Transparency
This page answers the question we'd want answered before paying for something like this: how much of it is AI, and who's actually accountable for what you receive.
Yes, this is AI-assisted. Plainly, not as a footnote.
Tanssor Research runs on a pipeline of AI systems that do the reading, structuring, drafting and self-checking at a volume manual review couldn't match — see About. That's the basis for the business existing at €349 a Note rather than a day-rate. We'd rather say this outright than let a professional-looking page create the wrong impression about how a Note gets made.
What actually happens, step by step
An automated reader pulls official award notices for the buyer or tender in question — TED, national registers — and structures the facts into a typed evidence set. Nothing narrative happens at this stage.
An AI model writes the Note from that structured evidence, constrained to cite specific facts rather than freely stating numbers or dates from general knowledge.
At least two AI models, from different providers, independently review the draft against the evidence set and try to find what's wrong: unsupported claims, weak inferences, missing context, overconfident wording. Neither is the model that wrote the draft, and they don't see each other's verdicts. Drafts get revised against this feedback, sometimes through several rounds.
The Note that comes out of that process goes through mandatory analyst sign-off — source verification against the underlying tender documentation — before it's sent. This is not a formality: Note #001 went through multiple revised versions before publication, with the change printed in the Note itself as a correction note. Nothing reaches a customer without this check.
What we won't claim
- We won't tell you a human analyst independently re-derived every number by hand — they didn't; the evidence structuring is mechanical by design, which is what makes it reliable rather than what undermines it.
- We won't claim zero error rate. We claim a labelled confidence level per claim, a stated verification path, and a public correction record when we get something wrong — see Methodology for a live example.
- We won't claim broader sector expertise than the sources actually give us. A Note reflects what the official record shows, not private knowledge of a market.
Known limitations, honestly
- Official records are incomplete by nature — below-threshold spend, informal call-offs, and some national identifiers simply aren't published anywhere we can reach. Where that happens, it's flagged as a gap in the Note, not silently omitted.
- Some public registries don't cover certain buyer types (e.g. state institutions established by their own founding legislation sometimes have no commercial-registry identifier at all) — again, flagged rather than guessed at.
- The pipeline is continuously refined based on adversarial review critique — the process above is real and running today, not a demo or a mockup. Every Note ships with a full refund guarantee, so you're never taking a risk on us; we're taking it on ourselves.