Data-driven AI: Your marketing superpower

This article is based on Sharif Karmally’s brilliant talk at the AI for Marketers Summit. As an RMA member, you can enjoy the complete recording here.

It’s been two years since ChatGPT launched, hitting a million users in under a week – a record-breaking debut. Since then, generative AI has become one of the most talked-about innovations in technology. The promise of AI transforming industries, streamlining workflows, and unlocking new opportunities has captured the attention of companies worldwide.

So, how well do you think your company is using generative AI compared to competitors? Are you way ahead, about the same, or lagging behind?

If you’re like me, your answer has probably shifted over the past two years. Sometimes it feels like you’re leading the charge; other times, it feels like you’re struggling to keep up. 

The truth is, as the industry evolves at breakneck speed, staying ahead isn’t easy. What worked a year ago might now only put you in the middle of the pack.

This article dives into strategies for staying competitive in the ever-evolving AI landscape. We’ll explore how to leverage retrieval-augmented generation (RAG), optimize your data assets, and build scalable systems that give you a sustainable edge.

Whether you’re just starting or looking to refine your approach, this guide will help you unlock AI’s full potential and avoid being left behind. Let’s dive in.

The era of temporary AI dominance

In the late 2010s, basketball entered what I call the “era of temporary dominance.” Teams would form “super teams” to eke out small but critical advantages, like being slightly better at two-pointers – closer, lower-risk shots. 

These teams might win a championship or two, but their dominance didn’t last. Why? After winning, the cost of retaining star players often became too high, or those players simply aged out of their prime. Another team would inevitably step up with their own superstars. 

This cycle of temporary dominance mirrors what we’re seeing in AI today. To see what I mean, take a look at Salesforce’s latest State of Marketing report. It highlights how companies are using AI, and the results are telling. 

The high performers – those who are completely satisfied with the overall outcomes of their marketing investments – have more AI systems up and running. 

Meanwhile, the underperformers – those only moderately or less satisfied with the overall outcomes of their marketing investments – are far more likely to still be in the evaluation phase, deciding what to implement.

But here’s the catch: almost everyone is in the game now. Only 3% of respondents said they weren’t using or considering deploying AI. That means the gap between leaders and laggards is razor-thin. It’s just a matter of being slightly ahead – a temporary lead.

McKinsey’s findings back this up. A year ago, just 33% of companies reported using generative AI in at least one department. Today, that number has soared to over 65%. 

In other words, if you were ahead of the curve last year, you might’ve been in the top half of your industry. Now, you might only be ahead of a third of your competitors.

Red oceans, blue oceans, and finding your edge

This all reminds me of the book Blue Ocean Strategy, which talks about the trade-off between value and cost in hyper-competitive environments. 

In a “red ocean,” everyone is fishing in the same waters. There’s so much competition – and so many sharks – that it gets bloody fast. Sound familiar? That’s where AI feels like it’s heading now, with companies rushing to gain an edge.

The goal, however, is to create a “blue ocean” – a space where you achieve both better value and lower costs through innovation. In a blue ocean, you’re not fighting for scraps with everyone else. You’re leading the pack with a sustainable advantage.

The challenge is figuring out how to get there in a landscape where temporary dominance seems to be the rule, not the exception.

Applying the value-cost curve to AI in marketing

Of course, with all exciting new technologies, there’s a trade-off to be made between value and cost. But how does the concept of value-cost balance apply to AI in marketing? Imagine a curve representing various approaches to using AI, ranging from low-cost, basic strategies to high-cost, advanced solutions.

The low-cost, basic approach: Leveraging base models

At the lowest cost, you can simply use a base model like GPT-4, Claude, or a consumer app tailored to your use case. If you’ve got a content flow, you can ask the model to convert a blog post into a LinkedIn post. It’s quick, cheap, and easy. 

The result won’t necessarily be the best LinkedIn post you’ve ever seen, but it’s an acceptable trade-off for the cost.

Adding a touch of prompt engineering

Taking it a step further, you can improve your results with prompt engineering. This involves giving the AI more specific guidance, like your tone of voice, examples of successful posts, and even examples of posts that didn’t perform well. 

The result? Better social posts at a slightly higher cost – but still relatively affordable.

High-cost, high-effort: Fine-tuning your model

At the other end of the spectrum, you can fine-tune your own model. This involves training a bot specifically for your needs, making it almost as effective as a skilled social media marketer. 

While this can yield impressive results, it’s expensive and time-consuming, requiring in-demand talent to set it up and maintain it.

Enter retrieval-augmented generation (RAG)

If all we do is move up and down this curve – choosing between cost and performance – we’ll remain stuck in the “red ocean” of hyper-competition.

However, there’s another option that offers a way out of the red ocean: retrieval-augmented generation (RAG). This technology, while perhaps less familiar, combines cost-effectiveness with high performance.

Instead of fine-tuning a model, RAG involves creating a workflow where a base model retrieves relevant data before generating a response. It’s an additional step in the process, but it’s much cheaper than fine-tuning. Plus, it leverages your data or a specific dataset, making the generated output far more relevant and effective.

The research shows that when you use RAG with a base model, the results are even better than those from a fine-tuned model. This means you can achieve high-quality outcomes without the hefty costs and long timelines associated with fine-tuning.