The challenge with current AI systems

GPT4, Claude and nearly every language model today were trained by learning to predict the next word using the entirety on the internet, and fine-tuned for human preferences. This process makes it such that they become great agents for having a conversation, retrieving information, structuring information or writing code – but there’s a downside to it. And this downside is crucial for marketeers.

write me a marketing message that…

Thousands of companies today are going to Jasper AI to do this exact thing: “Delve into…” “Explore the world of…” – “Elevate your cooking with…” are now dominating the internet, creating a world of bland white noise of text that targets everyone and no one at the same time. 

The irony is that AI can be an impressive tool to make marketing robust – but it’s being used in such a way that it has the opposite effect. Generic is the harmful, and for AI-generated marketing to be successful it needs to:

• Understand the product
• Understand who is the target audience
• Understand why the target audience would be interested in the product
• Figure out what kind of distribution channels are more likely to reach the targets
• Design messages that actually create an impact for the target audience
• Incorporate the brand identify, style and language in the campaign
• Learn and adapt from previous mistakes and successes

It’s not just text

An ideal marketing campaign starts by understanding what product we’re advertising: what is its value proposition, features, etc. Overall: why should I care about this product?

The ideal AI system for marketing will store all the information relevant to the product, but also understand the competitive landscape around that solution and be able to develop a very clean understanding of the “so what” of bringing this product to market.

The second step is understanding “who” will buy it: the ideal AI will also have access to the hundreds of thousands of ideal buyer profiles: which roles they have in the organization? What are their pain points? What problems do they face on a daily basis? What’s their favorite coffee order? This data is crucial to understand who are the best buyers who can be targeted, as well as how exactly to communicate the value proposition in a way that makes them genuinely interested.

That’s not enough, though: understanding who the ICPs are doesn’t tell us how we can reach them. Is it LinkedIn and cold e-mails, or could it be that the ideal ICP actually spends all day on hacker news and reddit? The ideal AI has access to data from hundreds of thousands of different potential marketing channels, from podcasts, youtube, to conferences and billboards – to quantify the reach of each channel for the particular audience we’re interested in reaching.

Only after understanding the exact context of the product, the company’s brand, who are the ICPs and how we’ll reach them, we are ready to write a marketing message. But even here there are subtleties: do we want to use any implicit behavioral nudge to get the ICP to wake-up? Perhaps FOMO, Exclusivity, emphasis on value, appeal to authority, social proof? Each ICP will have their triggers, and the ideal AI will not only understand these but also test different messagings to figure out which ones are most successful for each audience and channel.

After creating a campaign, our work is actually just beginning. This is when AI becomes even more important: we want to measure the outcome of each campaign, how they translate into new leads – and how new leads are being converted into revenue. After all, what’s the point in using a technology if you don’t know its ROI? 

To answer this question, the ideal AI will collect data from every campaign source and subsequent CRM interactions, and will attribute leads/revenue to different campaigns to quantify the performance of each one: “Are FOMO campaigns on instagram performing better than exclusive offers on LinkedIn?”. 

“People vote with their money”, so the fact that a specific type of messaging performs better than others gives us instant feedback – which is a lot more valuable than any other generic data used to train ChatGPT. Using that feedback the AI system can adapt and improve its ability to generate messaging that converts: optimizing for specificity – an AI that works exactly for your product, your brand, your ICPs and the channels to reach them.

Finally, by understanding the performance of each ICP/channel/messaging combination, the AI system can optimize the allocation of budget to get the most value – whereas that is measured in revenue, number of leads, brand awareness, etc.