Real numbers, or no page at all.
Every case study published here will meet the same standard: the baseline measured before anything was touched, the workflow that was built, the change measured afterwards in the same units, and the owner saying so in their own words. Nothing composite, no borrowed logos, no percentages without the figures behind them.
Instrumented from day one, so the results can be published.
The standard
Results are measured before they are published.
A write-up earns this page when it can show the baseline measured before anything was touched, the fixed scope that was agreed, what moved afterwards in response time, hours returned and conversion — and the owner's own words on whether it was worth it. That standard takes a full build cycle to meet, and nothing appears here until it does. The fastest way to see the work is from the inside: the diagnostic puts days of senior attention on your own workflow and ends in written findings that are yours to keep.
Anatomy of a case study
Three parts, in this order. A build that cannot fill all three does not get a page here.
01
The leak, priced in the client's own numbers
Where the money was going before anything was built: enquiries that went unanswered past the hour they were worth answering, the same details re-typed between systems, quotes that left days after they were asked for. Counted from the client's own inbox and calendar, not from an industry benchmark, and multiplied out to a monthly cost the owner recognises as theirs.
02
What was built, and how it behaves when things fail
The one workflow that was shipped, end to end, at the fixed scope agreed before the work started. What it does on a bad day matters more than what it does in a demo: where it retries and where it stops, what it refuses to guess at, and the point at which it hands a person the full context instead of going quiet.
03
The measured delta, in the same units as the baseline
The same instrument read a second time, after the build has been live long enough for the reading to mean something. Before and after side by side, measured the same way both times, with the figure the owner would confirm out loud on a call — and their own words on whether it was worth the money.
In the meantime
The journal is where the thinking sits.
Notes on what it takes for an AI build to survive real data, where automation pays for itself first, and what a production system actually involves. Less polished than a case study, and available now.