Selected work AI, fintech & enterprise ITSM
What I shipped
Six case studies · AI, fintech & enterprise ITSM
The brief is rarely the problem. The outcome is what has to hold up afterwards.
Each of these started somewhere other than where it landed. The research moved the brief — and the numbers held.
Four of these run at my current employer and sit under NDA. Those pages are encrypted: the narrative is sanitised, and it unlocks with a shared password I'm happy to send. Request access →
Case studies
I was asked to wire two systems together. I shipped a four-phase AI integration instead.
An AI/AIOps platform integrated into enterprise ITSM across six customer environments — sequenced by what de-risked the next phase, not by who shouted loudest.
6
enterprise customers integrated
Live
running in production
3 routes
one architecture
Read case study
Enterprise AI · ITSM
I was handed a card vault to build. I shipped a FinLife platform instead.
A fintech-and-lifestyle app for the UAE — everyday spending joined to the offers and rewards behind it, with Cencai as the AI assistant inside it.
24,000
users in under six months
Live
App Store + Play Store
Read case study
Consumer fintech · AI
Lawyers weren't going to switch tools. So we put the AI inside the ones they already use.
Twenty-plus interviews said nobody wanted another destination. The classifier shipped as a browser extension that works inline, where legal research actually happens.
30 sec
to understand a case, not fifteen minutes
$1M+
follow-on anchored
Read case study
Legal tech · Applied AI
Same move, three sizes
One demand, one programme, one portfolio — all at T-Systems, all under NDA.
Two thousand templates. I shipped an AI assistant for the ones people use.
The usage data said half the library was barely touched. Scope cut to what carried the volume — shipped in months, not years.
7 min
to open a change, down from 15
89%
first-pass compliance, up from 72%
Read case study
One demand
Everyone had an AI idea. I built the register that decided which were real.
Forty use cases, classified at intake: does this need a model, build or buy, and where does it sit against DORA. About one in ten wasn't AI at all.
40
AI use cases governed
509 hrs
a month back, conservatively counted
Read case study
One portfolio
Six customers wanted six surveys. I built one framework instead.
Almost every bespoke requirement was a parameter, not a product. What genuinely differed was privacy — and even that was a decision list I could write in advance.
1
framework, reused across tenants
GDPR · DORA
decided at onboarding
Read case study
One product line
Twelve years of this. If you want the version with your own product's problems in it, let's talk.