products that could never surface that same level of insight.

If a vendor’s product is based on JavaScript, and the customer success people on the account are unfamiliar with how to use Snowflake or cloud-based clean room products where ML models are applied for advertising, that’s a red flag, one major CPG marketing analytics leader told AdExchanger.

Marketers should also press for transparency into which large language models (ChatGPT, Anthropic, Gemini, etc.) are being licensed by a vendor.

Unfamiliar intelligences

Aside from navigating the nuances between AI and ML terminology, there are also other important things ad tech vendors, agencies or marketers must bear in mind when dealing with the new technology.

Ad tech is a field where “everybody is analytical, but not everybody is technical,” Kamakshi Sivaramakrishnan told AdExchanger in 2023 when she sold her data clean room startup Samooha to Snowflake.

The transition to ML- and AI-based tech often means learning new software languages, such as SQL and Google’s BigQuery. A big part of Samooha’s value to Snowflake was its ML features that can easily convert ad tech developer code inputs to SQL-based query outputs.

There is also a different pace for AI and ML tasks. Usually, cutting-edge tech is the quickest. But AI and ML tasks can take a while to complete.

During a developer Q&A session at Amazon’s unBoxed conference in Austin last October, ad tech engineers in the room bemoaned the long lag sessions their analytics teams would experience when they used the Amazon Ads ML product built on AWS, called Amazon Marketing Cloud.

Google’s cloud-based clean room product, called Ads Data Hub, has the same problem, they added.

Except, the lag isn’t a problem, Amazon technical product managers told the developers; it’s actually the solution.

These AI and ML models span vast amounts of data and may independently run simulations as part of providing a response.

AI products are built to be more like human intelligence, the Amazon product managers said. “They take a second to think about it.”

And marketers return the favor, by carefully evaluating the options before them with AI and ML tech.