Skip to content
Our Work
Telecom

KPN's collection workflow fired messages on schedule — not when customers were ready to pay

KPN's collection workflow fired messages on schedule — not when customers were ready to pay

Operational context

KPN had already moved overdue payment handling away from manual call centres into a large-scale digital communication operation. Luminos Software developed and operated the messaging infrastructure across email and SMS channels. The operational foundation was in place.

What was breaking

Communication timing was static. Messages were sent according to predefined schedules rather than based on how individual customers actually behaved. Payment behaviour varied significantly — by customer age, relationship duration, whether the overdue balance was a first occurrence or recurring, and by the size of the outstanding amount. The workflow treated all of them the same.

Why it mattered

Recovery rates were lower than the underlying data made possible. Overdue balances stayed open longer than necessary. More cases escalated to external debt collectors — adding cost and operational pressure downstream. The problem was not the volume of communication. The problem was execution timing.

What Luminos Software deployed

Luminos Software developed a custom AI prediction model integrated directly into the existing collections workflow. The model analysed a broad set of behavioural and operational features per customer and automatically:

  • predicted the optimal communication moment for each individual customer
  • adjusted email and SMS send timing dynamically — replacing fixed campaign schedules
  • executed within existing operational and regulatory time constraints
  • fed operational outcomes back into the model to improve prediction quality over time
  • ran continuously inside a live, high-volume production environment

Operational outcome

Successful monthly payment recoveries increased substantially, sustained across multiple months. Recovery rates improved, outstanding debt exposure dropped, and fewer cases escalated to external collectors. The improvement came entirely from optimising execution timing within the existing operation. No increase in message volume. No additional manual effort.

TelecomPayment RecoveryPrediction ModelCollections

Does this sound familiar?

If your operation depends on manual coordination, disconnected systems, or workflows that only move when someone remembers — we can help.