Skip to content

AEGISA churn decision system · Vinctum

Aegis turns churn risk into a decision, with its reasons and its cost: whom to intervene with, how, and when spending would lose money.

Working MVP · Looking for a first design partner

AEGIS · retention queuerepresentative data
CUSTOMERRISKDECISION
CUSTOMER#C-20417 · Monthly plan
MONTHLY VALUE₺400
01 · RISK
78/ 100
02 · REASON

Usage down 41% in 30 days · last login 19 days ago

03 · EXPECTED VALUE
Reminder email₺10+₺140
Personal outreach₺150+₺786
40% off for 3 months₺480−₺31
04 · DECISIONPersonal outreach

Expected value ₺786. The pricier options are not worth their cost.

5 customers · 3 act · 1 watch · 1 acceptEvery decision is logged with its signals and assumptions

01 Why Aegis

02 Platform

Most churn projects end with a score. In Aegis, the score is only one of six inputs to the decision.

01

Data

Customer history and churn outcome. No direct identifiers needed.

Data contract
02

Feature pipeline

The same feature state in training and serving. No train/serve skew.

Polars
03

Calibrated risk

If it says 78%, it means about 78%. Probabilities are measured and tuned.

XGBoost · LightGBM · CatBoost
04

Reasons

Why the risk is high and what would change it, in plain language.

SHAP · DiCE
05

Effect

The real effect of an intervention. Without treatment/control data it is marked as simulation.

T-learner uplift
06

Decision

Risk × effect × customer value − cost. Nothing is spent unless it is positive.

CLTV · margin · WACC
expected value=risk×effect×12-month value−cost3 interventions · computed per customer

03 Foundations

01

Role-based access

JWT and RBAC scopes. Every request runs in its own tenant context.

02

Redacted audit log

Every decision is logged; sensitive fields are masked before they are written.

03

Replayable decisions

Model and policy registries are enough to reproduce a decision later from the same input.

04

Evidence-guarded effect

Without a real treatment/control column, uplift is never presented as proof.

05

Drift and governance

Drift, policy and customer drill-down in one operations dashboard.

06

On your infrastructure

Runs in Docker, with Postgres, Redis and DuckDB integrations ready.

It runs at pilot scale today. Deployment, data access and retention are set in the pilot agreement.

Today

A working product.

Risk prediction, reasons, expected-value recommendations and the audit log work end to end.

Next

An effect not yet proven.

There is no verified retention or ROI figure. We will measure the real effect on a first design partner's data, with a control group, whatever the result.

05 Vinctum

Vinctum is an independent product studio building software that makes decisions explainable and verifiable after the fact. The principle behind Aegis, that every decision can be accounted for, runs through our open infrastructure too.

06 Pilot

One data table, a 30-minute call and a controlled measurement. See which intervention adds value for which customer, on your own data.