Predictive Intelligence

Predict. Assess. Optimise.

There is no shortage of forecasts. What matters is which actions are economically worthwhile and how reliable the underlying figures are. Two engines turn forecasts into informed decisions: the Retention Engine uses calibrated probabilities to identify customers at risk of churning and determines where action is worthwhile. The Forecast Engine predicts financial flows with quantified uncertainty and derives economically appropriate buffers, cash positions and reserve levels.

Explore the products

Ein Bildschirm mit einem Diagramm und Linien darauf

Our products

Two engines, one principle: they turn your data into calibrated probabilities and those probabilities into economically sound decisions. The Retention Engine determines which customers are worth targeting; the Forecast Engine calculates the required level of provision. Both are developed and implemented by us, with ongoing operation available as well.

Zur Produktlinie

Retention Engine

Identifies churn risk early and determines which retention measures are worthwhile.

It forecasts churn risk as a calibrated probability: a figure that is meant seriously and whose quality is continuously re-measured. To do this, it combines several models with different error structures into an ensemble that holds up more reliably than any single model. The decisive part is what happens next: the probability is linked to your cost situation. What does an unnecessary measure cost, and what does an overlooked churner? From both, the engine calculates the economically correct decision threshold instead of adopting a blanket default. The result is not a score, but a prioritised work list.

Calibrated probabilities – measured, not asserted
An ensemble of several models rather than trust in a single one
The decision threshold follows your cost situation, not a standard setting

Zur Produktlinie

Forecast Engine

Forecasts financial flows with uncertainty that can be measured.

It forecasts financial quantities such as liquidity, cash flows and funding needs as full probability distributions: the range of possible outcomes instead of a single figure that merely feigns certainty. Which time series are learned jointly and which separately is decided by a documented comparison on your data, not by model ideology; forecast quality is measured and reported per series. Because error costs are asymmetric, the engine derives the economically correct provision directly from the distribution and your cost profile: the liquidity buffer, the cash positioning, the level of reserves.

Distributions and quantiles instead of point estimates – the way finance thinks about risk
Local, global or grouped: the model architecture is decided by comparing your data
Provisions derived directly: buffers, cash positioning and reserve levels

Zur Produktlinie

The challenge

A forecast alone is not yet a decision. Knowing which customers are at high risk of churning does not tell you which are worth targeting with a retention measure. A forecast without quantified uncertainty cannot determine the appropriate level of buffers and provisions.

This is exactly where the costs arise: campaigns target customers where they cost more than they bring in, and financial forecasts follow point estimates even though the cost of error differs hugely in each direction.

Our answer

Two engines that take forecasts to their economic conclusion. They deliver not bare probabilities but decision bases: for which customer is the measure worthwhile? How high must provisions be under uncertainty?

The decision itself is always made by a human. Our products provide the robustly calculated, documented basis for it.

What we offer

From the first forecast to monitored operation – modular, and from a single source.

✓ Early detection of customers at risk of churning, with calibrated probabilities.
✓ Financial forecasts as full probability distributions.
✓ Documented model comparison on your data: local, global or grouped.

✓ Economic decision thresholds calculated from your cost situation rather than adopted by default.
✓ Sizing of liquidity buffers, cash positions and reserve levels.
✓ Ongoing quality and calibration evidence per customer, segment or time series.

In brief

Four key Predictive Intelligence terms explained briefly: probability distributions, economic decision thresholds, data drift, and local and global forecasting models.

What is a probability distribution?

A probability distribution describes the possible outcomes of a forecast and the likelihood of each one. This allows buffers and reserves to be planned while accounting for uncertainty.

What is an economic decision threshold?

An economic decision threshold is the point at which taking action becomes less costly than doing nothing. It is calculated from actual costs and expected benefits.

What is data drift?

Data drift occurs when customer behaviour, processes or market conditions change. This can reduce forecast quality and make model adjustments necessary. The engine continuously measures its own predictive quality and flags when adjustments are needed.

What is the difference between local and global forecasting models?

A local model is trained for an individual time series or segment, while a global model learns across multiple time series. A documented comparison using your data determines which approach performs better.

FAQ's

Does the Retention Engine decide which customers to approach?

No, the decision remains with your team. The Retention Engine identifies which customers are worth targeting and explains why. Decisions about whether, whom and how to approach are therefore based on transparent calculations.

How are forecasts turned into economically sound decisions?

The Forecast Engine translates forecasts into specific economic decisions. It considers the expected outcome, its uncertainty and the costs of potential forecasting errors. From this, it derives appropriate liquidity buffers, cash positions or reserve levels.

How does the engine detect when a forecast is no longer reliable?

Declining forecast quality is identified through continuous monitoring. The engines track predictive performance by customer, segment or time series and flag when adjustments are required.

When is a local or global forecasting model more suitable?

The appropriate model is determined through a documented comparison using your data. Local and global models are tested based on data volume and homogeneity. The deciding factor is demonstrated forecast performance, not model complexity.