The segments – and how Arc was built
Arc computes against two lenses today, with a third in build, and credits many more. Open an entry to see what each model sees, what it misses, and how Arc uses it.
How this was built
Arc is deliberately layered. The foundation is published segmentation science – More in Common's seven British segments, credited above, used exactly as published and never altered. On top sits the GBNH archetype lens: eight energy-specific motivational types developed for the book Green But Not Heard, from years inside the industry watching who buys, who refuses, and why the marketing misses. The two lenses answer different questions – who someone is in Britain, and why they would buy – and Arc keeps them separate on purpose.
Our evidence base combines primary research with UK households, run across three studies during 2026, with published third-party segmentation. Method notes, sample composition and field dates are available on request. The GBNH archetypes are sized and validated through our own research. Heat.day, Solar.day, EV.day and MyGreenPrint sit in the same Green But Not Heard family. The studies feed the engine's evidence grades; nothing joins the scoring tables without cross-examination – every table went to five independent AI models with a fixed challenge prompt before freezing, and the record of what they changed is on the method page.
The method is published; the recipe is not. What every page shows: the sources, the dates, the evidence states, the working behind each result. What stays in the engine: the full weight tables and blend constants. Enough transparency to be checked, not enough to be photocopied.
Where AI sits
The scoring is deterministic – the same inputs produce the same result, every time, with the working shown. AI sits above the engine, never inside it: it explains, adapts, and drafts from evidence packs the engine and the research supply. AI dresses the evidence; it never invents it. Where a page uses AI, it says so, and shows its inputs.
Two ways in
There are two routes into the Persona Builder. Describe a customer in plain words and a deterministic extractor reads what it can hear – showing each phrase it matched, guess by guess, for you to correct. Or skip the description and set the pickers directly. Either way, only what you confirm is scored, and anything unset simply says less rather than pretending to know.
Provenance
The book engages these frameworks as published at the time of writing. Research moves; Arc tracks current editions; where they differ, this site is the living version.