AI Insights: Generative AI to vet rental homes

We want to help you keep up with how brands across the travel industry are exploring - and using - generative artificial intelligence solutions from companies including OpenAI, Google and more. So we're surveying technology professionals at leading travel brands, and we'll be publishing their answers periodically here. 

Doron Meyassed, co-founder and CEO of rental home provider Plum Guide, is the latest to offer his insights on generative AI and its impact on the travel industry. 

We began working with generative AI … three years ago. In the past three years, we have been on a journey from near-total human vetting of homes to the predominantly AI-led curation of our collection. 

Our first foray into this world started by teaching a model how to vet homes for their aesthetic. In effect, we taught the model our detailed design standards and expectations, based on the thousands of homes we had previously passed and the millions we had rejected based on their design. Continuously improving and evolving the model has led to an impressive 91% match rate between what our creative director deems Plum quality and what the creative AI does. 

Our current work with generative AI is focused ... on three areas:

• To vet our homes: The core of the Plum proposition is based on personally vetting homes and selecting the top 1% at three price points in every market. We roughly accept only one in every 19 homes that we look at. We used to do it by sending hospitality experts into every home, spending hours in situ and collecting 500+ data points to make the decision. Today we use an assortment of AI models, which are able to do most of the vetting remotely with (almost) equal results. 
• Review synthesis: We pool every public review ever written on a home. We then use AI to pool keywords and collate consistent sentiment across stays. That allows us to build an initial picture of whether the home meets our basic Plum standards: e.g., cleanliness, Wi-Fi, responsive hosts, well stocked kitchens, luxury toiletries etc. The result is a confidence score against our standards and any red flags.  
• To produce our listings: With recent innovations in OpenAI and ChatGPT, we have taken this one step further in the last year and completely transformed our listing building process. We now leverage AI to name our homes, craft descriptions in the Plum style and tone of voice, and even analyze guest sentiment to craft our unique home truths section of the site (which lists all the negatives about the home). A listing used to take 60 minutes of copy writing to produce. Today it takes 15 and converts way better. 

These innovations have produced five times the number of homes added per person, slashed the cost of adding a home by 85%, all while maintaining our conversion rates and net promoter score. 

The biggest challenges for us related to generative AI are … inaccuracies. There have been some speed bumps along the way. For instance, the model sometimes generates confident responses unsupported by the provided data. Improving the prompts iteratively to achieve better quality output has also been a challenge. But with new tools emerging, we're excited about what the future holds.

Similarly, using AI to produce copy is far from perfect. We still need human input to ensure the AI doesn’t go rogue. One of my favorite examples is home naming. An AI names every home. But sometimes it gets it wrong. The AI has previously named homes "The Dustbin" or "I'm A Basement Dweller" - not exactly enticing names. That's why human vetting is still a crucial (albeit a smaller) part of the curation mix. 

For the travel industry overall, we see the most potential for generative AI … in everything from personalized trip planning to AI-powered customer service. 

One of the innovations we are most excited about is text-based searching. We imagine a future where searching for a home or hotel is done through long text descriptions. Where a user can describe in detail what they are after: the vibe, the feel, the amenities, even what they hope to get out of the trip. And the perfect home or hotel is surfaced. 

We are also blown away by the power of AI to spot mediocrity. Mediocrity that the naked eye fails to spot. Our AI often identifies homes that are “meh” in real life but look amazing on the surface (e.g .review rating of 4.9). If reviews killed the terrible experience, AI will kill the mediocre one. 

Travel is ripe for AI-led disruption. It has huge, very well structured data sets, a big consumer problem and a big prize for those that get it right. 

One year from now we expect to be using generative AI for … continuing to improve the model, getting as close as possible to 100% success rate vs. human vetting. I anticipate that we'll be using AI in even more innovative ways to assist our customers. 

At the same time, we recognize that as the presence of AI grows, so too does the value of genuine human interaction. Our goal is to strike the perfect balance, offering our hosts and guests the best of both worlds. With our exceptional care and matchmaking teams, we are well-positioned to do just that.