AVA — the Adaptable Valuation Approach — is a continuous backtesting and recalibration system for non-life insurance risk models. It compares the predictions of your current risk and retention model against observed loss data, isolates the segments where the model is mis-pricing, and produces an updated risk model that feeds straight back into Dynamic Pricing . AVA sits on top of Risk Reporting , reading the same diagnostic data.

This is how a static GLM tariff stops being static.

What it does

AVA targets a specific failure mode of GLM-based pricing. Even well-built risk models drift. As the book changes, as customer behaviour shifts, and as competitor pricing moves, the model’s calibration gets stale in some segments while remaining accurate in others. The hard part is not knowing the model has drifted. Risk Reporting tells you that. The hard part is knowing where to adjust, how much, and without breaking the segments that still work.

That is the problem AVA solves.

How the approach works

In short, AVA compares your risk model against observed losses parameter by parameter, identifies the segments where the divergence is statistically significant, and proposes targeted adjustments that bring the model back in line without disturbing the segments that are still calibrated correctly.

The mathematical detail of how the adjustments are computed and sequenced is something we walk through in person when an engagement starts. It is not what you should evaluate AVA on. What you should evaluate AVA on is whether the updated model it produces actually prices better.

What you get

The output of an AVA cycle is one thing: an updated risk model, delivered in the format your Dynamic Pricing engine consumes. Your team plugs it in, the tariff refreshes against the new model, and the loop closes.

The diagnostic artefacts AVA generates along the way (the parameter-by-parameter audit, the adjustment log, the model-vs-data plots) live with us. We use them to defend the changes when your team asks where they came from. We do not ship them as a deliverable unless you ask for them.

A real example

On a portfolio segmented by driver age, AVA has identified and corrected:

Frequency: repeated under-pricing of segments aged 81+, with upward adjustment of the risk loading.
Average claim: under-pricing of legal-entity policyholders, corrected upward.
Large-claim surcharge: reduction of overstated risk in segments that had been over-loaded.
Risk premium: targeted re-loading of old age segments and legal-entity policyholders, neutral or reduced loading elsewhere.

These are the kinds of moves that a manual quarterly review would either miss entirely or take six iterations to land. AVA lands them in one cycle.

How it runs

In production at clients today, AVA runs on a monthly cadence: once per cycle, after the latest Risk Reporting refresh, ahead of the next Dynamic Pricing release. The cadence is configurable per engagement.

How it fits with the rest

The products form a loop.

Risk Reporting captures what is happening in your portfolio, the data that AVA needs as input.
AVA identifies where the risk model is mispricing and produces an updated model.
The updated model feeds Dynamic Pricing, which prices each new quote against the refreshed tariff.

The cycle repeats. Pricing improves continuously, not in step changes.

Want to see it run

Write to info@com-pass.cz . We can walk through a real AVA cycle on a redacted comparable portfolio in the first call.