Everyone slapped “AI-powered” on their homepage in 2023. However, Parry Malm was building AI in eCommerce that actually worked in 2015.

Watch and listen to the episode

Before ChatGPT existed, before anyone could spell “generative AI,” Parry built Phrasee, a company that used machine learning to write email subject lines that outperformed humans.

Then he retired. Ate kimchi. Went on walks. And came back this year to build Drumbeat, which does the boring B2B marketing tasks that actually work but nobody wants to do themselves.

The problem nobody thought was a problem with AI in eCommerce

Parry spent years sending millions of emails. He’d write subject lines, test them, try to spot patterns in what worked. Some worked. Some didn’t. Nevertheless, there was no system to understand why.

When he joined Adestra, every single marketer asked the same question: “What should I put in my subject line?”

His answer? “I don’t really know.”

That’s when the light bulb went off. This wasn’t just his problem. It was everyone’s problem. Moreover, there was no good solution.

Specific words trigger specific responses in the human brain. When you have enough examples of words and phrases at scale, you can model human behavior. Consequently, you can predict how people respond to certain language.

That’s what Phrasee did. It built generative language models brand by brand, output human-sounding subject lines, ran tests, and created optimized models that predicted language performance before it was sent.

Why language, not something more “data-based” for AI in eCommerce

Here’s the thing most people miss: language IS data. It’s just not labeled that way in our brains.

We think of data as numbers and metrics. However, words are data points too. Each word choice has a response attached to it. Furthermore, string enough words together at scale, and you’ve got a dataset you can model.

The beauty of focusing on language? It was universally broken. Every marketer struggled with it. There was no good heuristic system or mathematical model to guide decisions. Instead, people just made educated guesses based on vibes.

And critically, subject lines were the most important part of email marketing that nobody had solved. Everything else in your campaign is worthless if people don’t open the email.

The uncomfortable truth about AI in eCommerce

Fast forward to now. AI in eCommerce is everywhere. Every vendor has “AI-powered” something-or-other on their homepage.

But here’s what most people implementing AI get wrong: they start with the solution instead of the problem.

Parry’s advice is simple but brutal: Don’t worry about AI. Worry about the problem you’re trying to solve.

AI is not the starting point. The problem is.

What large language models actually are (and why it matters for eCommerce AI)

When people talk about AI today, they usually mean large language models. Therefore, it’s worth understanding what they actually are.

LLMs work like this: companies scraped the entire internet and built a model that predicts what word is most likely to follow another word.

They’re called stochastic parrots.

The stochastic system follows a normal curve. Most of the time, the answer is within something you’d expect. However, every once in a while, the answer is completely bonkers.

This is why you can’t just plug in AI and walk away. Instead, you need to understand what it’s actually doing, where it’s likely to break down, and when you need human oversight.

What to do right now

Parry’s practical takeaway is beautifully simple:

When you’re trying to solve a problem, don’t worry about the solution. Think about the problem. Think about it deeply. Take in all the information you can.

Then go for a walk in the forest.

The answer will come to you.

(Or take a shower. That works too.)

Blatant plugs

Connect with Parry on LinkedIn.

Find out more about Drumbeat at thisisdrumbeat.com.

The gist

• Parry Malm pioneered AI in eCommerce with Phrasee, which optimized email subject lines using machine learning.
• He realized that marketers struggled with language, identifying a gap in existing solutions for crafting effective subject lines.
• Parry emphasizes that AI should not be the focus; instead, solving the underlying problem is paramount.
• Understanding large language models is crucial, as they can produce unpredictable results without careful oversight.
• For problem-solving, Parry advises deep reflection and taking breaks, as inspiration often follows.

The post What’s the big AI-dea? Parry Malm on teaching machines to understand eCommerce customers appeared first on Browse Basket Buy - eCommerce podcast.