Using GenAI to personalize PDP creative

An emerging use case for generative AI tools is serving more relevant product information and default images to different user personas.

Ash Rathie , co-founder of Bullsai has already built a solution for this case, and we caught up for a chat on how this tech works on PDPs (product detail pages).

Summary of how it works:

?? The app looks at types of customers that are coming to your store and defines personas based on visitor attributes (referral source, geolocation, profile data, clickstream etc)

?? Using a ChatGPT-style LLM (large language model), the app looks at your existing product title, description, attributes and tags to generate different product description versions that highlight the features and benefits most relevant to these personas, and stores them in its back end

?? Using a computer vision model, the app reads your product image set and selects the image that best fits a given persona when serving content to a given user

For example, a middle-aged shopper may be served a description highlighting anti-aging benefits with a relevant primary image:

And a teenage shopper may be shown variants focused on anti-acne:

Q: From a merchant’s perspective, how much effort is required to train the AI, or is it plug-and-play?

A: It’s a bit of both. We wanted to design the system in a way that it accommodates different levels of data sophistication. You don’t need a customer data platform (CDP) like Klaviyo or product information management system (PIM), but if you do have those things, it makes the personalizations even better.

Q: How does the app personalize to a brand new visitor you don’t know anything about?

A: Even with small amounts of data like just an IP address, we’re able to make really powerful assumptions about the shopper – their region or location can for example tell you how important SPF is in skincare, and boost these properties in the messaging. Of course with existing sales data, if you already have a very strong data story where you’ve done customer research, we can integrate with your CDP and feed that into the LLM.

Q: Is this just for PDPs?

A: We’re starting with PDPs but plan to expand to every surface of the online store including home pages, category pages, etc.

Q: One of the fears is if you just let Gen AI write a product page for you, you might be missing certain context like brand voice or something you know about your particular customer base. How can you protect that within the AI model?

A: We don’t believe it’s smart for retailers and brands to adopt a system where AI takes full control. AI is the system that manages all this stuff. The way our tool is built is more of a co-pilot. We understand it’s super hard to write all these different versions of product content for all these personas. It’s pretty much impossible, which is why it wasn’t being done before this.

So the AI’s role here is to make that process of creating different versions of each persona easy for you. But at the end of the day, it’s the ecommerce manager or marketer that has control, which is hey every single thing the AI does needs to be approved by a human first.

We have a section in our product called “Brand Voice” where we put all the knobs and dials for how to tailor how the AI speaks and set up safeguards to protect your brand. We have standard “banned” words and custom banned words you can enter. You can also add things like “ideal descriptions,” so thee AI will follow that style of writing to limit the number of edits and time it takes to review and approve them.

You can also put your brand story/history, general brand tone and custom domain knowledge about your products, industry, customers, etc. to guide the LLM.

Q: Where do you think this technology is going in the next 6-12 months?

A: Personalization has been a hot topic for years, but we thing it’s been an unfulfilled promise by recommendations and retargeting emails – that’s what we call personalization today. But we believe true personalization is when the website itself morphs to match who you are as an individual. And that includes changing images and text, and long term – how the website looks and feels. Can the UI really change to match your cognitive style.

So if you’re a very detail-oriented shopper, you see a lot of detailed specifications. But if you’re a more visual shopper, the images should be more prominent and the actual descriptions and specifications shouldn’t be as prominent.

For more info on GenAI for product pages, check out Bullsai or connect with Ash Rathie

Want more ecommerce ideas? ?? Check out the Ecom Ideas newsletter for more in-depth topics on optimizing your digital experience.