Pricing Optimization
“What premium maximizes retention and profitability?”
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
What premium maximizes retention and profitability?
Insurers face a constant tension: price too high and you lose customers to competitors; price too low and you write unprofitable business. Traditional actuarial models set prices using loss-cost models with broad territory and classification factors, but miss individual-level price sensitivity. A policyholder with $5,400 in premium and a 0.15 loss ratio will tolerate a 5% increase but leave at 10%. Meanwhile, a policyholder with $1,800 in premium and a 0.65 loss ratio is price-insensitive because competitors will not offer a better rate. Getting this wrong costs top-20 insurers $200-500M annually in either lost profitable customers or underpriced risky ones.
Quick answer
AI optimizes insurance premiums at the individual policyholder level by connecting risk profiles, claims history, competitive market rates, billing behavior, and renewal outcomes into a relational graph. Traditional actuarial models set prices by broad territory and classification factors, missing individual price sensitivity. Graph-based models predict the exact rate increase each policyholder will tolerate before switching, optimizing the portfolio for both retention and profitability. The result is 2-4 points of combined ratio improvement.
Approaches compared
4 ways to solve this problem
1. Traditional Actuarial Pricing (GLMs)
Generalized linear models with multiplicative rating factors: territory, vehicle type, driver age, credit tier. The regulatory standard. Prices are set by segment, not by individual policyholder.
Best for
Rate filings and regulatory compliance. GLMs produce the transparent factor structures that state DOIs require.
Watch out for
Segment-level pricing misses individual elasticity. Two policyholders in the same segment may have completely different price sensitivity based on competitive alternatives, tenure, and claims experience. One-size-fits-segment leaves money on the table.
2. Competitive Rate Monitoring
Track competitor prices in each market and adjust rates to maintain competitive positioning. Uses rate-comparison tools, quote aggregators, and market intelligence feeds.
Best for
Staying competitive in price-sensitive markets where rate position drives growth or retention.
Watch out for
Reacting to competitor prices is backward-looking. By the time you see a competitor's rate filing, they have already captured your policyholders. You also cannot see which specific policyholders are being targeted by competitor campaigns.
3. Price Elasticity Models (Demand Curves)
Econometric models that estimate price sensitivity by segment using historical renewal data at different rate-change levels. Predicts what percentage of a segment will renew at each price point.
Best for
Setting rate-increase caps by segment. Useful for portfolio-level decisions like 'do not increase any segment by more than 8%.'
Watch out for
Segment-level elasticity masks individual variation. A segment average says 85% renew at +8%, but that average hides the profitable customers who leave at +6% and the unprofitable ones who stay at +12%. Individual-level optimization requires individual-level predictions.
4. Relational Deep Learning (Kumo's Approach)
Connects each policyholder to their risk profile, claims history, competitive alternatives, billing behavior, service interactions, and renewal history in a relational graph. Predicts individual-level price elasticity and optimizes across the portfolio to hit combined-ratio targets.
Best for
Individual-level pricing that retains profitable customers with smaller increases and recovers margin on price-insensitive customers with larger increases. Optimizes the full portfolio simultaneously.
Watch out for
Regulatory acceptance varies. The model's output typically feeds into a GLM-compatible rate structure for filing purposes, adding a step between model prediction and final price.
Key metric: Individual-level pricing optimization improves combined ratios by 2-4 points while retaining 92%+ of profitable policyholders, generating $200-500M in annual value for a top-20 insurer.
Why relational data changes the answer
Flat pricing models treat each policyholder as an independent data point with static attributes: credit tier, territory factor, vehicle symbol. They estimate that policyholders with A-tier credit in territory 5 will tolerate a 7% increase with 87% retention. But within that segment, Jennifer Adams has a 0.28 loss ratio, three active competitor quotes in her zip code, and just absorbed a 12% increase last year. She will leave at 5%. Meanwhile, Mark Stevens in the same segment has a 0.65 loss ratio, no competitive alternatives, and renewed after a 4.7% increase without complaint. He will stay at 10%. The flat model sees them as identical. The relational model sees their completely different contexts.
Relational learning connects each policyholder to their full decision context: prior rate increases (cumulative fatigue), competitive alternatives in their specific market (not just segment averages), service-interaction sentiment (are they happy or frustrated?), bundling status (are they sticky or at risk?), and claims history (are competitors interested in poaching them?). The model learns that compound rate increases above 15% cumulative over two years trigger defection in low-loss-ratio customers at 3x the rate of single-year increases. It learns that policyholders in zip codes with active competitor conquest campaigns need 2-3 points less increase than the same risk profile in uncontested markets. These relational patterns turn pricing from a segment-level approximation into an individual-level optimization.
Setting insurance prices from a flat rating table is like a hotel charging the same room rate to everyone regardless of demand. Business travelers booking last-minute during a conference will pay full rate. Leisure travelers with six competitors on the same block will book the cheapest option. Revenue management in hotels solved this decades ago with dynamic, individual-level pricing. Relational pricing does the same thing for insurance: different prices for different contexts, optimized across the full portfolio.
How KumoRFM solves this
Relational intelligence built for insurance data
Kumo connects policyholders to their risk profiles, claims history, competitive market data, billing behavior, service interactions, and renewal outcomes. The model predicts the price elasticity for each policyholder: PH-6601 has high elasticity (competitive market, low loss ratio, recent rate increase) and should receive only a 3% increase, while PH-6602 has low elasticity (limited alternatives, higher risk profile) and can absorb 8%. The model optimizes across the portfolio to hit combined-ratio targets while maximizing retention of profitable customers.
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 | premium | loss_ratio | tenure_years | lines |
|---|---|---|---|---|
| PH-6601 | $3,200 | 0.28 | 6.4 | Home + Auto |
| PH-6602 | $1,800 | 0.65 | 2.1 | Auto Only |
| PH-6603 | $5,400 | 0.15 | 11.2 | Home + Auto + Umbrella |
COMPETITIVE_RATES
| policyholder_id | best_competitor_rate | rate_gap_pct | num_quotes_found |
|---|---|---|---|
| PH-6601 | $2,900 | -9.4% | 3 |
| PH-6602 | $1,950 | +8.3% | 1 |
| PH-6603 | $5,800 | +7.4% | 2 |
RENEWAL_HISTORY
| policyholder_id | prior_increase_pct | renewed | year |
|---|---|---|---|
| PH-6601 | +6.0% | Yes | 2024 |
| PH-6601 | +11.9% | Pending | 2025 |
| PH-6602 | +4.7% | Yes | 2025 |
PORTFOLIO_TARGETS
| line_of_business | target_combined_ratio | current_combined_ratio | retention_target |
|---|---|---|---|
| Home | 92% | 95% | 88% |
| Auto | 97% | 101% | 85% |
| Umbrella | 85% | 82% | 90% |
Write your PQL query
Describe what to predict in 2–3 lines — Kumo handles the rest
PREDICT BOOL(POLICYHOLDERS.RENEWED = 'True', 0, 0, days) FOR EACH POLICYHOLDERS.POLICYHOLDER_ID ASSUMING RATE_CHANGES.PCT_CHANGE = 0.05
Prediction output
Every entity gets a score, updated continuously
| POLICYHOLDER_ID | CURRENT_PREMIUM | OPTIMAL_INCREASE | RETENTION_PROB | PROFIT_IMPACT |
|---|---|---|---|---|
| PH-6601 | $3,200 | +3.0% ($3,296) | 0.91 | +$96/yr |
| PH-6602 | $1,800 | +8.0% ($1,944) | 0.88 | +$144/yr |
| PH-6603 | $5,400 | +4.0% ($5,616) | 0.95 | +$216/yr |
Understand why
Every prediction includes feature attributions — no black boxes
Policyholder PH-6601 (Home + Auto, $3,200)
Predicted: Optimal increase: +3.0%, 91% retention probability
Top contributing features
Competitive rate gap (below market)
-9.4%
28% attribution
Prior increase magnitude (already +11.9%)
Compound fatigue
24% attribution
Low loss ratio (attractive to competitors)
0.28
20% attribution
Active competitor quoting activity
3 quotes found
17% attribution
Tenure and bundling stickiness
6.4yr, 2 lines
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 pricing optimization
How does AI optimize insurance pricing?
AI optimizes insurance pricing by predicting individual-level price elasticity: how much rate increase each policyholder will accept before switching carriers. The model connects risk profiles, competitive alternatives, claims history, rate-change history, and service interactions to predict the optimal price point for each customer. The portfolio is then optimized to hit combined-ratio targets while maximizing retention of profitable customers.
What is price elasticity in insurance?
Price elasticity measures how sensitive a policyholder is to rate changes. A highly elastic customer (low loss ratio, many competitive alternatives) will leave with a small increase. An inelastic customer (higher risk, fewer alternatives) will stay even with a larger increase. Understanding individual-level elasticity is the key to pricing optimization: you can retain profitable customers with smaller increases while recovering margin on less price-sensitive ones.
How much revenue can AI pricing optimization generate for insurers?
Top-20 insurers generate $200-500M in annual value from individual-level pricing optimization. The value comes from three sources: retaining profitable customers who would have left under uniform rate increases (60% of value), recovering margin on price-insensitive customers (30% of value), and reducing acquisition costs to replace lost customers (10% of value).
Can AI pricing models satisfy insurance regulatory requirements?
Yes. The standard approach is to use the relational model's individual-level elasticity predictions as an input factor in a GLM-based rate structure that is filed with state DOIs. This gives you the optimization power of graph-based learning with the transparent factor structure that regulators require. Several states are also beginning to accept ML-based rate models directly with appropriate documentation.
Bottom line: Optimize individual-level pricing to improve combined ratios by 2-4 points while retaining 92%+ of profitable policyholders, generating $200-500M in annual value for a top-20 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.
