JLL
industry genAI assistant built on a proprietary vertical LLM
JLL GPT, launched in August 2023, is a vertical LLM for commercial real estate trained on JLL's internal transaction data; more than 11,000 employees used it within 48 hours and more than 47,000 professionals in 2024.
Objective
Give commercial brokerage teams a genAI assistant that meets JLL's security and confidentiality requirements, to research deals, model, and produce client deliverables faster on proprietary data.
The deployment
JLL GPT was launched on August 1, 2023 by the JLL Technologies division, presented as the first large language model built specifically for commercial real estate. It is trained on decades of JLL's internal transaction data supplemented by external sources, and deployed in a secure computing environment. Within 48 hours of opening, more than 11,000 employees out of a workforce of about 103,000 were using it. In October 2024, at the launch of the JLL Falcon platform, more than 47,000 professionals had already used it. JLL GPT had by then gained image understanding and twenty-five times more working memory than its initial version. Falcon adds access to several LLMs, multimodal models, and more than sixty AI functions by line of business. Brokers use it to research deals, model cash flows, and prepare client deliverables.
Results Proof C
Official JLL releases (T1) and established press (CIO Dive) consistent, with adoption figures at global scale. No isolated financial impact published, hence a level C rather than A.
How it works
Documented architectureThe stack in detail
- llm JLL GPT proprietary vertical LLM trained on decades of internal CRE transaction data, enriched with image understanding in 2024
- infra Microsoft Azure secure computing environment in which the model is deployed
- outil Couche RAG sur donnees CRE gouvernees retrieval from JLL's internal data before generation, key to the industry advantage vs a public LLM
- plateforme JLL Falcon platform launched at the end of 2024: access to several LLMs, multimodal models, and more than sixty AI functions by line of business
How it runs, concretely
For ops teams-
1Natural-language question human
The broker asks a question about a market, an asset, or a model.
-
2Retrieval on governed data AI
The system retrieves information from JLL's internal CRE data before answering.
-
3Deliverable generation AI
JLL GPT produces an answer, a synthesis, or a draft document, with image understanding since 2024.
-
4Human validation human
The broker checks and uses the output in their client work; the AI does not replace expertise.
The corpus of proprietary CRE data, cleaned and governed. Without it, the model answers generically and loses its industry advantage against a public LLM.
How your customers perceive this type of use
Sourced studiesLes consommateurs n'acceptent pas les chatbots par defaut : 64% prefereraient que les entreprises n'utilisent pas d'IA dans leur service client (Gartner, 2024) et pres d'un utilisateur sur cinq du service client par IA n'en retire aucun benefice (Qualtrics, 2025). L'acceptation se construit sur trois conditions mesurees par Salesforce : savoir qu'on parle a une IA, pouvoir escalader vers un humain, comprendre la logique de l'agent.
Acceptance conditions
- Etre informe qu'on parle a une IA et non a un humain (pres de 75% le demandent, Salesforce 2024)
- Un chemin d'escalade clair vers un agent humain (45% plus enclins a utiliser l'agent IA, Salesforce 2024)
- Une logique de l'agent clairement expliquee (44% plus enclins, Salesforce 2024)
Red lines
- Rendre l'humain injoignable : c'est la premiere inquietude des consommateurs sur l'IA dans le service client (Gartner 2024) et 50% craignent que l'IA les coupe du contact humain (Qualtrics 2025)
- Remplacer le service client par l'IA sans alternative : 53% envisageraient de partir chez un concurrent (Gartner 2024)
Sources: Salesforce 2024 · Gartner 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- large, structured proprietary industry corpus
- data governance and quality
- clear usage rights on client data
Org prerequisites
- dedicated ML and security team
- compliant computing environment
- change management with the business lines
Possible stack
- LLM on a secure cloud
- RAG layer over the internal document base
- compliance and confidentiality controls
The plan, step by step
- Step 1Build and govern the proprietary industry corpus (cleaning, usage rights, quality)Deliverable: Documented corpus with valid rights and identified owners
- Step 2Stand up the secure computing environment (access, logs, confidentiality) with the security teamDeliverable: Compliant cloud infra validated by security
- Step 3Connect the LLM and the RAG layer to the corpus, on one pilot business lineDeliverable: Assistant in alpha with a reference set of industry questions
- Step 4Pilot with a group of business users, measure answer quality and adoptionDeliverable: Quality feedback + usage rate of the pilot group
- Step 5Deploy broadly with change management and usage trackingDeliverable: Assistant open to the workforce, adoption tracked continuously
First step: Build and govern a clean industry corpus before connecting an LLM to it.
The same pattern, proven in other industries
Alaska Airlines
natural language search for travel inspiration
Etsy
agentic commerce (in-chat checkout through an open protocol)
Expedia
distribution through a third-party conversational assistant (app in ChatGPT)
Sources
- S1 JLL unveils first GPT model for commercial real estate Primary archive pending
- S2 JLL rolls out proprietary generative AI model to internal employees Established press archive pending
- S3 JLL Falcon kicks off new era of AI-powered CRE innovation Primary archive pending
- S4 How JLL built an AI platform to help employees and clients close the deal Secondary archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.