Decisioning Strategy
The hub that will receive policy applications, run all models and risk assessments; then make approval decisions.
Introduction
The Decisioning Strategy service is responsible for approving new policies.
It is the heart of the Probability Engine where a request to accept a new policy will arrive.
It then collects C-Scores (collateral), L-Scores (Loan & Policy) and IR-Scores (Internal/Portfolio) from the other services; and outputs a decision that will be forwarded to the backend.
It will be implemented as a highly configurable rules + decision matrix system.
For example: use predictions x, y, z; reject if x under a threshold t; apply decision matrix on y and z; use price formula f(x,y,z); etc.
Methodology
An aircraft is a good metaphor to describe the Decisioning Strategy Service:
Cockpit dashboard: C-Scores, L-Scores and IR-Scores are the metrics and tools to pilot the plane.
The Concrete team is the pilot who sets the destination, configures the flight in auto-pilot; and manages exceptions.
The guiding principles for this module are:
Maximize flexibility so that any strategy can be implemented with a simple configuration update.
Allow the team to design and deploy strategies at will; and react to market conditions as fast as possible.
V1 Implementation
For launch, the V1 implementation is a minimal function that was proven safe and profitable by the simulation engine.
It is published in this github repository.
V2 Future Plan
The Decisioning Strategy Service will integrate more models in the future, and will rely on a Pub/Sub queue where the backend (or any authorized consumer system) will publish request messages:
Backend publishes a decision request message to a queue, and indicates a callback endpoint.
Decisioning Strategy worker processes the query, publishes a response in the appropriate Pub/Sub queue; and calls the callback with the response (decision about the policy.)
Expectation set to 30s for running all the models and coming up with the response.
Last updated