Some Things Were Just Made to be Together
If you’ve ever had a gooey chocolate chip cookie, you know that craving for a cold glass of milk.
If you’ve ever had a bowl of macaroni noodles, it feels incomplete without the cheese sauce.
If you’ve ever had a piece of bread with peanut butter, a slice with jelly must be had to balance it out.
The list goes on: Tom and Jerry, Romeo and Juliet, yin and yang, burger and fries…
Some things are just destined to be together. And that’s how we feel about the combination of predictive analytics and generative AI. Apart from each other, they are useful and can even be revolutionary in the right use cases. Together, however, they can transform industries, democratize data, and spawn entirely new sectors of technology.
Why should we listen to you?
We’ve spent the past few years with our entire focus on perfecting our predictive analytics capabilities. We’ve generated over 35 billion predictive insights and have helped hundreds of organizations leverage them. Our infrastructure, or “engine”, balances speed and accuracy, allowing us to scale and develop predictive models at a rate unrivaled in the industry.
Tell Us About Your “Engine”
Our engine is comprised of 3 main components:
- Identity Resolution
- Instant Predictive Insights
- Custom Predictive Insights
The foundation of everything is identity resolution – the ability to take really sparse contact information for any contact and locate their real world counterparts inside of licensed third party data sources. Without this, the rest doesn’t work, so our next blog post is going to be solely focused on that. But for now, we are going to focus on the last two components of our system: Instant and Custom Predictive Insights.
Taking it back a step: what are predictive insights? Predictive insights (or predictive analytics) are models that are used to (you guessed it) predict people’s behaviors, demographics, preferences, or affinities. In our system, we use a combination of advanced machine learning models, with rule-based and statistical techniques to produce models that can be applied to our entire database of 240 million adult Amercians.
What is the difference between instant and custom models?
Our infrastructure supports two types of models: those that are pre-built based solely off of our data (instant) and those that are built off of segments provided by an organization (custom). Instant models are available “out-of-the-box”, meaning as soon as your contact goes through identity resolution, you can apply any one of these models to it and see the predictions. We outlined these instant predictions in a previous post (here). These instant models are great but they are generic; they are based on the entire US population and not specific to any one organization or region. This is where our second type of model comes in.
Custom models are models that have been trained on segments provided by an organization. For instance, an online retailer can provide us with customers who purchased once versus customers who purchased multiple times. With a few clicks of a button, our system takes those segments, uses our third party data and predictions, to build a model that can predict the likelihood of someone purchasing multiple times. This model is then available ONLY TO THE ORGANIZATION THAT TRAINED IT and can be applied to the entire US population or any other contact list that they would like.
Why are these useful?
The applications of these types of models are endless. The instant models provide a broad coverage of common predictions that can be used to prospect, to analyze past behaviors and drivers, or to personalize outreach. Instant models are perfect for organizations who don’t have enough information on people to make decisions or create their own models.
Custom models come in when an organization needs something that is specific to their clientele or industry. They have plenty of examples, and just need to know how to translate their past customers’ behaviors into their future customers’ decisions.
Both of these can also be combined with other types of analytics (forecasting, diagnostic, and descriptive) to create the “holy grail” of analytics, prescriptive analytics: a step by step guide on what to do next to get the result you want (aka you want to reach a sales goal by EoM, reach out to these 5 people, in this way, for this result).
Why isn’t everyone using predictive analytics?
As we’ve discussed before in previous blog posts, predictive analytics can be hard to adopt within organizations. Even if you trust the models, the outputs of these models tend to be highly technical and very nuanced. This is where the power and beauty of generative AI comes in.
Generative AI + Predictive Analytics: A Match Made in Heaven
It’s no secret generative AI has taken the world by storm. However, if you’ve ever asked ChatGPT specific questions about the behavior of the US population, it’s VERY apparent that it is still just a machine: a smart machine, but one with a basic understanding of human nature and almost no specific knowledge of the US population and its behavior. This is where the harmony of predictive analytics can make generative AI really sing.
Now imagine this…
Here’s how ChatGPT responds when asked a specific question about the behavior of the US population:
Here’s how the latest iteration of boodleGPT answers:
That’s a specific, actionable answer that is made possible by our predictive analytics engine. (Note: future versions of boodleGPT will give more verbose, more explanative answers — this is the equivalent of a toddler learning to walk.)
None of this is possible however, without the ability to identify people within different datasets. Next our CTO Ansel Teng will go over how we built our identity resolution engine and why it’s such an important component in building a Middle Layer AI like boodleGPT.