Processing & anti-fraud turnaround

Where it started

Two projects to replace vendor solutions had stalled. Anti-fraud: 2.5 years, nothing to show. Processing: 13 months, nothing to show. The leads, architects and delivery people had left the company.

What I did

I took over both teams and rebuilt how they worked:

• A steady rhythm - Kanban instead of artificial SCRUM cycles.
• A unified stack - starting with operations, moving away from a zoo of languages and frameworks.
• DDD - to synchronise the language and expectations of the team and the business.
• Rebuilt teams.
• Documentation and specs as code - the whole team writes and reviews requirements.
• Teams set their own rules through ADR/RFC.
• Automated routine - dailies, reviews.
• Explicit research and learning tasks on the board, synchronised with what the business needs.

Results

• Routine shrank sharply - people pick their own tasks.
• Research and learning activities produced consistent results: faster delivery, fewer incidents.
• Docs-as-code gave everyone even, high-quality context.
• Shared practices enabled inner source and easy movement of people between teams.
• DDD plus docs-as-code drove rework down to almost nothing.

The through-line

This is the foundation the AI SDLC transformation was later built on - DDD, docs-as-code and team-wide artefact review didn’t appear for AI, they were already how these teams worked.

© Dmitry Roytman