One model misses the full rate set
A model can perform well on its own while the combined rates change premium, loss ratio and risk mix across the book.
Portfolio simulation
Rapidly test and iterate on your rates.
Put rating structures head to head on the same risks, and see how your pricing strategy shapes your premium, business mix, margin, and adverse selection.
Simulation challenges
A rate change needs more than a model check.
Full-book tests need workarounds
Excel’s size limits and custom Python pipelines can turn each comparison into extracts, joins and reruns.
Averages hide pricing impact
Without segment views and what-if tests, it is hard to see where a change helps, where it hurts and what needs review before deployment.
Full portfolio testing
See the whole portfolio decision in one view
See how proposed rates compare with current rates across your full book before anything goes live. Bring in any model, from GLMs and rating tables to GAMs and GBMs, and combine them for a better view of risk.
See what changes across the whole book.
Compare current and proposed premium, loss ratio and auction outcomes on the same set of risks.
Average£ 591.26
Current £613.96−3.70%
Total£ 714.99 m
Current £742.44m−3.70%
Loss ratio51.46%
Current 49.55%+1.90 pts
Gini0.2234
Current 0.1901+17.53%
| Metric | Actual | Current | Proposed | Relative change | Absolute change |
|---|---|---|---|---|---|
| Portfolio result | |||||
| Average | £304.24 | £613.96 | £591.26 | −3.70% | −£22.70 |
| Total | £367,905,192.68 | £742,436,896.57 | £714,986,582.75 | −3.70% | −£27,450,313.82 |
| Loss ratio | — | 49.554% | 51.456% | +3.84% | +1.90 pts |
| Gini | 0.99449 | 0.19012 | 0.22344 | +17.53% | +0.03 |
| Competitive auction | |||||
| Auction average | — | £575.69 | £507.63 | −11.82% | −£68.07 |
| Auction total assigned | — | £281,398,294.66 | £365,627,977.41 | +29.93% | +£84,229,682.75 |
| Auction loss ratio | — | 64.42% | 51.05% | −20.73% | −13.37 pts |
| Auction profit | — | £100,181,957.32 | £178,963,433.55 | +78.64% | +£78,781,476.24 |
| Auction exposure share | — | 40.42% | 59.56% | +47.35% | +19.14 pts |
Illustrative monetary value
Understand the spread of every price change.
Plot exposure by the proposed-to-current premium ratio to see how much of the book moves, where the tails sit, and which risks need review before release.
MeasureProposed ÷ currentReference1.00 · no change
| Proposed-to-current premium ratio | Exposure | Approximate price change |
|---|---|---|
| (0.58, 0.62] | 460 | −40% |
| (0.70, 0.74] | 2,144 | −28% |
| (0.82, 0.86] | 7,234 | −16% |
| (0.90, 0.94] | 10,026 | −8% |
| (0.94, 0.98] | 10,065 | −4% |
| (0.98, 1.02] | 9,124 | No change |
| (1.10, 1.14] | 4,160 | +12% |
| (1.22, 1.26] | 1,535 | +24% |
| (1.38, 1.42] | 475 | +40% |
| (1.54, 1.58] | 105 | +56% |
Most exposure sits between an 8% reduction and no change. The thinner tails identify customers with the largest decreases or increases for targeted review.
Where do the current and proposed structures disagree?
Order risks by the proposed-to-current prediction ratio, then compare both estimates with the actual outcome in each equal-exposure bin. The same view can compare candidate models or complete rating structures.
Bins10 equal exposureOrderProposed ÷ current
| Portfolio bin | Average actual | Average proposed | Average current |
|---|---|---|---|
| 1 | 394 | 366 | 612 |
| 2 | 380 | 328 | 470 |
| 3 | 316 | 302 | 410 |
| 4 | 271 | 293 | 378 |
| 5 | 297 | 294 | 366 |
| 6 | 288 | 296 | 357 |
| 7 | 346 | 301 | 349 |
| 8 | 318 | 308 | 340 |
| 9 | 326 | 310 | 321 |
| 10 | 379 | 323 | 294 |
Proposed is closer to actual in 8 of 10 bins. The separation is strongest among the risks where the two structures disagree most.
Actual and predicted outcomes by driver age
Compare observed and predicted claim frequency with the fitted coefficient. Exposure is reflected directly beneath the curve.
VariableDriver ageMeasureClaim frequency
Actual, predicted, and fitted driver-age coefficients with exposure.
| Driver age | Actual | Predicted | Coefficient | Exposure share |
|---|---|---|---|---|
| 17 | 0.306 | 0.300 | 0.52 | 0.717% |
| 25 | 0.261 | 0.256 | 0.30 | 1.376% |
| 35 | 0.217 | 0.227 | 0.31 | 2.117% |
| 45 | 0.258 | 0.261 | 0.48 | 1.908% |
| 55 | 0.233 | 0.220 | 0.17 | 1.257% |
| 65 | 0.144 | 0.145 | −0.24 | 1.323% |
| 75 | 0.055 | 0.069 | −0.72 | 0.907% |
| 85 | 0.120 | 0.115 | −0.12 | 0.580% |
| 95 | 0.345 | 0.278 | 0.94 | 0.298% |
| 99 | 0.475 | 0.390 | 1.62 | 0.232% |
Proposed rate change by age and mileage
Rotate the two-way view to inspect every combination. Bar height shows the exposure supporting each cell; colour shows the proposed rate change.
InteractionDriver age × annual mileageHeightExposureColourProposed rate change
The illustrative view contains twenty combinations of driver-age and annual-mileage bands. Exposure is highest for drivers aged 35 to 49 travelling 5,000 to 10,000 miles. Proposed rate changes range from a 4.8% reduction to an 8.0% increase.
Three-dimensional driver age and annual mileage portfolio view.
| Driver age | Annual mileage | Exposure | Proposed rate change |
|---|---|---|---|
| 17–24 | 0–5k | 310 | −4.8% |
| 25–34 | 0–5k | 640 | −3.2% |
| 35–49 | 0–5k | 780 | −0.8% |
| 50–64 | 0–5k | 520 | +2.6% |
| 65+ | 0–5k | 260 | +5.4% |
| 17–24 | 5–10k | 420 | −3.1% |
| 25–34 | 5–10k | 920 | −1.4% |
| 35–49 | 5–10k | 1,120 | +0.7% |
| 50–64 | 5–10k | 810 | +3.8% |
| 65+ | 5–10k | 390 | +6.1% |
| 17–24 | 10–15k | 270 | −0.9% |
| 25–34 | 10–15k | 710 | +0.5% |
| 35–49 | 10–15k | 940 | +2.7% |
| 50–64 | 10–15k | 690 | +5.2% |
| 65+ | 10–15k | 310 | +7.4% |
| 17–24 | 15k+ | 140 | +1.8% |
| 25–34 | 15k+ | 350 | +3.1% |
| 35–49 | 15k+ | 460 | +4.9% |
| 50–64 | 15k+ | 320 | +6.8% |
| 65+ | 15k+ | 170 | +8.0% |
Proposed rate change by area
Compare the proposed and current structures across the portfolio. Colour shows rate movement and height shows the exposure supporting each area.
ColourProposed ÷ currentHeightExposure
Proposed ÷ currentLoading portfolio geography
Loading geographic portfolio data…
Portfolio explorer
Explore how a pricing change affects the entire portfolio.
Start with dislocation analysis to see the size and spread of price changes. Then use double lift to compare structures against observed outcomes, one way and two way views to inspect patterns, and geography to see where changes fall. Filter by segment, peril, model or rating structure.
Open portfolio viewsWhat-if analysis
Test what happens when costs change.
Use a custom expressions to test a change in cost for a specific group of risks. If electric vehicle repairs become 10% more expensive, apply that assumption only to electric vehicles, then compare expected claims, loss ratio and margin with the base case. Test any proposed rate adjustment against the same portfolio.
Illustrative scenarioEV repairs +10%
Custom expression
IfVehicle type is electric
ThenRepair cost assumption × 1.10
ElseKeep the base assumption
Compare with base case
See the portfolio effect.
- Expected claims
- Loss ratio
- Margin
Cross-package importance
See what drives each rating package.
Understand what variables drive the estimates for each underlying model in your rates.
Variable importanceIllustrative MTPL rating structure
8 variables · 2 packages| Rank | Variable | MTPL_Freq | MTPL_Sev |
|---|---|---|---|
| 1 | bm | 0.2118 | 0.0216 |
| 2 | coverage | 0.0249 | 0.0705 |
| 3 | ageph | 0.0915 | 0.0373 |
| 4 | postcode | — | 0.0423 |
| 5 | fuel | 0.0649 | — |
| 6 | power | 0.0637 | — |
| 7 | agec | 0.0413 | 0.0198 |
| 8 | fleet | 0.0078 | — |
Competitive rate auction
Test anti-selection risk.
Price the same portfolio with two rate sets. For each eligible risk, we assign the business to the lower price and carry the selected loss measure into the winning portfolio.
Find segments where a lower price attracts a worse expected business mix, then adjust the rating structure and run the auction again.
A lowest price auction isolates selection risk. It does not replace a demand model that accounts for customer behaviour, brand, channel, and product differences. But it can be used with competitor rates to see where you are losing business.
Saved and shareable views
Give reviewers the same versions, filters, and measures.
Share a link to the exact simulation view rather than rebuilding the analysis in slides. Authorised users open the same rating versions, portfolio filters, and selected measures. Access and data permissions apply to the shared view. Making it easy to review before setting it live.
- CompareNew rates, current rates, alternative models, and competitor rates when comparable data is available.
- FilterFocus on the segment, peril, geography, or version behind a result.
- InspectCheck how a rate change varies across exposure, age, mileage and geography.
Every portfolio decision stays tied to the rates that produced it.
When the selected strategy moves forward, the simulation evidence moves with it.