Featured Article:

Author: Shuvajit Basu

GenAI has taken the world by storm. I certainly don’t have to outline the numerous possibilities it brings to bear. Both GenAI and Predictive AI have their share of strengths andbopportunities. However, if we can combine the power of these Generative models with Predictive AI/ML models, we might be able to do something more transformative.

Let us take a few examples to corroborate this point.

First, if a user were to interact with a conversational AI interface to check out what could be ‘some choices for an evening wear in a formal get together’, the GenAI response would be the same for any user who would type this prompt. However, if this prompt were to be augmented with the same user’s likings, for example her preference for a certain style, brand, price, etc. the response would be more personalized for her and needless to say would seem more relevant and thereby translate to higher engagement.

Similarly, in ecommerce search, a search query would return the same set of search results regardless of the user. However, if we were to layer in user tastes and preferences, the results will differ considerably from user to user. While this has been in practise before the advent of GenAI; companies experimenting with GenAI search esp. in the case of theme-based search, would apply similar principles. Suppose a user is looking for items to go along with a ‘IPL watch party’. In this case, the query would be augmented with user’s preference such as a specific brand of beverage or chips or even healthy food options and return results from various categories that is relevant to the theme.

In the same vein, as GenAI continues to get adopted by Marketeers to design their campaigns with specific creatives and copies targeted at a certain profile/cohort of users, the same can be extended to n=1, whereby user features can be augmented to the system prompt to design very specific user level campaigns.

All of the above assume that there is enough and rich user-level data to make a difference. Clearly, the above examples illustrate that GenAI and Predictive AI complement each other and not necessarily compete against each other. While Generative models are trained on world knowledge, there is a need to complement with specialized or domain knowledge. Also, above examples illustrate that it’s not only retrieval that matters but retrieval & selection that makes the response for each user and each of their interactions very personalized. Hence, one needs to understand the strength and limitation of Generative & Predictive AI to use one or both effectively.

Title picture: freepik.com