Rating deployment

A home for the existing estate

Start with the rating logic you already have.

Bring current models or lookup tables into a structured platform through the supported migration path, then improve them without replacing everything at once.

Excel-like formulas

Actuaries can get started with familiar formulas, without learning a new coding language. Exakt adds operations beyond Excel's built-in functions.

Models and lookups

Combine supported legacy model outputs, Exakt models, and multidimensional lookup tables within the product limits.

One rating definition

Mix existing and new components while keeping the calculation versioned and testable.

Product workflow

Review the calculation, release, and production activity in one place.

Supported functions, model formats, lookup dimensions, and production controls are confirmed for each implementation.

Illustrative deployment workspaceRating v24

Build and inspect one versioned rating package.

Compose formulas, supported models, and lookups, then review the calculation and its dependencies before testing it for release.

Draft

1LTV=list_multiply(list_apply(Lapse_prob,lambda x: (1-x)),premium_delta)
2lapse_condition_mask = list_apply(Lapse_prob, lambda x: if(x <= 0.35, 1, 0))
3delta_condition_mask = list_apply(premium_delta, lambda x: if(x <= 150, 1, 0))
4filtered_LTV_values = list_multiply(LTV,delta_condition_mask,lapse_condition_mask)
5max_LTV = list_max(filtered_LTV_values)
6best_scenario = coalesce(scenarios[list_position(LTV,max_LTV)],scenarios[list_position(scenarios, 1.00)])
7office_premium = coalesce(round(best_scenario * Total_PP * 1/0.7,2),Total_PP)

Debug

This pricing package evaluates alternative premium scenarios and selects a recommended premium based on expected lifetime value (LTV) and eligibility constraints.

Pipeline flow

Input Node

Receives policyholder, policy, geography, and current-premium data, including:

age, gender, policy/product types
prev_premium
municipality identifier MUNI_ID
exposure and seniority measures

Enrichment

Converts coded fields into labels and enriches the input using lookup tables. For example:

new_business is converted from 'Yes' / other into 1 / 0.
Category labels are looked up for distribution channel, gender, product type, and policy type.
Municipality-derived risk variables such as C_C, C_GE_P, and IICIMUN are added.

PP calculation

Produces total pure premium:

Total_PP = pred.Serious_PP + pred.Non_Serious_PP

The two components come from an upstream prediction package.

Fix deployment for the pipeline with given errors.Write code for a translation layer

What would you like to know?

＋◎ Search✣ GPT Sol

Track production traffic and trace unexpected quotes.

Review quote volume, response times, failures, and defaults for the selected version, then open a calculation trace when a result needs investigation.

Overview

TimingsVariablesDefaultsLogs

Updated just now

Quote volume287,241Last 6 hours

Average response205.6 msRequest time

Success rate100.0%287,241 successful

Failures0No failed quotes

Default hit rate0.18%516 factor defaults

Quote volumes

Successful quote rate overlaid on hourly volume.

Quotes

Response times

Request processing and model inference time.

Milliseconds

Latest quote traceTest quote 1 · Rating v24

Input
Enrichment
Risk model
Response

Commit ID: 12b85639-e6fd-4c2a-af9c-97d50e78f872
Quote ID: Test quote 1
Calculation ID: Test result
Steps: 9
Calculation steps: 32
CodeNode formulas: Hidden

Trace payloadAll test-result keys, including timings

{ "quote_id": "Test quote 1", "calculation_id": "Test result", "duration_ms": 11.596 }

Step 1 · Input NodeDeployment Inputs

Step 2 · EnrichmentFreeform

Step type: Freeform
Node index: 1
Calculations: 20
Duration: 11.596 ms

Calculation steps

Calculations performed in the enrichment step
#	Variable	Formula	Inputs	Result	Type
1	new_businessComputed	`if(new_business='Yes',1,0)`	new_business 1	1	int
2	type_policy_dgComputed	`if(type_policy_dg='SE','S',type_policy_dg)`	type_policy_dg S	S	str
3	new_business_labelLookup	`lookup('new_business_lookup', new_business)`	new_business 1	New business	str
4	reimbursement_labelLookup	`lookup('reimbursement_lookup', reimbursement)`	reimbursement No	No reimbursement	str
5	distribution_channel_labelLookup	`lookup('distribution_channel_lookup', distribution_channel)`	distribution_channel A	Agency	str
6	gender_labelLookup	`lookup('gender_lookup', gender)`	gender F	Female	str

Excel-like formulas

Start with familiar formulas. Go beyond Excel.

Actuaries can get started without learning a new coding language. Use familiar formula syntax to build rates, then draw on operations beyond Excel's built-in functions, including supported models and lookup tables.

1LTV=list_multiply(list_apply(Lapse_prob,lambda x: (1-x)),premium_delta)
2lapse_condition_mask = list_apply(Lapse_prob, lambda x: if(x <= 0.35, 1, 0))
3delta_condition_mask = list_apply(premium_delta, lambda x: if(x <= 150, 1, 0))
4filtered_LTV_values = list_multiply(LTV,delta_condition_mask,lapse_condition_mask)
5max_LTV = list_max(filtered_LTV_values)
6best_scenario = coalesce(scenarios[list_position(LTV,max_LTV)],scenarios[list_position(scenarios, 1.00)])
7office_premium = coalesce(round(best_scenario * Total_PP * 1/0.7,2),Total_PP)

Lookup tables

Use lookup tables as first-class rating components.

Keep factors, limits, and reference data in versioned tables alongside formulas and models. This example shows one input dimension; Exakt supports lookups with more than ten input dimensions when a rating structure needs them.

Illustrative age-band lookup with motor frequency and severity factors
AGEC	MTPL_FREQ	MTPL_SEV
MISSING	1	1
<= 0.5	1.38474446986	1.38209121485
(0.5, 1.5]	1.1576664828	1.12728754208
(1.5, 2.5]	0.932941844956	1.01984157847
(2.5, 3.5]	0.918630059007	0.992860448939
(3.5, 4.5]	0.964375623241	0.97938859414
(4.5, 5.5]	0.980607024405	0.988751095881
(5.5, 6.5]	0.996200882077	0.983183507692
(6.5, 7.5]	1.01019374151	0.98348197432
(7.5, 8.5]	1.01762922884	0.991939767816
(8.5, 9.5]	1.02120892421	1.01003084668
(9.5, 10.5]	1.03074293379	0.994609764902
(10.5, 11.5]	1.04611554271	0.98904327953
(11.5, 12.5]	1.051498568	0.978498677205
(12.5, 13.5]	1.05789230742	0.95680864923
(13.5, 14.5]	1.0552176781	0.96804151045
(14.5, 15.5]	1.03016531111	1.01874737923

Built in AI assistant

Draft, explain, check, and debug the calculation.

The assistant can propose a formula, explain existing code, check syntax, and help investigate a failed test. Every proposed change remains visible for user review. The assistant does not approve or release a rate.

Pipeline flow

Input Node

Receives policyholder, policy, geography, and current-premium data, including:

age, gender, policy/product types
prev_premium
municipality identifier MUNI_ID
exposure and seniority measures

Enrichment

Converts coded fields into labels and enriches the input using lookup tables. For example:

new_business is converted from 'Yes' / other into 1 / 0.
Category labels are looked up for distribution channel, gender, product type, and policy type.
Municipality-derived risk variables such as C_C, C_GE_P, and IICIMUN are added.

PP calculation

Produces total pure premium:

Total_PP = pred.Serious_PP + pred.Non_Serious_PP

The two components come from an upstream prediction package.

Fix deployment for the pipeline with given errors.Write code for a translation layer

What would you like to know?

＋◎ Search✣ GPT Sol

Testing and production

Keep environments separate and versions identical.

Separate endpoints, API keys, data access, and permissions between testing and production. Run test calls and parity checks, record the approval, set the A/B allocation and stopping rule, and promote the validated version. Retain attribution and a rollback path.

Live monitoring

See what is happening with production quotes.

Monitor quote volumes and response times as requests arrive. Inspect calculation traces to see the input, formula, model, lookup, factor result, default, or failure behind an unexpected price.

VolumeTrack quote traffic by version and environment.
PerformanceInspect response time and endpoint behaviour.
FactorsAnalyse how rating factors behave in live traffic.
Trace controlsApply the configured delay, retention, search, access, and sensitive data rules.

API and model lifecycle

Run multiple versions without losing control.

Manage endpoints, API keys, rating versions, supported model formats, approvals, traffic allocation, rollback, and retirement in one lifecycle. Publish response time and throughput only with the test conditions and supported limits.

A good model implemented incorrectly is still wrong.

Exakt keeps the approved calculation intact from portfolio test to production request.

Deploy the rate you tested.

Start with the rating logic you already have.

Excel-like formulas

Models and lookups

One rating definition

Review the calculation, release, and production activity in one place.

Build and inspect one versioned rating package.

Promote the tested version.

Track production traffic and trace unexpected quotes.

Overview

Quote volumes

Response times

Calculation Trace

Start with familiar formulas. Go beyond Excel.

Use lookup tables as first-class rating components.

Draft, explain, check, and debug the calculation.

Pipeline flow

Input Node

Enrichment

PP calculation

Keep environments separate and versions identical.

See what is happening with production quotes.

Run multiple versions without losing control.

Monitor the rate, learn from production, and start the next model branch.