AI, Large Language Model
The Impact of Large Language Models (LLMs) Beyond Automation
I’d like to circle back to the recent webinar, hosted by National Mortgage News, and briefly highlight some additional snippets that I discussed with Josh Reicher, Chief Digital Officer of our client company, Cenlar FSB.
While the previous blogpost focussed more on how to get started with Generative AI and do so safely, there were some additional points made about how generative AI and Large Language Models (LLMs) have advanced business processes beyond just automation. Josh also answered questions about the impact of advanced AI on business and the return on investment from the technology, which is always a high priority for business leaders.
The Impacts of Generative AI and LLMs
There is a big difference between the Conversational AI that we are familiar with and LLM-based applications. Chatbots, for example, are good at answering requests, FAQs, and other routine queries whereas LLMs are good at a lot more than just providing answers, such as in:
1.Data Analysis: Finding the Needle in the Haystack
LLMs excel at analyzing vast amounts of data and picking out the needle in the haystack. For example, one well-known use case for artificial intelligence (AI) is in the healthcare industry, for analyzing patient scans. Applying AI to medical imaging supports the identification of abnormalities and potential problem areas that may be missed by the human eye. It has been proven that using AI in the analysis of patient scans and reports helps healthcare professionals create more accurate, efficient, and personalized diagnosis and treatment plans.
This is because LLMs are great at finding data and complex patterns in vast datasets. Imagine the potential this has for business in risk detention and prevention, compliance, and even in analyzing buyer behavior and preferences. It’s that needle in the haystack that can be crucial to better decision making.
2. Content Summarization
LLMs are also great at summarization. For example, taking a large document and summarizing it down to its essential components. One great use case for this in the contact center is in summarizing conversations between an agent and a customer so that the context is available for future conversations, for agent training purposes, and for updating customer records. Not only does this save on agent time spent on creating wrap-up notes following a call but the accuracy levels using generative AI are better.
For Cenlar FSB,
“Implementing such a solution has shaved off around 30 to 40 seconds on average handling time (AHT). That’s 30 to 40 seconds of additional time that the agent gets to spend helping the next homeowner versus having to spend the time typing up something.”
[Josh Reicher, Cenlar FSB]
Another application leverages the capabilities of LLMs in data analysis and summarization to perform due diligence on legacy datasets. LLMs excel at analyzing extensive volumes of historical data to review conformance to regulations, especially in regulated sectors such as financial services where adherence to compliance standards is essential.
LLMs can process vast amounts of unstructured data across disparate sources, such as live chat histories, transcribed calls, text chats, and bot conversations, detecting keywords and phrases that may indicate sensitive situations or improper practices. By flagging these interactions, AI helps businesses review and verify compliance more efficiently.
LLMs can also automatically redact sensitive information, protecting personal and confidential details during audits and ensuring data protection. This automation significantly reduces the time and costs associated with manual compliance audits, streamlining the process and enabling faster, more accurate identification of issues that are often then further reviewed by a human in the loop to determine next steps.
3. Personalization
LLMs are great at personalization. So think of mass personalization when you’re communicating with your potential customers.
The ability of LLMs to deliver deeper insights into customer preferences and do so in real time and at scale offers a whole new level of personalization when it comes to things like marketing, sales, product design & development. This then helps free up human workers to focus on higher value tasks such as resolving complex customer problems, channeling their talents into more innovative or creative products and services, and/or more strategic decision-making.
In customer service, LLMs make it possible to personalize recommendations and responses that align with individual customer profiles. Retrieval-augmented Generation (RAG) further enhances this by integrating real-time data retrieval based on up-to-date and relevant data sources. For instance, when a customer makes a specific inquiry, the system can retrieve the most relevant response using specific documents and a vast array of data sources, rather than generating less relevant responses from a larger and more generic public set of data.
The deeper insights delivered by LLMs also enable more personalized upselling. With timely insights into changes in a customer’s profile or situation, for example, a more tailored offer or action can be made at an individual level. This could be applied to detecting that a borrower is interested in moving house and needs a new mortgage, or that they could benefit from refinancing at a lower interest rate. It could even identify the need for additional insurance based on a borrower’s location near a flood zone.
4. Agent Assist
LLM-based solutions do not always have to be customer-facing. They can be used very effectively on internally-focused use cases which may be perceived as less risky than exposing generated responses to the consumer, especially if a company is starting out and testing the waters with the technology.
For example, LLMs can be employed in real-time to review chat as agents deal with a customer. They can flag potential risk for agent responses that don’t conform to compliance standards and give the agent prompts for compliant responses. In doing so they provide helpful feedback to the agent, who is often multi-tasking and may find it hard to be up to date on all aspects of compliance. This application of LLMs in this way also has significant implications for agent training.
And beyond agents in the contact center or customer service, LLMs can be used in many other internal employee-focused use cases, acting as Copilots, helping them navigate through workflows whether that is in automating back-office document processing or managing risk and compliance.
The Return on Investment in Generative AI and LLMs
The best way to sum up the ROI on LLMs is to offer Josh Reicher’s quote:
“The return is really quick. I think that if you’re looking at the impact, I always start with the application. So, picking a safe application is key. The most important thing for us to remember when we use AI is that our clients are looking at us, our regulators are looking at us, to make sure that we are constantly innovating, but that we’re doing it safely and with full transparency.”
[Josh Reicher, Cenlar FSB]
Cenlar FSB has reaped quick benefits on their AI implementations but this is also due to the way they approached the technology. Josh explained how at Cenlar they began with a relatively simple, internal, and straightforward automation project. They focused initially on educating their clients and staff to make sure that everyone understood what AI is, what it’s not, and how it could be used safely and responsibly. They also worked with a trusted vendor partner, ServisBOT.
In addition, they have a human in the loop on the process, even with their chatbots. They use natural language understanding and conversational AI to be able to respond, but the responses themselves are highly curated and approved by legal and compliance, operations, and marketing. Cenlar wants to make sure that:
“When we are providing responses to a homeowner, they’re getting the best, most accurate information at the time”
Generative AI is providing tremendously useful value to Cenlar’s call center and back office. They leverage AI to automate some of the more complex, but repetitive activities, allowing call center agents to spend less time in administrative work and more time being able to help homeowners.
Conclusion
There’s nothing quite as thrilling for me as to hear a client speak so passionately about how they embraced a new, and let’s face it, scary technology. It takes courage but the rewards are there for the taking, once it’s done right and done safely. Cenlar FSB is a great example of this.
As for the impact of Generative AI and LLMs. Don’t necessarily think of them as simply “automation everywhere.” They can be used in many different applications to improve quality, accuracy, decision-making, innovation, and human creativity. So think about automation, but think beyond it.
Finally, I’d like to quote Josh Reicher’s spin on AI as hr turns the focus on the human aspect:
“The technology is about how we get agents or human operations people to spend more time on oversight, on monitoring, and on support for homeowners.”
Download the EBook:The Use Cases for Generative AI in Mortgage