Tenant Churn Prediction
“Which tenants will not renew their lease?”
Book a demo and get a free trial of the full platform: research agent, fine-tune capabilities, and forward-deployed engineer support.
By submitting, you accept the Terms and Privacy Policy.
Loved by data scientists, ML engineers & CXOs at
A real-world example
Which tenants will not renew their lease?
Tenant turnover costs $3K-$8K per unit in vacancy loss, cleaning, repairs, and leasing commissions. For a REIT managing 50,000 units with 40% annual turnover, that is $60-160M per year in turnover costs. Predicting non-renewals 60-90 days ahead enables targeted retention offers that cost 10-20% of turnover expense. Traditional models use lease expiry date and payment history but miss the maintenance and satisfaction signals embedded in the tenant-property relationship.
Quick answer
Tenant churn prediction AI forecasts which tenants will not renew their lease 60-90 days before expiry, giving property managers enough lead time for targeted retention offers. By connecting payment patterns, maintenance request history, and building-level social dynamics into a graph, models detect the dissatisfaction signals that precede non-renewal. A REIT managing 50,000 units saves $30-60M annually by reducing turnover through targeted retention that costs 10-20% of the turnover expense.
Approaches compared
4 ways to solve this problem
1. Lease Expiry Tracking
Track lease expiration dates and send renewal offers to all expiring tenants at a fixed lead time (typically 60-90 days). The baseline approach used by most property management systems.
Best for
Ensuring no renewal window is missed. A necessary foundation that all property managers should have.
Watch out for
Treats all expiring tenants the same. A tenant who has already decided to leave gets the same generic offer as a tenant who needs a small incentive to stay. No prioritization means retention budgets are spread thin across tenants who cannot be retained and tenants who would have renewed anyway.
2. Payment History Analysis
Flag tenants whose payment timing has shifted (early payer becoming late payer) as churn risks. Payment behavior is a strong behavioral signal.
Best for
Detecting financial stress-driven churn and tenants who are mentally disengaging (deprioritizing rent payments).
Watch out for
Misses tenants who pay on time but are dissatisfied due to maintenance issues, neighbor problems, or better alternatives in the market. Payment history is one signal among many, and used alone it catches only 30-40% of non-renewals.
3. Survey-Based Satisfaction Scoring
Send periodic satisfaction surveys and use scores to predict churn risk. Direct measurement of tenant sentiment.
Best for
Identifying specific dissatisfaction drivers (maintenance responsiveness, amenity quality, noise) that can be addressed directly.
Watch out for
Response rates are typically 15-25%, creating severe selection bias. Tenants most likely to churn are least likely to respond to surveys. Also measures stated satisfaction, which lags actual behavior by weeks or months. By the time a tenant rates satisfaction as 3/10, they have already started apartment hunting.
4. Graph Neural Networks (Kumo's Approach)
Connect tenants, leases, payments, maintenance requests, and properties into a property management graph. GNNs learn churn patterns from the full tenant-property network, including building-level dynamics and maintenance satisfaction signals.
Best for
Large portfolios where churn patterns are relational: maintenance response times affect building-level satisfaction, neighbor departures trigger consideration, and market conditions affect different tenant segments differently.
Watch out for
Requires integrated data across property management, maintenance, and payment systems. Best value for portfolios with 5,000+ units where the network provides meaningful signal.
Key metric: Graph-based churn models predict non-renewals with 75-80% accuracy at 60-90 days lead time. Targeted retention offers cost 10-20% of the $3K-$8K turnover cost per unit, delivering 5-10x ROI on retention spend.
Why relational data changes the answer
Tenant churn is contagious. When 3 tenants leave a building in 90 days, remaining tenants notice. They see moving trucks, experience construction noise from unit turns, and wonder if they are missing something. The building-level departure rate is a stronger predictor of individual churn than any single tenant's own payment or satisfaction metrics. Similarly, maintenance responsiveness creates building-level satisfaction signals: when Tenant TEN002 has two unresolved maintenance requests with an average resolution time of 10 days (vs. the portfolio average of 3 days), that dissatisfaction affects not just TEN002 but their neighbors who hear about it.
Flat churn models see each tenant as an independent prediction problem. Graph-based models see the tenant-property-building network and learn that TEN002's 76% non-renewal probability comes from the compound effect of unresolved maintenance (attribution 0.30), late payment trend (0.24), short tenure (0.19), and the recent departure of similar tenants in the building (0.16). SAP's SALT benchmark shows 91% accuracy for graph models vs 63% for gradient-boosted trees. RelBench confirms at 76.71 vs 62.44. For tenant churn, this means predicting non-renewals with 75-80% accuracy at 60-90 days lead time, when retention offers still work. At 30 days, the tenant has already signed a new lease and no offer will retain them.
Predicting tenant churn from individual tenant data is like predicting employee turnover by looking only at each employee's performance review. You would miss that three people on their team just quit, their manager is unresponsive to concerns, and a competitor is hiring across the street at 20% higher salary. Tenant churn works the same way: the decision to leave is shaped by the building community, management responsiveness, and market alternatives, not just the individual tenant's payment history.
How KumoRFM solves this
Graph-powered intelligence for real estate
Kumo connects tenants, leases, payments, maintenance requests, and properties into a property management graph. The GNN learns churn patterns from the tenant network: when maintenance response times deteriorate, when similar tenants in the building leave, and when payment patterns shift from early to late. PQL predicts non-renewal probability per tenant with 60-90 days of lead time, enabling retention offers before tenants start searching for alternatives.
From data to predictions
See the full pipeline in action
Connect your tables, write a PQL query, and get predictions with built-in explainability — all in minutes, not months.
Your data
The relational tables Kumo learns from
TENANTS
| tenant_id | name | unit_type | tenure_months |
|---|---|---|---|
| TEN001 | J. Martinez | 2BR | 24 |
| TEN002 | K. Patel | 1BR | 8 |
| TEN003 | L. Johnson | 3BR | 36 |
LEASES
| lease_id | tenant_id | monthly_rent | expiry_date | renewal_offered |
|---|---|---|---|---|
| LSE101 | TEN001 | $1,850 | 2025-05-31 | Yes |
| LSE102 | TEN002 | $1,400 | 2025-06-30 | Yes |
| LSE103 | TEN003 | $2,200 | 2025-04-30 | Yes |
PAYMENTS
| payment_id | tenant_id | amount | days_late | timestamp |
|---|---|---|---|---|
| PAY401 | TEN001 | $1,850 | 0 | 2025-03-01 |
| PAY402 | TEN002 | $1,400 | 5 | 2025-03-06 |
| PAY403 | TEN003 | $2,200 | 0 | 2025-03-01 |
MAINTENANCE_REQUESTS
| request_id | tenant_id | type | resolution_days | date |
|---|---|---|---|---|
| MR501 | TEN001 | HVAC repair | 2 | 2025-01-15 |
| MR502 | TEN002 | Plumbing leak | 8 | 2025-02-10 |
| MR503 | TEN002 | Pest control | 12 | 2025-02-25 |
PROPERTIES
| property_id | name | units | occupancy_rate | avg_rent |
|---|---|---|---|---|
| PRO01 | Oak Park Apartments | 200 | 94% | $1,750 |
| PRO02 | Downtown Lofts | 120 | 88% | $1,650 |
Write your PQL query
Describe what to predict in 2–3 lines — Kumo handles the rest
PREDICT BOOL(LEASES.renewed = False, 0, 90, days) FOR EACH TENANTS.tenant_id WHERE LEASES.expiry_date < '2025-07-01'
Prediction output
Every entity gets a score, updated continuously
| TENANT_ID | UNIT | LEASE_EXPIRY | NON_RENEWAL_PROB | RISK_TIER |
|---|---|---|---|---|
| TEN001 | 2BR | 2025-05-31 | 0.12 | Low |
| TEN002 | 1BR | 2025-06-30 | 0.76 | Critical |
| TEN003 | 3BR | 2025-04-30 | 0.08 | Low |
Understand why
Every prediction includes feature attributions — no black boxes
Tenant TEN002 -- 1BR unit, 8-month tenure
Predicted: 76% non-renewal probability (Critical)
Top contributing features
Unresolved maintenance requests
2 open, avg 10 days
30% attribution
Late payment trend (last 3 months)
Increasing
24% attribution
Short tenure (less than 12 months)
8 months
19% attribution
Similar tenants in building left
3 in last 90 days
16% attribution
Rent-to-market ratio unfavorable
105% of market
11% attribution
Feature attributions are computed automatically for every prediction. No separate tooling required. Learn more about Kumo explainability
PQL Documentation
Learn the Predictive Query Language — SQL-like syntax for defining any prediction task in 2–3 lines.
Python SDK
Integrate Kumo predictions into your pipelines. Train, evaluate, and deploy models programmatically.
Explainability Docs
Understand feature attributions, model evaluation metrics, and how to build trust with stakeholders.
Frequently asked questions
Common questions about tenant churn prediction
How much does tenant turnover really cost?
The fully loaded cost of tenant turnover is $3K-$8K per unit, including: vacancy loss (average 30-45 days at $50-75/day = $1,500-$3,400), unit turn costs (painting, cleaning, repairs = $500-$2,000), leasing costs (marketing, agent commissions = $500-$1,500), and administrative costs ($200-$500). For a REIT with 50,000 units and 40% annual turnover, that is $60-160M per year. Reducing turnover by even 5 percentage points saves $7.5-20M.
What retention offers work best for at-risk tenants?
The most effective offers match the churn driver. For price-sensitive tenants: rent concessions or lease term discounts ($50-100/month costs $600-1,200/year vs $3-8K turnover cost). For maintenance-dissatisfied tenants: priority maintenance resolution and unit upgrades. For market-driven churn: matching competitive offers or adding amenities. The key insight is that the cheapest retention offer is addressing the specific dissatisfaction driver, not blanket rent reductions.
When is the best time to make a retention offer?
60-90 days before lease expiry is the sweet spot. Earlier than 90 days and the tenant has not yet started thinking about renewal. Later than 60 days and they may have already started apartment hunting or signed elsewhere. Graph-based models provide early warning at 90+ days, giving property managers time to address maintenance issues before making a formal retention offer at 60 days.
Can churn prediction AI work for commercial leases?
Yes, but the signals differ. Commercial tenant churn is driven by business performance (revenue trends, employee headcount changes), lease economics (rent vs. market rates, escalation clauses), and space utilization (subletting activity, reduced occupancy). Graph-based models connect tenant financials, market data, and building-level dynamics. Commercial prediction typically needs 12-18 months of lead time given longer lease terms and the complexity of commercial relocations.
How does tenant churn prediction integrate with property management software?
The model outputs a churn probability score per tenant that integrates into existing property management workflows. Most implementations create a weekly priority list of at-risk tenants for property managers to review, with recommended actions based on the predicted churn driver. Integration with major PMS platforms (Yardi, RealPage, AppFolio) typically takes 4-6 weeks. The model updates daily as new payment and maintenance data arrives.
Bottom line: A REIT managing 50,000 units saves $30-60M annually by reducing turnover through targeted retention. Kumo's property graph detects the maintenance-satisfaction-payment patterns that predict non-renewal 60-90 days ahead, when retention offers still work.
Related use cases
Explore more real estate use cases
Topics covered
One Platform. One Model. Infinite Predictions.
KumoRFM
Relational Foundation Model
Turn structured relational data into predictions in seconds. KumoRFM delivers zero-shot predictions that rival months of traditional data science. No training, feature engineering, or infrastructure required. Just connect your data and start predicting.
For critical use cases, fine-tune KumoRFM on your data using the Kumo platform and Research Agent for 30%+ higher accuracy than traditional models.
Book a demo and get a free trial of the full platform: research agent, fine-tune capabilities, and forward-deployed engineer support.
