Case study · Market Intelligence

A Self-Hosted Weekly Market Intelligence Digest for a Neuroscience Instrumentation Manufacturer

Live in 6.5 weeks against eight quoted, with 100% of claims source-attributed and zero hallucinated citations.

Scientific instrumentationSelf-hosted · weekly cadence
6.5 wksto first live digest
100%claims source-attributed
0hallucinated citations
01 · The situation

The situation

A UK manufacturer of electrophysiology instrumentation — patch-clamp and in vivo rigs sold into research institutions across five markets. The Managing Director already had a daily AI digest from an off-the-shelf assistant, and it failed on the things that mattered: no sources, no way to tell a quiet week from a broken pipeline, and every signal reported as an event rather than read against the market it happened in. A fall in Chinese sales that reads as demand collapse in London is channel substitution in Shanghai.

The requirement was narrow and testable. One weekly digest, every claim traceable, every movement interpreted where it occurred.

02 · What we built

What we built

  • Funding coverage across eight public bodies — NIH, NSF, UKRI, Wellcome, Horizon Europe, ERC, DFG and ANR — with per-award reads and eight-quarter trajectory by route. DFG publishes no API, so it was solved with a verify-then-publish subsystem
  • Competitor monitoring of eight companies across two clusters, tracking launches, funding, partnerships, distribution, regulatory and leadership moves, each run through a five-step strategic “so what” chain
  • Trade, regulatory and publication signal — ECJU, USTR and OFSI cross-cut against export channels, regulatory discovery across five territories, and paper-volume trajectory from PubMed, OpenAlex and Europe PMC
  • A territory lens: five market profiles that re-frame every finding in local context — industrial policy, buyer dynamics, local competition, trade exposure
  • An entity-verification gate that screens out same-surname look-alikes before anything reaches the digest
  • A self-hosted pipeline on UK infrastructure — the AI models run on dedicated machines we manage in the UK, not on a third-party AI service — with eight independently replaceable modules, a Qdrant vector store, synthesis on a UK-resident model with no third-party model API in the production data path, and a reply-to-approve gate between the Sunday pipeline run and the Monday 08:00 send
03 · The outcome

The outcome

Live in 6.5 weeks against eight quoted, with 100% of claims source-attributed and zero hallucinated citations. Every issue ships with an audit companion, so the digest can be checked rather than trusted. Two hundred findings were read in-market across five territories — and where a framing would be wrong, the system degrades to “no reframe” rather than inventing one. The entity gate screened fourteen same-surname false matches out of the MD’s inbox. A quiet week is now distinguishable from a low-signal one, and slow-building trends surface instead of “nothing changed”. Confidential documents never leave the controlled host; only the derived digest transits the mailer.

Market intelligence systems of this type start at £7.5k to build. Running it costs from £200 a month, which covers the dedicated AI computing power that produces the digest each week plus ongoing operation and improvement. Learn more about Market Intelligence.

Facing something similar?

Tell us what you are dealing with. We will tell you honestly whether AI helps.

Talk to us →

← All case studies