Like Finding a Needle in a… Well, Pile of Other Needles

finding-needle-in-a-box-of-other-needles-boodlegpt

Imagine that you are making a phone call to your bank to discuss a loan. Your bank has a ton of information about you that will enable them to approve your loan. However, before they can discuss any of these matters with you, they will first need you to prove your identity by asking you a number of questions. Once they establish your identity, they can have a meaningful conversation with you about your loan based on the information they have about you.

Similarly, at boodleAI, we have 35 billion predictive insights about the 250 million adults in America that can answer many questions about their consumer and philanthropic behavior. Those insights are useful in their own right, but they become even more meaningful when they can be connected to the contacts in an organization’s records (think prospect list, customer list, or donor list). Making that connection is the process called identity resolution, and it is the foundation of what boodleAI does.

How our identity resolution engine works

Our identity resolution engine takes in a few pieces of personal information, such as a name and an email or residential address or phone number, and uses our proprietary, patent-pending algorithm to find the best match against the 250 million individual records in our database.  Once the best match is determined, an enriched profile of 1200 data points and over 120 predictive insights is generated. 

The engine also provides a match confidence indicator so that the user can choose to use only the high confidence matches to maximize accuracy, or both high and medium confidence matches to maximize reach.

The challenge of identity resolution

While it sounds very straightforward, identity resolution is a notoriously difficult problem. The difficulty lies in the noise in the input data (the donor data could have typos or outdated information), and the incompleteness of our database (we may not have all email addresses and phone numbers an individual has). While the theoretical foundation in solving this difficult problem was well established in the late ‘60s, implementing it in practice is no small feat.   Adding to the challenge is the scale we need to operate on: 250 million people. Our identity resolution engine was created from day one of boodleAI because of its foundational role. Over the years, our engineers have enhanced and evolved the engine as we learned, and applied rigorous methodology in tuning and validating the results.  

Why customers trust our identity resolution engine

We’ve extensively tested our identity resolution engine using independent datasets. Our identity resolution engine can match 70 to 80% of individuals based on name and email alone, and over 90% with up to date residential addresses. It matches at an impressive speed of over 10,000 individuals per hour. Hundreds of customers have used our identity resolution engine to gain insights into their contact lists. Their feedback has allowed us to continuously improve our identity resolution engine.  

Why identity resolution matters to Middle Layer AIs

Our identity resolution engine allows a Foundation AI like GPT to do something it can’t do today: generate responses that incorporates insights about the people in an organization’s records.    

Let’s look at boodleGPT compared to a spreadsheet or CRM:

What a spreadsheet or CRM can tell you:

Our company had 567 sales in Q3. All sales were of golf products. 

What boodleGPT can tell you once a company’s data is ingested:

It’s the same information (although boodleGPT allows for a natural language query – which is convenient but not groundbreaking).  

Here’s the difference. This is what boodleGPT can tell you by using identity resolution to connect the 567 customers in Q3 to their identities in our dataset of 35 billion insights about 250 million adult Americans:

This is useful!  Generating the above profile is simply not possible using an organization’s records without identity resolution and access to a data set of insights about the US population. 

boodleGPT can do even more. For example, it can identify markets with similar profiles:

Now, let’s leverage the power of GPT’s generative AI combined with boodleAI’s predictive AI to get a recommendation for an advertising campaign:

The above response could not be generated by either the CRM, boodleAI, or OpenAI by itself.  It requires the fusion of (1) the company’s sales records in the CRM with (2) boodleAI’s predictive insights using identity resolution and (3) OpenAI’s generative AI capabilities.  

From here, you can start providing prompts to boodleGPT to generate advertising ideas:

Okay, this part of the Middle Layer AI may still need some work…

The above are screenshots of actual prompts and responses from an early version of boodleGPT, which we’re currently training, testing, and tuning to improve its responses.

Though it’s vital to have a unique data set and the infrastructure to make it usable, where the rubber meets the road is the team that makes it all come together. Next our Co-CEO Shawn Old will tell you about the journey boodleAI’s rare talent took to make all this “AI stuff” possible.

CTO of boodleAI, Ansel is a senior technology leader since the dot com days. He has a Ph.D. in computer science and has been managing software product and system development in organizations of all sizes, from small startup to global enterprise.

Connect with Ansel on LinkedIn. He loves to connect with fellow technology leaders.