Artificial intelligence in real estate: increased conversion with machine learning

Challenge: Optimise the customer journey of finding a new home by building a recommender system for properties and a segmentation system for customers based on artificial intelligence and machine learning.

For Co-libry we created a unique artificial intelligence that guides people to find their perfect home in no time. We developed an AI that shows users personalised search results, based on a combination of:

  • the similarity between properties (rule-based learning model)

  • the text descriptions of properties (natural language processing)

  • and clickstream analysis of the user (customer journey)

Who is Co-libry and what is their mission?
Co-libry is a start-up, based in Ghent, with a mission to help and assist the customers in their search for the perfect home. The home-buying process can be a drawn-out affair, time to use some technology innovation to transform this into the joyful life-event that it should be

The search mostly happens online, yet the Belgian real estate industry seems to be lacking in innovative solutions to support the consumer in finding the perfect home. We used machine learning to make the AI-driven decision-making process of the house buyer faster and easier

“The data scientists of Vectr have the expertise and know-how to deploy the artificial intelligence, including the architecture and structure we need to build.”

— Wendy Geeraerts, founder Co-libry

How data science changes the house-hunting struggle

Data-science techniques change the house-hunting struggle into an opportunity by finding the perfect house that matches the needs and wishes of the customer.

By utilizing data science techniques like artificial intelligence and natural language processing, the search results expand and the user enjoys more relevant results.

How? Our algorithm computes similarities between properties, based on characteristics such as price, square meter, location. Thanks to a supervised machine learning algorithm, houses and apartments can now be linked to each other and recommendations of similar properties can be given the user

Result of the rule based AI engine in the dataset of the premises. Search results are exact matched and properties that are linked to the exact match, independent from the specific original search.

Result Of The Rule Based AI Engine In The Dataset Of The Premises. Search Results Are Exact Matched And Properties That Are Linked To The Exact Match, Independent From The Specific Original Search.

Besides computing similarity, we also applied natural language processing

Unique in the Belgian real estate sector is our natural language processing on the descriptions of the properties: “A traditional search only includes results based on basic characteristics (price, square meter, and location) falling within the set criteria. Thanks to the NLP, the descriptions of the houses are also taken into consideration when presenting search results and recommendations.” Ignaz Wanders, project lead.

“The utilization of natural language processing of the descriptive texts of the listings in the datasets ensures a relevant and inspiring search result for the user. It allows the user to discover houses or apartments that regular search results wouldn’t have shown.” Wendy Geeraert, founder of Co-libry

The AI recommending premises

The AI Recommending Premises

Customer-centric AI approach: clickstream analysis on the customer journey

As no customer journey is the same these days, we need to understand the behaviour of the users, and tailoring interactions to specific segments of users is crucial.

Therefore, Vectr.Consulting tested the possibilities of clickstream analysis techniques within the real estate context. In production phase this would imply collecting, analysing and reporting aggregate data about which page and in which order a user navigates the website.

The ultimate goal is to combine clickstreams with users’ search histories. Creating opportunities for content personalisation, content targeting, targeted advertising segmentation, and profiling. This way, users will see unique, relevant content that is engaging and helpful in the context of that particular step of the customer journey.

Artificial intelligence on the customer journey of the user enables user profiling and segmenting.

Artificial Intelligence On The Customer Journey Of The User Enables User Profiling And Segmenting.

For example: Is the user an expert house hunter? Someone who needs guidance? A user who needs financial guidance? Is the user still in the ideation phase or is he already considering? The content will vary according to these features and will guide the user during his quest.

Targeted content can be:

  • a blog item

  • a newsletter

  • contact an expert

  • a calculator

  • a property

  • ads

Recap: the machine learning powered search ranking

Our AI shows users personalised search results, based on a combination of:

  • the similarity between properties (rule-based learning model)

  • the text descriptions of properties (natural language processing)

  • and clickstream analysis of the user (customer journey)

Co-libry is now a partner for those who are searching for their next home

Result: Recommendations have been optimised by using NLP techniques on free-text property descriptions, giving higher accuracies than recommenders based on metadata only. Segmentation allows for targeted advertisements, increasing paid click-through rates.

the AI engine for Co-libry: recommender system for properties and a segmentation for customers

The AI Engine For Co-Libry: Recommender System For Properties And A Segmentation For Customers