Exalted CTO Henry Li on understanding the limits of AI

Editor-in-chief Sarah Wheeler sat down Exalted CTO Henry Li to talk about the limits of artificial intelligence and how it is reshaping his company’s roadmap.

Sarah Wheeler: What sets Lofty’s technology apart?

Henry Li: Lofty is a platform product; that has been our vision from day one. We want to be the operating system for our users and their essential operational staff – for brokers, brokers and brokers, no matter their size.

Platform can mean many different things. At Lofty, it’s an end-to-end experience: from searching for consumers on the IDX portal to maintaining the customer relationship in a CRM and converting that relationship into a real transaction. Completing that transaction generates all kinds of operational insights into the system to make smart business decisions.

Platform also means that we are not alone. We have a pretty powerful marketplace bundle with our product, which has two parts. One of these is of course that we have an open API system, so that our partners can build their own apps on top of the system. We also allow them to use other partner services through our open API system and our marketplace offers many different value-added services.

One thing we’re a little different in is that we have pretty strict control over the user experience. So for the important value-added services – marketing automation services, automated campaign lead generation services – these services are essentially created by us, but we also have a pretty open system. We work with select partners to distribute their services through a native experience at the top of our marketplace.

The other thing about being a platform is that when we design the product, we believe the product should be generic; we are not just building a product for the real estate sector. We’re a proptech company, so it’s not just real estate agents that can use the platform, but also adjacent businesses like mortgage brokers and property management companies. We’re not trying to build something like Salesforce, where you have to hire two Salesforce engineers to implement the whole system. We would need very minimal configuration to run the business on top of the platform.

So the last thing is that we are a scalable platform, which means we can provide an experience from a single user all the way up to an enterprise user with 10,000 users under that account.

SW: At our recent AI Summit, we broke AI down into three tiers: automation, traditional AI, and gen-AI. How do you use AI in the context of these three levels?

HL: Automation has of course been one of our selling points from day one. And if you think about the traditional definition of AI, we certainly have a lot of that, like predictive analytics. We score lead insights to help the customer understand the quality of the score and we have integration with third party data enrichment services. There are also many recommendation systems embedded in our system, for example the listing alert.

The last part is gen AI: we’re talking specifically about large language models (LLM). I think the most important thing is to really understand the limits of the technology. We were one of the first vendors to introduce AI-generated content (AIGC) features to help our customers generate listing descriptions, ads, blogs, text messages, emails, you name it.

I think AI is very exciting because we’ve had to completely tear down our roadmap three times at this point as the technology and the technology service providers are making tremendous progress. Because technology is evolving so quickly, we must evolve with the technology. And when you talk about AI features, it’s never a complete feature. Building AI functions is different from the old concept of building software functions because it doesn’t necessarily involve writing the code. You tune the AI ​​to give the AI ​​more capabilities. So it is constantly evolving.

We’ll start by looking at what kind of work our users hire out. They often hire an ISA company [Inside Sales Agent] or a remote virtual assistant. The main thing we’re focusing on is the AI ​​assistant. We also have a fairly large call center that serves as frontline support for customers, and one of our AI features is a service bot that helps with all product-related queries. We have a lot of conversations with users about how to diagnose a particular problem, so we use those documents to train an AI assistant.

But this is actually just the beginning. The exciting thing is that we can gradually replace some of the tasks that they have transferred to our call centers. It won’t happen 100%, but with this kind of infrastructure we can replace and enable that AI system piece by piece, gradually providing a real, human-like service to our users. This is actually what we consider phase two.

The challenge for us, and the interesting problem we have to solve, is that we think software is a perfect workbench for agents to have their data and we want to provide all kinds of tools to help them perform certain tasks. So what we need is a scalable approach to taking new capabilities for AI agents and implementing them into our product. We are not talking about two AI assistants – what if it is 100, 200 or 300 different types of AI assistants that help our user perform different types of tasks?

I think the biggest challenge for us going forward is that we need to redesign our software to make these capabilities accessible to our users in a very intuitive way. So we talk about some of the features that we’ve already released, but compared to what we’re going to release, this is really just the beginning.

SW: How are you adapting to the changes that are part of NAR’s settlement agreement?

HL: From a software vendor’s perspective, it won’t change much of the core of the software. There are some small, very fundamental things we’ve already changed, but now we’re having discussions about building features that make sense for our users in this environment.

First, we need to help buyer’s agents better present their value. There is now a completely different process: from lead acquisition to the marketing side. It is important for our users to connect with the customer as quickly as possible, so you will need to provide a buyer with the experience of quickly reviewing the agreement and agreeing to use you as a buyer’s agent. This is what we want to optimize. We can help our users improve their speed of agreement.

SW: What keeps you awake at night?

HL: Understanding the limits of this technology. That’s really changed, so there are some fundamental questions to think about. Can AI replace the agent? Will AI replace software? Our current answer is no for both.

As an agent, you build a business and don’t perform a job. It’s maintaining a customer relationship, it’s raining. Is AI so powerful that it can run a business? No. Can AI replace software? Software isn’t really in conflict with AI. Software is where users store their data to manage their business. Software is essentially a workbench. So that doesn’t hold me up anymore.

What keeps me up at night today is that we don’t really know what the industry will look like in the future and how quickly technology will evolve. As I said before, we have already had to abandon our roadmap three times. As a product company, we look at the new paradigms or frameworks for building a product. Building software right now is very different than it used to be, so we remain vigilant and very focused. Honestly, it can be quite overwhelming as we receive new information every day. I don’t really sleep much!

It’s stressful planning, but at the same time it’s very exciting because as the environment changes and technology changes, we get new opportunities. Making sure we don’t miss those opportunities is what keeps me up at night.