Customer support · Production

StreetEasy deploys two LLM-powered features for instant property FAQs and personalized agent introductions

The problem

Home shoppers spent less than 61 seconds on StreetEasy property pages and struggled with traditional FAQ search engines that returned slow, inefficient, and irrelevant results. Separately, finding a well-matched real estate agent was a complex and intimidating process for buyers.

First attempt

Traditional web-based FAQ systems depended on slow, inefficient search mechanisms that required users to navigate complex search engines and irrelevant results, causing frustration.

Workflow diagram · grounded in source
1
Analyze submitted shopper questions
trigger
“we dove into the details of the questions that shoppers submitted to our agents. We took all the property-for-sale-related questions that were submitted between December 2021 and April 2023 and visualized the data as word clouds.”
2
Pre-generate FAQ answers via LLM
ai_action
“We decided to pre-generate answers to the most frequently asked questions using available data instead of responding to inquiries in real-time. This approach not only minimized computation cost and latency, but also ensured the reliabili…”
3
Instant FAQ answer displayed
output
“our goal was to create a seamless, intuitive experience that provided instant, contextually-relevant answers”
4
Agent attributes computed from data
ai_action
“We computed these attributes based on the agent's deal history (based on their StreetEasy profile), the user's home preferences (inferred from their search activity), and data about the specific property they are browsing.”
5
Chain-of-prompt bio summary generation
ai_action
“we used LLM-based evaluation to score how well the summaries comply with the requirement, and human-in-the-loop review to rate the level of informativeness and creativity. Based on the evaluation, we adopted a chain of prompts, iterating…”
6
Human-in-the-loop review
human_review
“human-in-the-loop review to rate the level of informativeness and creativity”
7
Personalized agent intro displayed
output
“we display an AI-generated bio summary to highlight the agent's professional experience and qualifications as a successful buyer's agent”
Reported outcome

StreetEasy deployed two LLM-powered features: an instant, contextually-relevant property FAQ experience and 'Easy as PIE', a personalized agent introduction tool that won the best AI award at Zillow Hackweek 2024, designed to foster trust between shoppers and agents.

Reported metrics
Average shopper time on property pagesless than 61 seconds
Successful buyers who hire an agent9 in 10
Shoppers caring about agent expertise in building or neighborhood46%
Shoppers who like the agent personally37%
Show all 5 reported metrics
average shopper time on property pagesless than 61 seconds
successful buyers who hire an agent9 in 10
shoppers caring about agent expertise in building or neighborhood46%
shoppers who like the agent personally37%
shoppers caring about agent's past deal-making experience32%
Reported stack
large language models (LLMs)LLM-based evaluation
Source
https://www.zillow.com/tech/revolutionizing-the-real-estate-experience-with-llms-streeteasys-ai-journey/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

StreetEasy deployed two LLM-powered features: an instant, contextually-relevant property FAQ experience and 'Easy as PIE', a personalized agent introduction tool that won the best AI award at Zillow Hackweek 2024, des…

What tools did this team use?

large language models (LLMs), LLM-based evaluation.

What results were reported?

Average shopper time on property pages: less than 61 seconds; Successful buyers who hire an agent: 9 in 10; Shoppers caring about agent expertise in building or neighborhood: 46%; Shoppers who like the agent personally: 37% (source-reported, not independently verified).

What failed first in this deployment?

Traditional web-based FAQ systems depended on slow, inefficient search mechanisms that required users to navigate complex search engines and irrelevant results, causing frustration.

How is this customer support AI workflow structured?

Analyze submitted shopper questions → Pre-generate FAQ answers via LLM → Instant FAQ answer displayed → Agent attributes computed from data → Chain-of-prompt bio summary generation → Human-in-the-loop review → Personalized agent intro displayed.

Related customer support cases