The voicebot, which was launched in beta mode in August, can help customers get answers to mortgage application inquiries, call them back if they abandon an application and ask them about the weather, while it connects borrowers to a loan officer.

Betsy took nine months to develop and is built on the company's proprietary Tinman loan origination system. Its main goal is to keep potential borrowers on track while filling out their loan applications.

Vishal Garg, the CEO of Better, wouldn't disclose the monetary investment of building such a system out, but said it would be worth it in the long run.

"We think the return is going to be extraordinary," said Garg. "While the investment in Betsy and the time and energy we've spent on it is extremely costly upfront, we believe it will enable us to serve 10 times more customers." 

Garg pointed out that Betsy can also contact borrowers from lead referral sites such as Lendingtree, Credit Karma, Nerdwallet and Bankrate.

"Frequently the customer is most ready to transact when they're submitting their information on one of those platforms," he claimed. "Traditionally it would take minutes for our loan officers to contact the customer by which point in time they're out doing other things, so we needed something that could contact hundreds or thousands of consumers whose leads are generated, so Betsy is able to reach consumers right at the time they're most interested."

Garg says that since consumers have already provided some part of their information, Betsy contacting them won't be in violation of soon to be implemented robocall restrictions.

"Betsy is an intelligent loan assistant that is asking the consumer for the pieces of data that are missing, and the consumer has provided consent to be reached out and started an application, so it's not a robocall to someone who's not expecting it." 

Better's CEO noted the company experimented with AI chatbots, similar to what many competitors have introduced on their websites, but that borrowers don't seem to be fans.

"The chatbots simply don't work," he claimed. "We've tested chatbots before, and consumers don't like them because the latency is too long. And then when it doesn't know the answer, it's not able to transfer it to a person who actually does." 

"We recognize that consumers are far more comfortable interacting on the phone than they are through a chatbot and so we raced ahead and skipped it, keeping in mind how we can use true generative AI," added Garg. "[We] started thinking about how we can build the first voice assistant for the mortgage industry." 

The voice assistant was built using large language models, "which have been fine tuned and taught based on all the data we have on our tinman platform, plus the underwriting guidelines across 32 mortgage investors" the CEO said.

"The answers are driven by generative AI and are natural. It is not scripted. It's fine tuned to follow pathways to get the missing information that's what helps us become faster, but the responses are true generative AI," he claims

Some industry stakeholders have previously expressed skepticism about AI voice assistants in the mortgage space, pointing out that there are no products out there yet that truly understand the financial services space and can be used with confidence.

"The models are not yet in a place yet where you can offer real customer service," said  Jason Bressler, chief technology officer at United Wholesale Mortgage, in a previous interview.

Others have expressed worries about regulatory hurdles and data privacy concerns.