Cross-Sell Optimization
“Which policyholders should receive a bundling offer?”
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 policyholders should receive a bundling offer?
Multi-line policyholders have 90% retention rates vs 70% for single-line (J.D. Power), yet only 35-40% of personal-lines customers bundle. Each additional line adds $800-$1,500 in annual premium and reduces churn risk by 15-20 percentage points. Insurers spend $50-100 per outbound sales contact, but untargeted campaigns convert at just 2-4%. The signals for bundling readiness are spread across policy records, claims history, life events, service interactions, and competitive pricing data. A targeted approach could double conversion rates while halving contact volume.
Quick answer
AI optimizes insurance cross-selling by connecting policyholder coverage portfolios with life-event signals, quote activity, discount eligibility, and competitive market context. Instead of blasting untargeted bundle offers to all single-line customers, graph-based models identify the specific policyholders who are actively shopping for additional coverage, rank the right product to offer, and time the outreach to maximize conversion. Targeted cross-sell campaigns convert at 8-12% versus 2-4% for untargeted approaches.
Approaches compared
4 ways to solve this problem
1. Untargeted Campaign Blasts
Send bundling offers to all single-line policyholders at renewal. Simple to execute using existing marketing systems. High volume, low conversion.
Best for
Building awareness of multi-line discounts across the entire book when you have no targeting capability.
Watch out for
2-4% conversion rates at $50-100 per contact mean you spend $25-50 to acquire each additional line. Most contacts are wasted on policyholders who are not in the market.
2. Propensity Scoring with Logistic Regression
Score policyholders based on demographics, tenure, and current coverage. Target the highest-propensity segment with bundling offers. A meaningful step up from blast campaigns.
Best for
Moderate targeting improvement when you have clean policyholder demographics and can segment by life stage.
Watch out for
Misses real-time shopping signals, life-event triggers, and competitive context. A 35-year-old auto-only customer might have high demographic propensity but zero intent right now. You need timing signals, not just demographic fit.
3. Collaborative Filtering (Recommendation Engine)
Borrow techniques from e-commerce: 'customers like you also bought home insurance.' Uses policy-purchase patterns to identify likely next products.
Best for
Large books with diverse product portfolios where purchase-pattern data is rich enough to find meaningful similarities.
Watch out for
Insurance is not e-commerce. Purchase decisions are infrequent (annual) and driven by life events, not browsing behavior. The cold-start problem is severe: new policyholders have no purchase history to match against.
4. Relational Deep Learning (Kumo's Approach)
Connects policyholders to coverage portfolios, life-event signals (new vehicle purchase, home closing), quote activity (competitor shopping), discount eligibility, and service interactions in a single graph. Predicts both propensity and optimal product in one model.
Best for
Identifying high-intent cross-sell moments: a home-only customer who just bought a car, a renter who just closed on a house, an auto-only customer shopping for home quotes. Timing the right offer to the right customer.
Watch out for
Depends on access to life-event data sources (DMV records, property records, digital signals). Without intent signals, the model falls back to demographic propensity, which is less precise.
Key metric: Multi-line policyholders retain at 90% vs. 70% for single-line (J.D. Power). Targeted AI cross-sell converts at 8-12% vs. 2-4% untargeted, generating $80-150M in incremental annual premium.
Why relational data changes the answer
Flat cross-sell models score each policyholder based on their own attributes: age, tenure, current coverage, premium level. They can predict that 35-year-old homeowners are statistically likely to add auto insurance. But they cannot see that Jennifer Adams just purchased a new vehicle (DMV record match), has been requesting auto insurance quotes from two competitors (digital signal), qualifies for a 15% multi-line discount she does not know about, and has a low loss ratio that makes her a highly profitable customer to retain. These signals come from four different data sources, and their combination makes Jennifer a 81% likely converter right now, not just a generic 'someday might bundle' prospect.
Relational learning connects these signals in real time. The model walks from policyholder to life-event signals, to quote activity, to discount eligibility, to risk profile. It learns that the combination of life-event trigger + active shopping + available discount + strong risk profile predicts bundling conversion at 4-5x the rate of demographic propensity alone. More importantly, it identifies the specific product to offer (auto + umbrella for Jennifer, not just auto) and the specific discount structure that maximizes both conversion and lifetime value. This precision turns cross-sell from a numbers game into a relationship-building conversation.
Cross-selling from a flat policyholder table is like a waiter recommending dessert to everyone at every table. Some people want it, most do not, and you annoy the ones who are watching their diet. Relational cross-selling is like a waiter who noticed you looked at the dessert menu, knows you ordered a light main course, and remembers you had the chocolate cake last time you were here. The recommendation is personal, timely, and welcome.
How KumoRFM solves this
Relational intelligence built for insurance data
Kumo connects policyholders to their coverage portfolio, life-event signals, billing patterns, service interactions, and market context. The model identifies that Policyholder PH-6601 (home-only) just purchased a new vehicle (DMV record match), has been searching for auto insurance quotes (digital signals), and has a low-loss-ratio profile that would qualify for a significant multi-policy discount. These cross-table signals produce a bundling-propensity score, ranking the right offer (auto + umbrella) for each household.
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
POLICYHOLDERS
| policyholder_id | name | current_lines | total_premium | tenure_years |
|---|---|---|---|---|
| PH-6601 | Jennifer Adams | Home Only | $2,100 | 6.4 |
| PH-6604 | Carlos Reyes | Auto Only | $1,400 | 3.2 |
| PH-6603 | Diana Lee | Home + Auto + Umbrella | $5,400 | 11.2 |
LIFE_EVENT_SIGNALS
| policyholder_id | event | confidence | detected_date |
|---|---|---|---|
| PH-6601 | New Vehicle Purchase | High | 2025-09-05 |
| PH-6604 | Home Purchase | Medium | 2025-09-12 |
| PH-6603 | None Detected | N/A | N/A |
QUOTE_ACTIVITY
| policyholder_id | line_quoted | competitor_quotes | last_quote_date |
|---|---|---|---|
| PH-6601 | Auto | 2 | 2025-09-10 |
| PH-6604 | Home | 1 | 2025-09-14 |
| PH-6603 | None | 0 | N/A |
DISCOUNT_ELIGIBILITY
| policyholder_id | multi_line_discount | loyalty_discount | claims_free_discount |
|---|---|---|---|
| PH-6601 | 15% if adds auto | 5% (5+ years) | 10% (0 claims) |
| PH-6604 | 12% if adds home | None | 5% (1 small claim) |
| PH-6603 | Already applied | 8% (10+ years) | 10% (0 claims) |
Write your PQL query
Describe what to predict in 2–3 lines — Kumo handles the rest
PREDICT BOOL(POLICYHOLDERS.LINES_ADDED > 0, 0, 90, days) FOR EACH POLICYHOLDERS.POLICYHOLDER_ID WHERE POLICYHOLDERS.CURRENT_LINES != 'Home + Auto + Umbrella'
Prediction output
Every entity gets a score, updated continuously
| POLICYHOLDER_ID | CURRENT_LINES | BUNDLE_PROPENSITY | RECOMMENDED_LINE | EST_PREMIUM_ADD |
|---|---|---|---|---|
| PH-6601 | Home Only | 0.81 | Auto + Umbrella | +$2,200 |
| PH-6604 | Auto Only | 0.54 | Home | +$1,800 |
| PH-6603 | H+A+U (full) | N/A | Already Bundled | $0 |
Understand why
Every prediction includes feature attributions — no black boxes
Policyholder PH-6601 (Jennifer Adams, Home Only)
Predicted: 81% bundling propensity (Auto + Umbrella)
Top contributing features
New vehicle purchase detected
Sept 2025
30% attribution
Active auto insurance shopping
2 competitor quotes
26% attribution
Multi-line discount opportunity
15% savings
19% attribution
Strong retention profile (low loss ratio)
0.28
14% attribution
Tenure and loyalty discount eligible
6.4 yrs, 5%
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 cross-sell optimization
How does AI improve insurance cross-sell conversion rates?
AI improves cross-sell conversion by targeting policyholders with active intent signals rather than demographic propensity alone. Graph-based models identify customers who are actively shopping for additional coverage (life events, competitor quotes) and match them with the right product offer at the right time. This doubles conversion rates from 2-4% to 8-12%.
What is the retention benefit of insurance policy bundling?
Multi-line policyholders retain at 90% versus 70% for single-line customers (J.D. Power). Each additional line adds $800-$1,500 in annual premium and reduces churn risk by 15-20 percentage points. The combined effect makes cross-sell the highest-ROI retention strategy in personal lines insurance.
How do insurers identify the best cross-sell opportunities?
The best opportunities combine three signals: a life event that creates a coverage need (new car, new home, new baby), active shopping behavior (competitor quotes, coverage research), and available discounts that make bundling attractive. Graph-based models identify all three signals from connected data sources and prioritize the highest-probability opportunities for outreach.
What data do insurers need for AI-powered cross-sell?
Effective cross-sell models need policyholder coverage data, life-event signals (DMV records, property records, digital activity), quote activity logs, discount eligibility rules, and service-interaction history. The more data sources connected, the more precise the targeting. Even with just coverage data and rate-change history, models outperform untargeted campaigns by 2-3x.
Bottom line: Increase multi-policy households by 15-25% and reduce churn by 15-20 points per converted household, generating $80-150M in incremental annual premium for a top-10 personal-lines insurer.
Related use cases
Explore more insurance 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.
