Proof

Four engagements. What was broken, what I changed, what the numbers did. No adjectives.

Case 01 — Informa 2017–present

Quality platform and zero-downtime delivery

Situation A monolithic CMS behind a global B2B events and information business — content and e-commerce across thousands of events and 5,000+ sites. Fortnightly releases. Regression by hand. No automated safety net under a platform the business ran on.

Intervention Designed the test architecture for the monolith-to-multi-service migration and built the in-house frameworks to carry it: custom JavaScript to TypeScript suites, Puppeteer to Playwright, Percy visual regression, Lighthouse performance, Artillery load and stress. Embedded all of it as CI/CD quality gates rather than a suite someone remembers to run. Then designed dependency-aware Jenkins pipelines across 9 services and 2 shared-library repositories, with blue/green deployment, explicit proceed and block conditions, and rollback protocols.

  • 0rollbacks in five years
  • +400%automation coverage growth
  • 70 → 15minutes to validate a release
  • −90%suite flakiness
  • 2 wks → 1planned release cadence, plus on-demand
  • 3 → 11QA engineers hired, trained, retained

Also on the record: traceable coverage across 100% of delivered user stories, roughly 65% more defects caught before production, and about 15 minutes from an approved pull request to a releasable build. Median pipeline time fell to about two-thirds of its previous duration, and to under half on tuned paths. The zero-rollback figure covers releases run through these processes.

Case 02 — Informa 2024–2026

AI-augmented quality engineering a QA lead can actually sign off

Situation AI in the delivery pipeline as a slogan rather than a system. Acceptance criteria arriving vague or contradictory and only failing at release. Manual test design consuming the QA time that should have gone into judgment.

Intervention Defined and drove human-in-the-loop agent workflows across three teams on a multi-service CMS, built with a QA colleague and tuned against real tickets with the teams that use them — Claude Code and Codex, custom skills and agents, the Atlassian, GitHub and TestRail MCP servers, and browser automation. One agent reconstructs story, linked-work, comment and decision context to stress-test acceptance criteria before development starts. A second drafts and verifies manual test cases and publishes them to TestRail with an auto-created Test Run, with QA review and approval retained. Pull requests are analysed against the specific criteria they claim to implement, and every “implemented” finding has to carry file-and-line evidence. Output is a readiness verdict — ready, caution, not ready — plus coverage and blocker views. The workflow shows diffs and waits for confirmation. Nothing merges, publishes or releases on its own authority.

  • −60%missed or incomplete acceptance criteria reaching release
  • −45%manual test-design effort per story
  • +70%test cases and test runs created or updated

In practice its output fed human decisions: raise the bug, flag the pull request, decline the merge, block the release, send the acceptance criteria back to the product owner. That is the point. An agent that can ship without a person in the loop is a liability wearing a productivity costume.

Case 03 — K&C (Krusche & Company) 2021–present

A QA department built to survive its founder

Situation A small QA team serving multiple client engagements with no shared definition of what a QA engineer at any level was supposed to be able to do. Hiring by instinct. Quality varying by whoever happened to be staffed.

Intervention Company-wide QA strategy, then the machinery to run it without me: a hiring and assessment pipeline with designed interview exercises, onboarding standards, a five-level role grading framework for manual QA professionals, and a Senior QA capability framework derived from selected SFIAplus source material and tailored to the organisation’s actual expectations. Tooling standardised, internal documentation written, skill matrices maintained. I remain the escalation point for internal and external project leads — but the system decides, not me.

  • 500%QA org growth over three years
  • >90%quality of hire
  • 15direct reports across five projects
  • 5level grading framework, still in use

The Senior QA capability framework is an internal framework built on SFIAplus source material — not a SFIA certification or accreditation. Its content is published on this site at /senior-qa/, because a capability framework nobody can read is just a filing cabinet.

Case 04 — Wargaming.net Persha Studia · 2011–2016

Release coordination across four regions, and a game launcher

Situation AAA online games shipping to players in Belarus, Australia, the United States and the EU, with builds, localisation and release windows that all had to line up.

Intervention Ran the full testing domain for World of Warplanes — A/B, localisation, performance, client/server, network, peer-to-peer distribution and hash verification — while the QA department scaled around the programme. Led QA for the WDC Unified Game Launcher: installer and auto-updater with rollback and delta-update support, peer-to-peer distribution, hash-sum verification, network optimisation and failover, multilingual UX. Later, as QA Process Coordinator, owned the cross-team process, KPIs and shift-left standards.

  • 10 → 40QA engineers as the programme scaled
  • 4regions coordinated per release
  • 4QA team led as Lead QA Engineer

Rollback-safe delivery is not something I picked up from a conference talk. I shipped a consumer game launcher with delta updates and rollback in 2014, when getting it wrong meant bricking someone’s install.

Published by the client, not by me

Two of these engagements were written up as case studies by K&C and published under their own name:

Recognised internally with Top Team 2022 at the Informa Awards.