Investment Insights from Two Years of AI-Powered Postmortem Analysis

Zalando recently published a report revealing how the company turned its postmortems (if you’re still living on the Scrum planet, think of them as Retrospectives!) — once seen as static reports of failure, written actions buried somewhere (hopefully on a Confluence page!) that teams never revisited after discussions — into a powerful source of strategic insight.

Traditionally, every lesson learned ended with a human-written postmortem or some other format meant to capture what went wrong and how to avoid it in the future (come on! this is the business world! those that went well were definitely forgotten at the speed of light!). However, as these reports accumulated over time, their collective value remained locked away — too vast and inconsistent for manual analysis. Here exactly is the moment that an Idea comes to the people’s species mind:😁

👉 What if this bulk of non-processed information could make our entire system smarter?

This was exactly the question Zalando’s SRE team asked when they looked at their vast collection of infrastructure failure data. To answer it, they built a multi-stage AI pipeline powered by Large Language Models (LLMs). This system automatically summarized thousands of postmortems, identified the technologies involved (like AWS S3, DynamoDB, or Postgres), analyzed the root causes, and surfaced cross-system failure patterns. What once took weeks of human effort could now be distilled into actionable insights within hours.

The results were striking: the AI uncovered recurring pain points — from misconfigurations and scaling issues to weak change management — helping the company target investments and prevent similar incidents in the future. Although Zalando emphasizes that AI didn’t replace human judgment (and as a human, I truly hope that’s true!), AI-driven systems have clearly helped them a lot — especially when we look at the big picture. It’s about scale, accuracy, performance, reliability, maintainability, etc. — all working together. 🚀