Geiger's random forest flags twice as many at-risk customers as gut feel

Geiger GmbH, a German distributor of promotional products, was losing drop-shipment customers without seeing it coming. With professors and students from Hochschule Düsseldorf, it turned transactional, demographic and interaction data into a random-forest model that assigns every account a churn-risk score each month.

The scores feed a Power BI dashboard that sales representatives use to decide whom to call first. The deployed model identifies more than twice as many future churners as random targeting, so retention effort goes to accounts that would actually leave.

The project began as an academic collaboration and became part of daily sales operations, showing that a structured analytics project can pay off in a small B2B company without a big data team.

Score every account monthly so reps intervene before loss

Churn signals hide in transaction frequency, order size and contact patterns that reps cannot track manually.

Monthly automated scoring makes retention proactive instead of reactive to a cancelled account.

Power BI puts the model where reps already work, removing the adoption barrier of a new system.

A university partnership supplied modeling skill that the mid-sized firm could not hire internally.

In mid-market B2B, a model embedded in the sales team's existing dashboard beats a standalone analytics project.

The model is firmly anchored in Geiger's daily sales practice and continues to be refined. The case was published in INFORMS Journal on Applied Analytics, and the Hochschule Düsseldorf project became a template for how structured analytics projects create measurable value in mid-market B2B environments.