What shipped, what it changed, 2020 to today

AI automation case studies, and the numbers behind them.

These are the systems and platforms Paul Rahme has shipped: six AI systems in production at a US marina-software company, 2025–2026, a dealership data pipeline, and QA and delivery work for a bank, a national tourism platform, a global consultancy, and a media streaming service. Every number below has a case behind it, and every case names what was built, for whom, and when.

Headline numbers

~60%
of incoming support tickets resolved autonomously, zero human touch, 24/7
AI support agent · the marina-software client · 2025–2026 · read the client's support-agent case →
~80%
of manual QA effort removed, with tests generated, run, and maintained by AI on a schedule
AI test automation · the marina-software client · 2025–2026 · read the self-testing QA case →
~70%
less manual regression testing via end-to-end frameworks, before AI entered the picture
Cypress / WebdriverIO frameworks · Eurisko · 2024–2026 · read the Shasha Media case →
2×
faster delivery from ticket to review-ready pull request
AI dev agent · the marina-software client · 2025–2026 · read the ticket-to-PR case →

The marina-software client figures are rounded from production data at the client, 2025–2026; the Eurisko figure is from the Shasha Media engagement. Definitions and measurement notes are on each case study.

Every one of these numbers started with one call.
The Automation Audit estimates yours before you spend anything: free, 30 minutes, written Automation Map included.
Book the Automation Audit →

Case studies

Case 01 · the marina-software client · the client · 2025–2026

Support stopped being a cost center

The work: a support queue absorbing engineering time, first responses taking hours.
The system: an autonomous agent with guardrails and confidence thresholds that triages, answers, and closes tickets, escalating to humans only when unsure.

Read the client's support-agent case →

~60%
of tickets resolved with no human in the loop
Seconds
to first response, down from hours
24/7
coverage without adding headcount
Case 02 · the marina-software client · the client · 2025–2026

QA that runs itself

The work: a manual QA cycle before every release, slow, repetitive, easy to skip under pressure.
The system: an AI pipeline that generates and executes automated tests on a schedule, catching regressions before they ship.

Read the self-testing QA case →

~80%
less manual QA effort
Every release
covered by full regression, zero-touch
0
dedicated manual QA cycles before each ship
Case 03 · the marina-software client · the client · 2025–2026

Engineers review. The machine builds.

The work: hours of setup per ticket before real engineering could start.
The system: a pipeline that reads Linear tickets, writes the code and the tests, survives the build gate, and opens review-ready pull requests.

Read the ticket-to-PR case →

faster ticket-to-PR turnaround
Review-only
developers approve instead of building from scratch
Human
on every merge, always
MORE CASES

Which of these is your queue? The audit writes its first page, free, in 30 minutes. Book the Automation Audit →

Screenshots from the systems

Illustrations of the systems' output, not production screens
Illustration: a nightly regression run passing
Regression, zero manual steps. A scheduled AI-generated suite finishing green before a release.
Illustration: an engineering board with the agent's pull request under review
A machine-opened pull request. Scoped from the ticket, tests included, waiting on human review.
Illustration: a helpdesk queue with tickets closed by the agent
The queue, resolving itself. Tickets triaged, answered, and closed with no human touch.

Where the work happened

QA-automation, development, and delivery roles at these organisations from 2020 to 2026. The AI systems were built and are run at a US marina-software company.

The marina-software clientmarina-management software — six AI systems in productionAI AUTOMATION · 2025–2026
McKinsey & CompanyInvest Kenya — investment-promotion platform, DubaiFRONT-END + QA · 2023
Saudi Tourism AuthorityVisit Saudi — national tourism platform; senior QA automation through Eurisko on the PwC-managed programmeQA AUTOMATION · 2024–2025
EuriskoShasha Media — automated testing across web, TV & mobileSENIOR QA + DEV · 2024–2026
mdgroupUK pharmacy platform — QA automation within an agile squadQA AUTOMATION · 2023–2024
AbirootFull-stack development and QA automation, LebanonDEV + QA AUTOMATION · 2020–2024

Web platforms shipped end to end (EpikDubai, Fresh Greens & Proteins, Yala Next) are on the Build page →

Questions about these case studies

Are these numbers audited?

No. They are rounded from production data at a US marina-software company, 2025–2026, and each carries its measurement definition on its own case page, so you can see exactly what was counted. Nobody external has verified them, and this page says so rather than implying otherwise.

Which of these systems are in production today?

Six: the AI support agent, AI test automation, the AI dev agent, the autonomous bug fixer, documentation automation, and the voice agent on the support line, all at a US marina-software company on its marina-management product. The voice agent is the most recent to go live, answering around 100 calls a month; its resolution figures are not yet published.

Can I speak to a reference?

Ask on the audit call. Referencing work done under a contract engagement needs the client's agreement, so it is arranged case by case rather than promised on a web page.

Which case study is closest to my situation?

If the cost is a support queue, start with the Marina-software support agent. If it is a slow release, read the self-testing QA pipeline. If it is an engineering backlog, read ticket to pull request. If it is bad data in a CRM, read the dealership enrichment pipeline.

Get your own numbers, in writing.

The free Automation Audit maps which repetitive work to remove first, and what it returns. 30 minutes, written Automation Map included.

Audit: free, 30 min →