Skip to main content
The infrastructure that didn't exist: how Brandmine uses AI
Most platforms use AI to produce content faster. We built it to do something previously impossible: read 600+ primary sources per brand in the brand's own language, assemble the evidence, and present the result in a form institutional investors can use — across dozens of markets, in dozens of languages. Today we published the full methodology.
The brands Brandmine covers are not secret. A founding crisis documented in Коммерсантъ in 2009 is publicly available. The founder interview in 澎湃新闻 is indexed. The evidence is there.
What has never existed is the infrastructure to find it, read it, assemble it, verify it, and present it in a form institutional investors can use — across dozens of markets, in dozens of languages, at the depth serious due diligence requires.
That infrastructure is what we built. AI is the engine that makes it viable.
Nine functions, not one
AI at Brandmine is not a content generator. It operates across nine distinct functions in the production process.
Research — 600+ primary sources per brand run, in the brand’s native language. Not translated summaries. The originals: trade press, regulatory filings, founder interviews, consumer media, across 16 markets in Russian, Chinese, Hindi, Portuguese, Mongolian, and other languages. The linguistic moat is structural: machine translation of a source you don’t know exists produces nothing.
Synthesis — aggregating evidence from hundreds of sources into structured English: founding dates, ownership structure, crisis events, decisions made under pressure, verifiable outcomes, export markets, partnership history. It connects what a founder said in 2011 with what the trade press reported in 2015 with what a regulatory filing confirmed in 2019.
Analysis & classification — narrative arc scoring against the six-phase transformation arc, plus systematic classification across 16 attributes and signals, with documented rationale for each assignment.
Writing — first-draft profiles, sector spotlights, and report sections at defined register standards: Economist in English, Сноб/Коммерсантъ in Russian, 财新/三联生活周刊 in Chinese. Analytical, not promotional. Every claim sourced. Every signal evidenced.
Translation — the trilingual layer, calibrated by native-language editors for register, idiom, and cultural fit. Before AI, this was economically impossible at scale. It is now table stakes for any platform that takes founder trust seriously.
Inspection — a systematic second-pass audit over every finished profile and report after human editorial sign-off. Not to generate content. To inspect it: factual inconsistencies between sections, claims without source support, meaning drift between language versions, micro-errors in trilingual PDFs. The analogy is machine vision in precision manufacturing — sensors catch what human attention cannot sustain at scale.
Location intelligence — maps built around explicit arguments about what the geography reveals, with automated colour-blind accessibility review before publication.
Web layout — maintaining consistent formatting, taxonomy tagging, and cross-language linking across hundreds of profiles and reports simultaneously.
PDF production — compiling institutional reports with consistent layout, typography, citation structure, and design standards across all language versions. Publication-ready documents, not formatted Word files.
The human role
AI handles volume, language access, systematic consistency, and auditing at scale. Humans make the calls that AI cannot — and that institutional buyers require a human to have made.
Inclusion decisions: does this brand meet Brandmine’s threshold? Is the crisis genuine or cosmetic? NDD gate review: does the evidence satisfy the four-element test? Source arbitration: when sources conflict, which version is credible? Register calibration: does the Russian version read as native? Final sign-off: every profile and report Brandmine publishes carries human editorial approval.
AI inspects. Humans decide.
The full methodology
We have published the complete methodology on the Brandmine site — what each function does, how the research pipeline works, why trilingual publication is a condition of the relationship rather than a product feature, and what the two-layer review process looks like in practice.