AI Assistant for Real Estate Brokerages: Give Agents a Second Route to an Answer
A large, independent Canadian real estate brokerage built an internal AI assistant on CustomGPT.ai for its REALTORS®, managers and staff. The assistant draws on more than 70,000 questions and answers its managers recorded over the past decade.
Once agents could ask it directly, monthly questions grew from about 170 to more than 2,200 within a few months. The brokerage shared its story anonymously.
One route to an answer, and it ran through a manager
A REALTOR® has a client waiting on an answer. A manager is fielding the same question again. A staff member needs to double check a legislative detail.
Before CustomGPT.ai, all three had one route to an answer. They went to a manager.
The brokerage is locally owned, has operated for more than 40 years and describes itself as one of Canada’s largest independent brokerages. It serves hundreds of REALTORS® and staff, and those agents depend on managers for guidance on practice, policy, procedure and legislation. That route worked. The brokerage wanted to give its agents a second one.
A second route, with the manager still on the other end of the phone
Managers are still fully available for support, and the brokerage wanted to keep it that way.
What it added was an option for agents who prefer to get an answer on demand and on their own schedule. The goal was to help agents move their careers and deals forward faster, whenever an AI answer was appropriate. An agent can now get an instant answer or still pick up the phone.
In the brokerage’s own words:
“It’s been a very helpful tool to keep up with the AI wave, offering a great option for agents to use as they need and want to, in a world where technology is changing fast.”- A Canadian real estate brokerage
Managers went first
The brokerage chose CustomGPT.ai because the support felt human and the customization options were broad and easy to understand. Then it took its time.
The assistant ran internally with managers for several months before any agent had access. The people who had been answering these questions for years were the ones who vetted how the assistant answered them. The team wanted a product it was proud to offer, and it waited until it had one.
Once the managers were satisfied, access opened to agents. That is when the numbers moved.
A decade of manager advice is the knowledge base
The content that matters most is the brokerage’s own history.
Managers have recorded more than 70,000 questions and answers over the past decade. Those answers cover practice, policy, procedure, legislation and regulatory guidance. They sit in the same assistant as the brokerage’s internal policies and forms and its professional development video library.
This is the capability that gives the assistant its value. A REALTOR® asking a question is, in effect, searching what the brokerage’s managers have already said, alongside the documents those answers rest on.
Five kinds of source in one assistant
The knowledge base combines five source types: recorded manager answers, website crawls, Google Drive documents, YouTube video transcripts and uploaded files. They work together in one assistant, without a separate tool for each.
Keeping it current is somebody’s job. A manager who curates the knowledge base adds new sources as things change. An in-house programmer sends weekly updates from the brokerage’s internal question and answer database, so new manager answers flow in. Many sources refresh on a schedule and some sync automatically.
Changes are tested in the CustomGPT.ai sandbox before they reach agents. The team is also removing outdated answers, so that older advice does not override newer documents, and it reviews conversation history and analytics regularly to spot weak answers and decide what to fix.
From about 170 questions a month to more than 2,200
Monthly question volume is the main measure the brokerage tracks. It rose sharply once agents could use the assistant directly, from about 170 questions a month to more than 2,200 within a few months. That is 13 times the volume.
The brokerage also reviews conversation data to judge answer quality. More agents check the assistant first with each passing week, and then call a manager to verify an answer when they want to. The on-demand route suits agents who want to double check legislation or find a quick fact they can rely on.
Trust is building, and use is rising every week.
A search bar and a compliance check on the same account
The first assistant was followed by a second use case. The brokerage extended the same CustomGPT.ai account to power a virtual office search bar, which makes the resources available to agents faster to find.
One assistant also uses the web search MCP connector to support Canadian anti-money laundering compliance. It runs adverse media checks, searching public news and web sources for negative coverage of a client. The team gets a quick first look before the manual review.
What this means for your brokerage
The advice your managers have already given is the knowledge base. This brokerage had a decade of recorded answers, and those answers now reach every agent who asks.
Let the people who know the answers go first. Managers here vetted the assistant for several months before agents saw it.
Keep the human route open. Agents choose the assistant when it fits and a manager when it does not, and more of them check the assistant first every week.
Make keeping it current a named job. One manager curates the sources, a programmer sends the weekly updates, and changes go through a sandbox before agents see them.
The REALTOR® with a client waiting can now get an instant answer. The manager has a second route that takes some of the repeat questions, and the staff member checking legislation has a reliable source on demand. Each of them can still call a person when the situation calls for one.
The brokerage is continuing to clean out outdated answers, and it plans to look at Plan & Act in more detail to see how it could help its team and agents.
Figures as reported by the brokerage. Shared anonymously. We publish a customer’s name only with their written agreement.
Related Resources:
- How a Canadian Real Estate Brokerage Grew from 170 to 2,200+ Monthly Questions: Read the full case study behind this post, with the at-a-glance table, the results and the PDF.
- How to Make a Real Estate Chatbot: See the client-facing side: a chatbot for buyers and sellers built on your listings and brokerage FAQs.
- What Is the Fastest Way to Onboard New Employees Using an Internal AI Knowledge Base?: See how an internal assistant answers the questions new agents and staff ask in their first weeks.
- How Do I Set Up an Internal AI Search for My Team?: See how to build the kind of internal search this brokerage added as its virtual office search bar.
- AI for Compliance: The Complete Enterprise Guide: See where an AI first look fits beside manual review, as in the brokerage’s anti-money laundering checks.