Suggestions
Kevin Karty
Innovation & Insight Champion - Product and Strategy Fanatic - Searching for the Good in Everyone
Kevin Karty is a proficient individual who specializes in launching disruptive products and services to revolutionize industries through cutting-edge technologies.
His expertise lies at the convergence of analytics, marketing research, technology, and product innovation, fueled by an insatiable curiosity that constantly motivates him to expand his knowledge base.
With a keen interest in consumer behavior and decision-making processes, Kevin leverages a blend of academic insights and practical experience to develop innovative solutions.
His ultimate joy lies in reigniting imagination and passion in individuals striving to make a positive impact on the world.
Highlights
At a recent conference, someone was telling me how they used ChatGPT to create segments that were just as good as a million-dollar segmentation study that another supplier had run. Two thoughts flashed through my mind:
1) Oh no… AI is going to make me irrelevant… 2) A million dollars? Intuify would have done a MUCH better job for a lot less!
Well, I finally got around to testing it out. I asked ChatGPT to build me 5 personas for home telecom customers. And it did. And they weren’t completely terrible.
But they weren’t GOOD. They were just… bland. And kind of weird. They certainly didn’t map onto the segments we delivered, and they weren’t really actionable. They were caricatures – stereotypes that were essentially lifted from online blogs and reddit posts. For example, here was one segment:
The Student Renter
Demographics: Aged 18-24, college or university students, often in shared housing or dormitories.
Behavioral Traits: High internet usage late into the night, uses the internet for academic research, social media, gaming, and streaming. Budget-conscious and often seeks short-term contracts.
Values: Affordability, high data allowances, reliability during high-usage hours.
Preferred Product Features: Affordable plans with high or no data caps, discounts or promotions for students, options for seasonal suspension of service during holidays.
So here’s the problem with this segment. It actually DOES NOT EXIST. The vast majority of students don’t control their home internet. They get it through their dorms, their apartment buildings, their frat houses, or their parents’ home. Even then, this segment is basically just an age group with some stereotyped needs.
I then asked ChatGPT to create a full persona for this group, and it gave me Alex the Tech-Connected Student. Alex is exactly who you would expect Alex to be. No surprises at all. Literally, not one single thought-provoking detail. I mean, if you closed your eyes and imagined Alex, that’s him.
The best objection to what I did was that I used a general purpose LLM. What if the LLM had been trained on other survey data – for example, other attitude, usage, shopper, demographic, and behavioral datasets from research the client had done in the past 6 months? Would it have done better? Well, yeah – almost definitely.
It also would have done better if you’d given it the actual dataset from the survey we ran for the segmentation.
But that’s kind of the point, isn’t it? At that point, LLMs are doing the same thing our other clustering and segmentation models do, but faster…
Just to be clear, I love AI and LLMs. I'm just not sold on fake people. To simplify, this is where I stand currently:
LLMs + real people = amazingly awesome
LLMs - real people = awkwardly average
