AI, Large Language Model
A Discussion on Generative AI – Doing it Right, Doing it Safely
Following a recent webinar, hosted by National Mortgage News, that I had the pleasure of participating in with Josh Reicher, Chief Digital Officer of our client company, Cenlar FSB, a number of great discussion points around Generative AI implementations emerged.
I feel the highlights of our discussion about approaching AI initiatives in a safe and secure way are worthy of sharing with you in this blog (more to follow in another blog).
It’s one thing to read about research on Generative AI and all the interesting articles that are out there on AI topics, but it’s quite another thing to hear firsthand how a mortgage servicing company like Cenlar has approached their AI implementations. Josh gave some great insights into how to approach AI in ways that deliver fast business value but without the high risks and security implications that are so commonly feared.
The Future of Generative AI will be Driven by Business
It’s become clear that Generative AI is ubiquitous, both in today’s business landscape and in our personal lives. A recent Gartner survey found that Generative AI is the most frequently deployed AI solution in organizations, creating a window of opportunity for business leaders to deliver value at scale. The power of Gen AI is undeniable and is expanding automation capabilities for businesses, far beyond traditional automation technologies.
While AI has been around for decades, the last 18 months have brought a seismic shift. I compare this to the transition from dial-up internet to the emergence of broadband internet. AOL and dial-up internet seemed really cool and then suddenly we had broadband internet services everywhere. It’s that kind of seismic shift.
It’s similar to Artificial Intelligence. What seemed like all of a sudden, the real power is there with Gen AI and it’s everywhere. This is the shift that is changing the opportunities and the conversations in business. We’re seeing conversations being held at a more senior level now where it’s about the business cases. It’s not about whether the business should do it, but how fast the business should do it.
Millions of people are using ChatGPT daily, so there’s so much more understanding and appreciation for the technology. Now, it’s the business that is going to drive the technology forward from this point. There’s no putting the genie back in the bottle.
However, businesses still have concerns about security, data privacy, and job displacement. Even though the pace of innovation from companies like Amazon, Google, and OpenAI is driving rapid change, there are some good tips from Cenlar on addressing these challenges. Indeed, at ServisBOT, we spend time with clients helping them understand and implement the necessary safeguards so that these risks are mitigated.
Overcoming Security and Data Privacy Concerns
Tip #1: Start Small & Use a Partner
The key to overcoming security concerns is starting small and working with partners who understand the technology. This approach allows businesses to test AI in low-risk scenarios and gradually scale up.
But how do you get it into your organization when you’re not ready to make the large-scale investment in staffing, in infrastructure, in knowledge, in governance and education, and everything else that is necessary to make it work?
“Don’t be afraid to start small and don’t be afraid to start with a vendor partner.“
[Josh Reicher, Cenlar FSB]
That’s exactly how Cenlar started on their AI journey. Josh highlighted how they began with a relatively simple, internal, and straightforward automation project. By doing so they were able to spend more time building out the internal skills, infrastructure, governance, and guiding principles. 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.
“We were able to create an entire program around that because we had a partner in place that was certified. This allowed us to hone our own skills while at the same time delivering value to the business – day one.”
[Josh Reicher, Cenlar FSB]
Tip 2: Look for Low Risk Use Case(s) and Scenarios to Start
For Cenlar, one of the biggest guiding principles was not to put generative AI in front of a homeowner. Since it is a new technology and given that generative AI can hallucinate in providing an answer, this was very important. However, as they mature in their approach and as AI’s maturity grows, this might change for them. But for now, the important thing is that generative AI is not something that they use in a sensitive situation, for example, helping a homeowner when they are in the process of purchasing a home.
“What generative AI can do well with a human in the loop is it can improve automation. It can remove the repetitive tedious work that, say, an operations person or an agent may have to do, but they can spend more time overseeing the results and the recommendations that come from AI and make sure that it looks good. And more importantly, spend more time with homeowners and help do that. So it’s much more of a back-end solution at this point for us, a back-end automation of things that we’ve been traditionally automating.”
[Josh Reicher, Cenlar FSB]
Tip #3: Start with a Simple Use Case and an AI Partner
When asked about doing a sandbox, Josh says “start with the right use case that helps you get going with what’s there today. You don’t have to spend $10 million building a sandbox, training a model and have it doing this and that and the other.”
EBook:Discover the Use Cases for Generative AI in Mortgage
Don’t forget that there is a difference between using OpenAI’s ChatGPT that we all are exposed to versus some enterprise solutions that are out there. ChatGPT on Azure, for example, has a lot more security controls. This is why Josh advises working with a partner.
“Don’t be afraid to start with a vendor partner. There are many vendor partners out there that are building at a scale that is difficult for an individual company. They can help you deliver and generate AI solutions that will provide value.”
[Josh Reicher, Cenlar FSB]
Start with a use case and a partner and you’re off and running. Start small. When he says simple, he’s talking about removing PII data, for example, and having AI do the easy repetitive tasks that it is really good at. Then you don’t have to worry about losing data.
“A partner can help you, for example, make sure that redaction happens, so that PII is never going out, sensitive data is not going out.”
Josh also emphasized the importance of choosing flexible vendor partners for AI implementations. As the technology evolves, businesses need partners willing to co-innovate and explore new applications. This flexibility allows companies to adapt and scale their AI initiatives effectively.
Highlight Use Case from Cenlar FSB: Starting with Generative AI
This first implementation of Generative AI for Cenlar FSB was intentionally simple. The idea was just to minimize risk, allow more control, and ensure governance was put in place. So, really straightforward but with a quick return on investment.
At the end of every conversation and every live chat session that an agent has with a homeowner, they have to stop what they’re doing, and spend time summarizing the conversation that they just had. That information has to be captured and correctly categorized for subsequent conversations. This can be put directly into the Large Language Model (LLM) and be prompted to have it feed back in the right language, the right standard, the right size to fit the field, etcetera – in other words a perfect summary of the results.
That summary is then presented back to the agent in real time, and the agent can then decide whether or not they want to put that into the system or make changes to it along the way. This has shaved off around 30 to 40 seconds on our 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 they just typed up in the live chat conversation.
This small and simple use case helped build internal skills and governance while delivering immediate business value.
Conclusion
There were more takeaways from this webinar besides the above and I will summarize these in a follow up blog. Josh and I agree on the fundamentals of a safe and valuable approach to generative AI implementations, as highlighted above.
In order to help overcome the fear and not “miss out” on the immense potential of generative AI, the key to getting started is to start small. Don’t try to boil the ocean when it comes to choosing a business use case. A simple internal application for Gen AI can be a great way to test the waters, gain buy-in, and understand the nuances of the technology.
Choosing a knowledgeable AI partner is also wise. The technology is relatively new and fast evolving. An experienced vendor has built up AI knowledge and insights into what works and what doesn’t, and can help the business implement solutions that address security and data privacy while generating high value. Business FOMO is real when it comes to generative AI. ServisBOT is here to encourage businesses not to miss out and to help them be part of the seismic shift that is Artificial Intelligence.
For more insights, listen to the webinar recording “Harnessing AI: Boost Efficiency, Cut Costs and Enhance Customer Experience“