AI for Leasing - Is it a good idea?

The pros and cons of using AI for scattered-site leasing
Jul 25, 2024
AI for Leasing - Is it a good idea?

Short answer:

Yes, but not yet. Current AI models hallucinate answers and give prospects a clunky experience. But they're getting better quickly and you should start paying attention.

Long answer:

It's 2024 and AI is everywhere. Even Chipotle is getting in on the action with AI that makes better burritos.

Can AI help me with my scattered-site leasing?

In this piece we're going to cover what AI can actually do, its drawbacks and the best tools on the market.

What AI thinks AI-powered leasing looks like - very meta

First Line of Defense

When we talk about AI we're talking about Large Language Models (LLMs) that have been trained on billions of paragraphs of human writing (Wikipedia, blog posts, emails, etc.). They've "learned" how humans talk and can answer most general knowledge questions.

ChatGPT is a great example.

You can go a step further and "finetune" your AI model on your own data. For example, you could finetune your model with data about your available properties (beds, baths, features, listing description). Now when you ask the model how many bedrooms are at 123 Main Street, it should respond correctly.

AI bots can respond to the easy questions. But they hallucinate and can't handle harder questions

They're a decent first line of defense when the prospect asks a clearly worded question. And it can save your team a few hours each week by tackling these questions first.

Unfortunately, not every prospect interaction is this simple. AI struggles with making up answers, dealing with complicated questions and handling phone calls.

A sample of the current real estate AI vendors

Not too intelligent (yet)

These LLMs are getting better every day as venture capitalists pour billions into buying chips to train better models. But there are few major unresolved issues when deploying customer-facing AI.

Hallucinations

This is the biggest challenge that AI engineers are trying to fix. When an LLM gets a question that it hasn't seen before, it will "guess" the answer based on similar data it has seen before - called a hallucination. This isn't a big deal if you're asking ChatGPT what temperature to bake your cupcakes at.

But it's a major liability if a prospect asks "Can I pay $300 less for rent on this property?". The AI hasn't seen this exact question for the property and while trying to helpful it responds "yes".

Now your company has technically accepted an offer for $300 lower rent - not good!

A recent example of ChatGPT hallucinating. ChatGPT is the most powerful LLM currently available

Can you repeat that?

Sometimes the AI doesn't hallucinate, but it can't understand the question. That might be because the question includes a lot of personal context, is multiple questions lumped together or is weirdly worded.

The LLM will try it's best and usually that means that it answers the wrong question or only part of the question.

Unfortunately, that created a bad experience for the prospect who feels like they're not getting the attention they deserve. Even the best AI models have to forward that difficult question to a human who can only respond during business hours.

A little too artificial

While the LLMs have been trained on human-written content, they don't always sound too human. If you've spent any considerable time using ChatGPT you might have noticed that it sounds a little robotic, with weird sentence structure and uncommon words.

Prospects don't want to feel like they're talking to a robot | Credit: Freepik

Young prospects who have used AI tools can easily spot when they're talking to an AI. This puts up their guard and makes them less willing to engage if they known they're talking to a robot instead of a person. They'll just move onto the next property where a person answers their questions.

Not Omnichannel

Our data shows that 90% of prospects prefer texting to calls and emails. But 65% of prospects who complete an application have called at least once. They call if they have a complicated, multi-part question that takes too long to send as a text or if they have trouble completing the application.

But AI can't handle phone calls. So you'll still need someone who's available around the clock to help prospects who call in. Don't sacrifice your leasing conversion rates with a poor prospect experience.

Do your homework first

If you're still ready to experiment with AI for leasing, here are some questions you should ask the sales rep:

  • How does the LLM know if it can confidently answer a question without hallucinating?

  • What happens when it encounters a question it can't answer? How is that question passed to me and what does the prospect see?

  • Do they have any warranties that pay for hallucinations which cause damages (like promising a lower rent to a prospect)?

  • What are the CSAT scores from prospects who use the AI? (this is an industry metric that measures customer experience)

RentEngine and AI

Here at RentEngine we are actively investing in AI for faster leasing.

But we're not there yet.

The key is in training the most powerful LLMs on proprietary data of actual conversations with prospects. That's a painstaking process of collecting, cleaning and organizing millions of data points (all anonymized). We are confident this will deliver much better results than pre-trained models that use contrived examples of prospect communications.

In the coming months we'll be sharing more details.

If you're a scattered-site property manager, get better results than AI with RentEngine's around-the-clock leasing operations solution. Schedule a call with the RentEngine team today.

Share article