The client is Tivaco, an expert firm valuing damages and assets for international arbitrations. Top-5% damages experts globally, $15B+ in valued assets. Highest claim in the portfolio — $8B+.
The valuation targets are private (non-public) companies. That's the key constraint: they have no direct market multiples — peers must be picked among public companies in the same geography and sector.
Before us, the work was done manually by senior analysts — $50/hour+, several days minimum per document. Financial statements run hundreds and thousands of pages, and the same metric can have several values within a single document.
The task — reliably extract financial metrics for specific dates, resolve collisions, and pull in market context.
Extraction pipeline built around the structure of financial statements: PDF → sections → tables → values with source context. Tables, body text, and footnotes are processed together.
Collision resolution across multiple values found for the same metric on the same date — with reasoning explaining why a particular variant was chosen.
Temporal dimension: one metric across historical / current / forecast slices, on specific dates. Technically harder than «just pull the number»: a single metric exists in multiple slices simultaneously.
Public peer selection by geography and industry, with multiples gathered (cost / market / income approach).
Analyst UI. Every value comes with a quote from the source PDF, an explanation of the choice, and a link to the specific page. The analyst can verify, edit, accept.
The human stays the source of decision — the machine produces a defensible draft. This is critical for high stakes: every number ends up in an arbitration.
Send a brief description of what you have. From there — in correspondence: we get the context, scope the boundaries, and agree on a PoC in days–weeks.
