Opta6
They were data, integration or process problems with an AI budget attached. We run fixed-price assessments that tell executives which is which — with a working prototype to prove it.
The useful question isn't "where can we use AI?"
It's "where is intelligence the bottleneck?"
Track record
Opta6 has delivered software and architecture into regulated, multi-entity operations. The AI assessment is new; the ground it covers is not.
What we do
We'd rather be known for one thing that works than list eight services. This is the one that's live.
Two weeks to establish where intelligence is genuinely the bottleneck across your operating model, ending in a one-page executive heat map with a defensible reason recorded for every candidate — in and out. Then three to six weeks building a working prototype on the one or two processes worth proving, using synthetic or sanitised data.
The method, run end to end
A VAT reclaim pipeline, built on a synthetic corpus of 120 invoices across 40 counterparties and three jurisdictions, so that every correct answer was known before the system saw the document.
The finding worth your attention: automating 26 exception cases created 47 new ones. Different in shape, different in where they concentrate, and invisible until the automation exists.
Synthetic data, simulated review decisions, no client engagement. We say so here because a proof point without its limits stated isn’t one.
Why us
Which means the answer can be "don't". The most valuable output of an assessment is usually the list of things you've decided not to do — and that's a hard finding for anyone whose next quarter depends on the build.
The failure modes here are integration, data ownership and adoption. Those are architecture problems, and they're what we've spent twenty-five years doing in regulated, multi-country operations.
Working software on a sanitised sample, with measured failure modes — not a slideware pilot that needs another pilot.
Who we are
Opta6 is a small UK consultancy. Engagements are run by the person who does the work, and capacity is deliberately limited to two at a time.
Enterprise & data architect — leads the AI Opportunity Sprint
Twenty-five years designing the systems this assessment interrogates: regulatory reporting at Northern Trust, master data and integration architecture at Walgreens Boots Alliance, and group enterprise architecture across 22 countries at Brammer.
The AI half isn't advisory. Patrick founded, architected and built MyAdminMate — a voice-first AI platform for UK tradespeople — and still runs it in production: LLM pipelines, Whisper speech recognition, semantic search on pgvector, the mobile apps and the infrastructure under them. Model cost, latency and accuracy are a monthly operating problem here, not a slide.
That is the view the Sprint is built on: an assessment written by someone who has had to make models work against real data, on a real budget — not a survey of what they are said to do.
The monthly note
Two hundred words, no sequence, no pitch. Reply if it's a live problem and we'll pick it up; unsubscribe and we'll stop the same day.
Corporate addresses only. We keep a legitimate interests assessment on file and never share the list.