Use cases · Finance & admin

How it works

This is a batch job that runs overnight; it cannot answer while you wait. Where the existing clause-finder reads one contract at a time, this holds your whole estate in view at once — every supplier, lease, insurance and SaaS contract, with their amendments and side letters — and answers questions that cross documents: liability caps that stack, indexation clauses that compound, a cluster of auto-renewals falling in the same month. Reasoning across a whole estate is a job for a very large model — too big to fit on a graphics card at all. So its bulk sits instead in the much slower system memory of ml2, and only a small part of that model works on any one word. That is what lets it get through the estate overnight rather than while someone waits.

Data you need

Every contract you want read together as digital text — Word files or text-layer PDFs — with each amendment and side letter gathered alongside the contract it changes, or a superseded clause gets read as if it still stood. Contracts that exist only as scans or photos must go through a separate OCR step first, since no vision model is served here, and OCR mistakes carry through into the answers. A contract sitting in a filing cabinet does not exist for this system at all.

What to expect

The cross-document questions are the whole point, and they are also where it goes wrong in ways worth naming. It will state a confident conclusion stitched from two clauses that never actually interact — reporting a stacked liability cap where the second contract's cap was never engaged — so treat every cross-document claim as a lead to verify, not a finding. An amendment filed apart from its parent contract gets missed, and a cap that a side letter raised will be reported at its old value; compounding indexation is arithmetic, and arithmetic is where these models slip, so read the clause and work the figure out yourself. It has no view of your national law. On a dense Dutch or German clause it will hand you fluent English that quietly changes the scope — a limitation-of-liability carve-out read as absolute, a conditional obligation read as unconditional — and the English reads clean enough that the error stays invisible until someone checks it against the original language.

Where people stay involved

A person owns every answer and signs nothing on the strength of it — this points you at the clauses to check; it is not a compliance review or legal advice. Read each cross-document conclusion back against the quoted clauses in the original, and for anything with money, liability or a termination deadline attached, that reading is your lawyer's job.

• Pull the details off supplier invoices into your bookkeeping

  Pulls supplier name, invoice number, dates, line items and VAT out of PDF invoices into records your accounting software can import.

• Turn expense receipts into draft expense entries

  Turns OCR'd receipt text into draft expense entries — merchant, date, amount, VAT and likely category — for the submitter and finance to confirm.

• Sort expenses into your bookkeeping categories

  Reads cryptic bank and card descriptions and proposes a category from your own chart of accounts, backed by your past bookings as evidence.

• Match incoming bank payments to open invoices

  Proposes matches between bank statement lines and open invoices when references are mistyped, names differ or one transfer covers two invoices.