JLL: AI Referral Attribution in Real Estate for 2026

Artificial intelligence is changing how commercial real estate firms figure out where their business actually comes from. For a long time, especially with complex referral networks, knowing the true origin of a deal has been a guessing game. For firms like JLL, AI referral attribution isn’t about counting clicks. It’s about untangling the messy web of human connections and digital breadcrumbs that lead to a closed deal. The real challenge is getting sophisticated AI models to correctly identify the source of a referral when there are so many different interactions along the way.

1. Establish a Centralized Data Repository for All Referral Information

For AI referral attribution to have any chance of working, it needs a strong, centralized data infrastructure. If you don’t have one single source of truth, your AI model is going to choke on fragmented and inconsistent data. For a real estate firm, this means pulling everything together from your CRM, marketing automation software, email servers, and even old-school paper referral forms you’ve scanned. I’ve watched plenty of firms try to do this with data living in a dozen different spreadsheets and ancient databases, and the results are always a mess.

Think about it. A firm might be using Salesforce for its clients, HubSpot for marketing, and some custom-built thing for broker-to-broker referrals. The first real job is to build APIs or connectors to get all of that into one data warehouse, probably something cloud-based like Amazon Redshift or Google BigQuery. Make sure every single piece of data has a clear timestamp and a unique ID for the client or property. This sounds basic, but this is where projects go wrong. You absolutely have to standardize your naming conventions. If a referral is called “Broker Network” in one system and “External Partner” in another, your attribution model is going to get confused and produce garbage.

Pro Tip: Before you let any AI model near your data, you need to run a serious data cleansing operation. Get rid of duplicates, fill in missing information where you can (maybe with rule-based guesses for known referral types), and standardize all your formats. Dirty data will always give you biased or wrong attribution, and it doesn’t matter how fancy your AI is.

2. Implement Natural Language Processing (NLP) for Unstructured Data

In commercial real estate, referrals rarely show up as clean, structured data entries. They’re buried in emails, call notes, and casual conversations. For AI referral attribution, this is exactly where Natural Language Processing (NLP) is essential. Your old-school attribution models are definitely missing these unstructured data points.

Let’s say a JLL broker gets an email from an old client pointing them to a new prospect. The email might say something like “John Doe from XYZ Corp mentioned you to me,” or “I was talking to Sarah Smith, and she suggested I reach out.” An NLP engine, like the ones from Google Cloud Natural Language AI or AWS Comprehend, can be trained to spot the key people and companies and understand the relationships (like “referral” or “introduction”) inside that text. The system reads these messages, pulls out the potential referral sources, and ties them to the right client record in your central database.

For instance, if a broker types “Met with client, referral came from Jane Doe at ABC Realty” into a CRM activity log, the NLP model can flag “Jane Doe” and “ABC Realty” as the source. This gives you a much richer picture of your referral network than just relying on whatever someone types into a dropdown menu. It augments what your people are already doing by automatically pulling out the important context.

Common Mistake: Thinking you can just use a generic NLP model without any specific training. Real estate has its own language and referral habits. A generic model won’t understand industry jargon or abbreviations. You have to invest time in training your NLP models on your firm’s own historical emails and notes to get the accuracy you need.

3. Select and Configure a Multi-Touch Attribution Model

Simple “first-touch” or “last-touch” attribution models are completely outdated for high-value commercial real estate deals. Modern AI referral attribution requires a multi-touch approach that can handle the long and complicated path a referral often takes. This is a job for AI, which can process tons of touchpoints and assign fractional credit way better than any simple rule-based system.

A few advanced models are good for this. Shapley Value attribution, which comes from game theory, is really effective here. It looks at the marginal contribution of every single touchpoint, the first email, a follow-up call, a meeting, an ad they saw, to the final deal. It calculates this by running through every possible sequence of events to see how much each touchpoint added on average. This method distributes credit fairly, giving some to an interaction that didn’t start or close the deal but was still influential.

Another way to go is a Markov Chain model, which looks at the probability of a client moving from one state to the next, helping you see the most common and effective paths to a closed deal. You’d typically implement these models using Python libraries (like scikit-learn or Shap for Shapley values) or through marketing attribution platforms that can connect to your data warehouse.

When you’re setting this up, define “touchpoints” as broadly as possible. Include website visits and email opens, but also calls, meetings, and even offline things like industry conferences (as long as they’re logged). The more granular your data, the better your AI model will be. I’ve seen firms completely underestimate how much small interactions matter. A well-trained AI will find those hidden connections for you.

Pro Tip: Don’t just pick one model and call it a day. Try a few different ones (Shapley, Markov, Time Decay) and see what they tell you. The “best” model might change depending on the deal type or business line. You should A/B test your attribution models against actual business outcomes to make sure they’re actually working.

4. Integrate AI-Driven Insights into Operational Dashboards

Attribution data that just sits in a report is useless. The insights from your AI model need to be fed directly into the dashboards your brokers, marketers, and managers use every day. You’ve got to use dynamic, real-time visualizations instead of static reports.

Tools like Tableau, Microsoft Power BI, or Google Looker are perfect for this. You should build dashboards that show, at a glance:

• The top referral sources by deal value and sheer volume.
• The average time it takes to close a deal from different referral types.
• The “influence score” of different touchpoints, according to your multi-touch model.
• Geographic referral patterns (for example, you might see that referrals for Class A office space in Midtown Atlanta are almost always coming from one advisory firm in Buckhead).

A good dashboard might reveal that while direct broker intros make up 40% of your deal volume, the referrals that come from a specific industry event, while fewer, have a 25% higher average deal size and close 15% faster. This is the kind of insight that helps leadership reallocate resources effectively, maybe by putting more money into that event’s sponsorship. The point is to use data to inform strategic decisions, not just to generate reports on what already happened.

Common Mistake: Just dumping the raw AI model output onto a dashboard. Dashboards need to have clear, simple summaries and actionable recommendations. The goal is to make these insights usable for people who aren’t data scientists so they can make better decisions without needing a PhD in algorithms.

5. Continuously Monitor, Validate, and Retrain AI Models

AI models require constant attention. The real estate market changes, clients behave differently over time, and new referral channels pop up. So, you have to continuously monitor, validate, and retrain your AI referral attribution models if you want them to stay accurate.

You need to regularly check the AI’s attributed sources against sources confirmed by your brokers. Take a sample of transactions and have your team manually verify where the referral came from. If you see a big difference, you have to dig in. Is the AI missing a certain kind of unstructured data? Is there a new referral channel the model doesn’t know about yet? This human validation is what refines the model.

Retrain your models every so often (maybe quarterly) with all the new data you’ve collected. This makes sure the model learns from the latest market trends and referral behaviors. For example, if a new social media platform suddenly becomes a hot source for leads, your model needs to be retrained to give it the right amount of credit. You also have to watch for data drift, which is when your input data starts to look different than the data the model was trained on, causing its accuracy to drop.

You should also build in feedback loops. Let brokers give feedback on the attributed sources right in the CRM. Their notes can be incredibly valuable for spotting when the AI got it wrong, which lets you make targeted fixes to the NLP rules or model weights. This ongoing process of deployment, monitoring, feedback, and retraining defines a successful AI project.

The initial work to set up an AI referral attribution system is heavy, but it’s the ongoing maintenance and tuning that really deliver the value over the long haul. It’s a continuous improvement process.

AI-driven referral attribution is a big step up for commercial real estate firms. By building a solid data foundation, using advanced NLP and multi-touch models, and plugging the insights into daily work, firms can finally get real clarity on their referral networks. This clarity helps allocate resources better and build stronger, more profitable business relationships.