AI in Marketing: Let The Robots Do What They Do Best

AI in Marketing: Let The Robots Do What They Do Best

Image of a smiling robot.

Generative AI—we can’t seem to escape it.

It seems like every other day, there’s a “new” tool that claims it will propel my business into the future. A “new” guru who wants me to sign up for their LinkedIn newsletter (thanks, but no thanks). A “new” study that either pontificates this technology’s transformative potential or announces our arrival at the Ninth Gate…

Can we all just relax for a second?

Yes, it’s fair to say that this era has been marked by a profound shift, not just in how we work, but in how we perceive and interact with the world. And yes, I recognize that AI is one of the factors at the core of this shift. 

This is particularly true when it comes to my professional wheelhouse, financial services and fintech marketing.

In a nutshell, here’s what I see:

Marketers are concerned about the possibility of being replaced by the machines. Company executives don’t want to overpay for marketers to use the machines.

With respect to the former, I can nip that concern in the bud rather quickly by reminding our team that CSTMR is the sum of its parts. We’re a family. We depend on creative critical thinking, real-world professional experience, instincts, and sound judgment. AI isn’t there yet.

I like people.

Real people.

And if we ever get to the point that AI can replace people, well, I’m inclined to go down with the ship.

Now, concerning clients, it’s here that the AI in marketing topic gets a little more complicated. Clients care about a lot of things, but in the world of fintech and financial services marketing, they are mainly concerned with performance and ROI. 

In other words, they care about the inputs ($) and the outputs ($).

They are willing to pay a premium for marketing services and expertise that drive business growth. And yet, if ChatGPT can produce the same outputs as a grizzled marketing veteran (it can’t, by the way), why should they invest in high-cost human expertise?

Because even if we let the robots do what they do best (and I think we should), the use cases are limited, the outputs are rather vanilla, and the lack of human creativity and intuition is clear.

So, does CSTMR leverage the power of AI? Yes. 

How? To simplify tasks, streamline workflows, and increase efficiency.


Generative AI, for all the hyperbole and hoopla, is just another productivity tool. It’s a complement, not a replacement.

Large Language Models (LLMs) for Dummies

To better understand how generative AI can best be leveraged in digital marketing, it’s important to also wrap our heads around the technology that powers tools like ChatGPT: Large Language Models (also known as “LLMs”).

Does anyone reading this remember what it was like to go to the library?

For those of you who don’t, here’s a refresher.

Libraries are institutions dedicated to the storage, preservation, and dissemination of knowledge. Back in the day, if you found yourself in need of a quiet place for studying or access to a wide variety of informational resources, your local public library was the answer.

Now, let’s imagine that you’re at a huge library with billions of books. 

You have a question, and there’s a librarian who’s read every single book. You ask your question, and the librarian gives you an answer based on what they’ve read. 

This is somewhat like an LLM.

LLMs are like super-smart computer programs that have read (or been trained on) a vast amount of text from the internet, books, articles, and more. They’re designed to understand and generate human-like text. When you ask a question or request a piece of writing, they respond and make predictions based on the patterns and information they’ve learned from all that text.

The key word here is “predictions.”

Perhaps the best analogy is the predictive text function on our cell phones. If you have a smartphone (which I imagine 99.99% of you do), when you start typing a message, your phone suggests the next word based on what it has learned from you and others. LLMs do this too, but on a much larger scale.

Also, similar to predictive text, LLMs are far from infallible.

Here’s an example.

I’m passionate about staying physically active, and I love to start my day with a morning workout. Often, I share updates on my fitness routine with friends, family, or coworkers—not to brag, but to keep them informed about my whereabouts.  

What I want to send is “Be back soon, just finished up a great RUN.”

What I end up sending is “Be back soon, just finished up a great RUM.”

One message makes total sense at 10 in the morning, the other… not so much. 

Predictive text can’t think or reason like humans. Predictive text, like LLMs, provides an output based on patterns in data, not real-world experience. It’s used for efficiency, often with limited success. 

So, while LLMs can provide information that seems knowledgeable, responses may not always be accurate or reliable, especially on topics that require specialized expertise or real-time information.

What LLMs Do Well

Having said that, large language models, with their advanced capabilities in processing and generating human language, do prove valuable in a variety of domains.

These models demonstrate remarkable proficiency in simplifying complex language tasks and aiding in some creative and technical endeavors.  

Most experts agree that LLMs are effective in the following areas:

  • Language Understanding and Generation: LLMs excel in understanding and generating natural language, making them useful for tasks like summarizing texts, answering questions, and providing QA/revision assistance.
  • Language Translation: LLMs are quite adept at translating languages, even though they may not always reach the accuracy of specialized translation models.
  • Knowledge Retrieval: LLMs can provide information on a wide range of topics, albeit with limitations regarding the recency and specificity of data.
  • Assistance in Education and Research: LLMs can be valuable tools for learning and research, offering explanations, generating examples, and helping with problem-solving.

This list is not exhaustive. 

For example, LLMs seem to do an OK job with certain types of advanced technical assistance, like understanding and generating code. 

While the jury is still out on how good of a “coder” generative AI actually is, it doesn’t matter much to me—it’s a novelty that doesn’t impact our work at CSTMR.

What LLMs Don’t Do Well 

Turning to the limitations of LLMs, there are a number of areas where generative AI falls short (like, really short), especially when compared to the nuanced understanding and creativity of real humans. 

Having an open and honest conversation about these shortcomings is a crucial first step to better harnessing their potential.

Here’s where LLMs prove to be rather ineffective:

  • Providing Real-Time Information and Updates: Most  LLMs lack the ability to access or provide real-time information, as their knowledge is static and limited to the time of their last training. Even with LLMs that leverage popular search engines (e.g., Bing via ChatGPT), the results are hit or miss and often misaligned with the “search intent” of the user’s query.
  • Creative Critical Thought: While LLMs can easily generate content, they lack the originality and emotional depth of real people (remember, outputs are based on patterns in existing data). Also, regardless of how specific and detailed your prompts are, these models cannot provide personal insights or experiences.
  • Understanding Complex Contexts: Play around with generative AI long enough, and you will realize that LLMs sometimes struggle with interpreting and responding to complex situations or questions, leading to responses that may be irrelevant, overly simplistic, or just plain wrong
  • Imitating Emotional Intelligence: Related to the point of creative critical thought, LLMs lack the ability to truly understand or empathize with human emotions. This is a HUGE issue when talking about the use cases for AI in digital marketing (especially copywriting), as the goal of most digital interactions is to stimulate a deep emotional connection. 

This list is also not exhaustive.

Any discussion focused on the limitations and drawbacks of LLMs inevitably leads us to the topics of data privacy and the moral and ethical considerations of generative AI. 

I have some thoughts here too, but this isn’t the time or place to discuss further. If you want to know what I think, reach out via CSTMR’s website or my LinkedIn, and let’s chat. 

3 AI Use Cases in Digital Marketing

If you’ve made it this far, you’re probably saying to yourself: 

“We get it, we get it, the robots are smart but boring… how does CSTMR use them?

Below you will find three real use cases for generative AI at CSTMR. 

No theory. No BS. No fluff. 

1. Buyer Persona Research (With a Caveat)

In the context of buyer persona research and audience segmentation, I think generative AI has a lot of potential. 

Although I’m never going to not chuckle when I see job offers with the title “Prompt Engineer,” I do believe that effective “prompting” (if that’s the appropriate verb) can help guide generative AI to create detailed, hypothetical buyer persona profiles that more or less accurately represent the various segments of a brand’s ideal target market. 

This method leverages the AI’s ability to synthesize and extrapolate information based on its extensive training across diverse datasets.

Here’s a brief summary of how we effectively use prompts to develop basic buyer personas with ChatGPT:

  • Define Demographic Parameters: We start by setting the demographic context. For instance, the prompt might be: “Create a persona for a 40-45-year-old male, living in a suburban area, with an annual income of $80,000, working in a managerial position.”
  • Incorporate Psychographic Elements: Next, we add details about interests, values, and lifestyles relevant to financial services and fintech. For example: “This persona is interested in long-term investment opportunities, values financial security, and is cautious but open to new investment strategies.”
  • Specify Behavioral Traits: Next, we describe behaviors that are relevant to how they might engage with financial services organizations. For instance: “Prefers doing thorough research before making financial decisions, regularly reads financial news, and is active on professional social networks like LinkedIn.”
  • Include Goals and Challenges: Finally, we highlight their financial goals and challenges. For example: “Aims to secure a comfortable retirement, is concerned about college funds for children, and faces challenges in understanding complex investment options.”

After reviewing the initial outputs, we refine the prompts to explore different aspects of the persona or adjust based on any gaps in the first iteration.

OK—now for the caveat.

This process is not a replacement for hard-earned experience and firsthand knowledge of an industry and target audience. Instead, it should be viewed as a complementary process that enhances and refines the insights gained from years of skin in the game.

When we develop a brand or go-to-market strategy for clients, we lean on our in-house strategists.

Why? Because they know the financial services industry inside out, they are experts in market analysis, and they can articulate the pain points of a client’s target audience better than anyone (sometimes even better than the client). 

And that’s where the real value lies. 

Generative AI can provide a basic understanding of buyers but cannot replace the deep, contextual insights that come from boots-on-the-ground experience in a specific industry. 

Again, it can be a complement, but not a replacement.

2. A/B Testing Efficiency

For me, paid media is a blend of art and science, where the artist—part scientist, part storyteller—understands how to leverage data to create campaigns that 1) engage the target audience, and 2) drive brand awareness.

Here, success depends on trial and error through effective A/B testing. 

A/B testing, at its core, involves comparing two versions of a landing page, search listing, ad, or other paid media asset to determine which one performs better in terms of user engagement or conversion metrics. 

While I think the suggestion that generative AI can drive the analysis portion of the testing process is rather dubious (I still prefer a real paid media expert to lead the charge), I do believe that it can significantly streamline the process of “failing forward.”

How? By assisting in the production of derivative ad creatives for the purposes of A/B testing. 

The initial phase of any paid media campaign involves a substantial amount of human input, requiring original thought for the development of the campaign’s core concept, copy, visuals, and overall design. This is where the human element is crucial—marketing professionals leverage their expertise, creativity, and understanding of a brand’s target audience to craft an asset that resonates with that audience.

Once this original ad creative is developed, generative AI becomes a powerful tool for efficiently producing derivative creatives. 

In other words, it makes the process scalable.

For instance, if the original creative features a specific messaging theme, AI can create variations tailored to different audience segments while retaining the overall theme. This could include, for example, rephrasing sentences, incorporating synonyms, or tweaking the creative’s call to action.

The ability to rapidly produce variations means that iterative testing can be conducted much more efficiently. Marketers can easily identify which elements of the ad are performing well and which aren’t, and then use AI to further refine for additional testing.

Last thought (but an important one): 

If you A/B test two mediocre creatives, one is still going to win. Be sure to put in the necessary work to avoid celebrating the lesser of two evils.

3. Blog Post Development (Anything But Copy)

Over the past year, there’s been a noticeable uptick in articles circulating throughout my network that seem overly mechanical… some might even say “robotic.”

Considering that ChatGPT made its debut in 2022, I find the timing to be more than a little suspect.

And I’m not alone.

There is a growing concern among marketing stakeholders about the authenticity and uniqueness of the content populating our feeds. 

This concern also extends to how audiences perceive such content. 

In financial services, consumer trust is paramount. An overreliance on generative AI for content creation will lead to a homogenized, insipid landscape of boring, often valueless ideas that erode the very essence of brand identity that sets one financial services company apart from another.

Here’s a novel idea: let’s avoid the content apocalypse.

How? By distinguishing between using AI as a productivity tool and letting it take over the creative process entirely.

I can’t believe I have to say this, but no, you should not publish AI generated content and pass it off as thought leadership. That would be disingenuous and damaging to your and your company’s brand. 

However, yes, you (and your team) should be leveraging AI in a supporting role to optimize content development. 

AI can be an incredibly efficient tool for tasks such as ideation, research, outline development, and QA/revisions.

Brendan Bresnahan, CSTMR’s Senior Content Strategist, does exactly this. As the lead on all things content for CSTMR’s clients, Brendan researches and produces dozens of articles each month across a wide range of financial services and fintech verticals.

Here is how he leverages generative AI to streamline his production workflow:

  • Ideation and Brainstorming: Coming up with fresh, engaging topics for blog posts can be challenging, especially in verticals that don’t naturally lend themselves to compelling storytelling. Generative AI can aid in the ideation process by suggesting a range of potential topics based on parameters such as industry, target audience, previous articles, and related subjects.
  • Research Assistance: Generative AI can be an invaluable tool for conducting preliminary research to determine the merit of an idea or topic. It can quickly sift through large volumes of information and present summaries, key facts, and insights. 
  • Outline Development: Outlines are essential if you want to produce compelling content. For writers, they serve as a roadmap, guiding them through the development of ideas and ensuring that the content remains focused and coherent. With a little bit of guidance (i.e., you have done your initial research), generative AI can help create detailed outlines, suggesting a logical flow for the article as it highlights key points to cover in each section. 
  • QA/Revisions: While the nuanced task of copy editing is best handled by professionals (we have real humans—Lucas, Whitney, David, Dan, Lau, etc.—who do exactly this at CSTMR), generative AI can provide some initial helpful suggestions for improvements. This can include grammar and spelling checks, suggestions for better word choice, or even recommendations for sentence restructuring to improve readability and flow. 

Notice that I didn’t mention Search Engine Optimization

Why? Because although there are ways that generative AI can help boost SEO performance (e.g., semantically categorizing keywords), there are many marketing tools built specifically for SEO. 

They do it better. Use them.

Now, before we wrap, let’s quickly address the elephant in the room: 

Your writers.

It doesn’t matter how compelling an idea is, if your writers don’t have the chops to bring the idea to fruition, well, yawn. 

It is here that generative AI can serve as an effective litmus test for measuring the quality of your talent pool. Allow me to propose a worthwhile experiment. Take your top performing article. The one that delivers maximum ROI for Organic Search. I’m talking about benchmark-setting traffic and engagement.

Take the title and ask ChatGPT to build out an original blog post. 

If ChatGPT can put together an article that delivers the same value or more (in the form of substance and original creative thought) as your go-to writer, it might be time to reassess your team. 

Generative AI: A Complement, Not a Replacement

OK, there you have it—my in-the-moment, probably incomplete, subject-to-change thoughts on generative AI in digital marketing.

I think AI is here to stay, but as marketers, we must tread carefully. 

Not because the machines are going to take over, but because human creativity is something to be cherished. As we integrate AI into our workplaces, we should see it as a productivity tool that enhances and extends our innate abilities, not one that replaces them. 

At CSTMR, the real magic happens at the intersection of technology + human imagination.

It’s crucial that we maintain a balance where technology amplifies human potential without overshadowing the qualities that make us unique—our intuition, empathy, and ability to think creatively. 

The goal for any marketing agency should be to find harmony. To strike the right balance. To leverage AI as an extension of our hard-fought capabilities and experience.

In other words, as a complement to the work that we do, not a replacement. 

Picture of Rory Holland
Rory Holland
Rory Holland is CEO and Co-Founder of CSTMR. For more than 20 years, he has made it his passion to help Fintech and financial companies leverage digital marketing and advertising to drive growth.

Get Fintech Insights for Growth

Please enter your email address below to receive our thoughtful and timely Fintech marketing insights and information.

Ready to build your brand into something better?

Marketing insights, news, and strategy from the CSTMR Team.

In depth Interviews of the innovators, the disruptors, and the trailblazers from the world of finance and fintech.