Case studies · From the build

Proof, not promises.

The editorial is the thinking. This is the doing. The same methods, run against real systems: data platforms, BI, CRM rollouts, reusable AI skills and right-sized adoption. Each one written up plainly: the problem, the approach, what shipped, and what it actually changed. No vendor gloss, no vanity metrics.

05
Real builds
1,400+
Sites, one warehouse
~150
Shadow AI users heard
Days→min
Compliance run
A drafting desk at night: a cream systems schematic unrolled across it, four tabbed binders standing behind, one route across the drawing marked in acid-lime under a single lamp.
From the build · AI Sustained
01 · Think

The editorial.

Every method starts as a written argument, tested against evidence rather than enthusiasm. Read the issues.

02 · Apply

The day job.

The idea gets run in a real business, on real systems, with real constraints and no demo data.

03 · Prove

The case study.

What shipped, what it cost, what broke, what changed. The receipts, written up below.

Tile art for case study 001, asking the workforce about shadow AI use. Case study 001
Agentic AI · Shadow AI AI adoption

Before you write an AI policy, ask the people already using it.

The workforce is already using AI on personal and business accounts. Rather than a policy written in the dark, we ran an anonymous workforce survey, and handed the build to an AI agent that designed and configured it through a browser in about fifteen minutes.

~150
Shadow AI users
~15 min
Agent build time
12
Survey questions
Read the case study
Tile art for case study 002, many source systems joined into one governed warehouse. Case study 002
Data platform · Integration Multi-site · 1,400–1,800 sites

Before you overlay AI, build the single source of truth.

A group of 1,400–1,800 sites taking till and app sales, running on disconnected systems and load-bearing spreadsheets, joined up into one governed warehouse (finance, CRM, sales and more), so AI finally had trustworthy data to stand on.

1,400+
Sites/stores
£180–300m
Revenue
~£2–3m
To one source
Read the case study
Tile art for case study 003, Charles, a semi-autonomous virtual agent. Case study 003
Agentic AI · Virtual agents AI adoption

Bolster the team: hire a semi-autonomous agent.

"Charles", named after Charles Babbage, is my first virtual hire: he runs my Jira backlog, drafts meeting notes and books meetings. When the email and Teams connectors turned out read-only, a Power Automate, OneNote and SharePoint bridge did the sending. No engineers required.

001
First agent
Jira
Backlog owned
No-code
Send bridge
Read the case study
Cover art for case study 004 — a tabbed binder of written procedures under a single desk lamp. Case study 004
Practical AI · Skills AI adoption

Prompts forget. Skills remember.

Twelve reusable skills, written in plain English by an analyst with no engineering background, turned repeat work into one-line commands. That includes a safety-critical compliance run that went from days to minutes, and once refused to proceed when the source contradicted the instruction.

12
Skills, no code
Days→~5min
Compliance run
30–60 min
To build one
Read the case study
Cover art for case study 005, a laptop showing a grid of quiz results with one row highlighted in acid yellow, a phone propped beside it mid-screenshot. Case study 005
AI Made Simple · No-code HubSpot · Quiz analytics

Screenshot. Paste. Build.

The AI Survival Quiz needed somewhere to put 55+ results, so a complete HubSpot beginner built a CRM data warehouse that scores every submission against the field and replies on its own. Built by pasting screenshots into an AI, with zero training, and the quiz stays live to keep feeding the analytics.

55+
Quiz submissions
0 hrs
HubSpot training
~0 min
Ongoing admin
Read the case study
Get in touch

Do you want to hear more?

Every build here was done inside a normal working week, without an engineering team. Happy to talk through how any of it was done. Drop me a line at [email protected].

Contact me