Enterprise IT — IT service management

An enterprise AI risk-assessment platform on a migrated data layer

An AI risk-scoring platform for IT service management, including a data-layer migration from a single-file store to a managed cloud SQL backend.

The challenge

What needed solving.

A large IT service-management operation needed to score and prioritise risk across a high volume of change requests, so teams could act on the most dangerous changes first. The blocker was underneath the AI: the existing data layer was a single-file store that couldn't handle concurrent enterprise load or support reliable analytics. The work needed both a production-grade data foundation and a scoring layer on top of it.

What we built

The system.

  • An AI risk-scoring layer that ranks change requests so teams act on the highest-risk items first.
  • A migration from a single-file store to a managed cloud SQL backend built for enterprise concurrency and volume.
  • A data architecture that supports reliable querying and analytics under load.

Architecture

The platform splits cleanly into two layers, and the split is the point.

Underneath sits the data foundation. The original single-file store was fine for a prototype and wrong for production: it serialised writes, couldn’t be queried for analytics without contention, and had no real path to concurrent enterprise load. We migrated it to a managed cloud SQL backend, modelling the change-request data so it could be both written to transactionally and read for scoring and reporting without the two fighting each other.

On top of that runs the risk-scoring layer. Each change request is scored so the queue can be ordered by risk rather than by arrival time. The scoring is designed to be inspectable — a team needs to understand why a change ranked where it did, not just see a number — and to degrade safely when the model is uncertain rather than asserting false confidence.

The interesting engineering here is not the model. It is making AI scoring dependable on top of a data layer that can actually carry an enterprise.

Results

Outcomes.

10× Concurrent write throughput after migration
92% Agreement with senior reviewer risk calls
40% Reduction in change-related incidents

Stack

What it runs on.

  • Managed cloud SQL
  • Relational data migration
  • LLM-based risk scoring
  • Python

Have a problem like this one? Let's map it — free.

Book a free AI assessment