Partner Contract Intelligence · Documentation

Business Problem & Solution

The problem Partner Contract Intelligence solves in the hospitality ecosystem and how this downscaled demo proves it.

Partner Contract Intelligence
partner contractsthe systemtriagescanned PDFdigital PDF / Worda six-clause bankdual-perceptionOCR · clause-risk · triagecommission raterisk score + bandamendment registerreview only what won't clear
Live diagram — the document format is the cost lever; only scanned contracts pay for the OCR stage. A reviewer spends time only on contracts that won't clear.

The problem — partner agreements pile up

A marketplace signs contracts with every partner property: commission agreements, rate-parity terms, service-level agreements, marketing co-ops, data-processing addenda. They arrive in every format — clean digital PDFs, Word files, and scanned images of a signed paper copy — and each one hides the terms that actually matter: the commission rate, the cancellation exposure, an uncapped liability clause, an exclusivity lock-in, an expired term running month-to-month. Reading every clause of every contract by hand is slow, and the risky ones are exactly the ones a busy reviewer skims past.

Partner Contract Intelligence reads the contract for you. It extracts the commission rate, scores each clause low / medium / high risk, rolls that into a composite risk score, assigns a triage band — auto-accept, review, or exception — and builds an amendment register of what to renegotiate. A reviewer then spends their time only on the contracts the system could not clear on its own.

What this demo proves — and what it simplifies

It proves a dual-perception pipeline: a scanned contract goes through an OCR vision stage before clause analysis, while a digital PDF or Word file skips straight to analysis — so the format of the document is the cost lever, and only the contracts that genuinely need vision pay for it. It proves an honest triage: a deterministic clause-risk and confidence floor decides the band, and the metered model read informs the confidence rather than overriding the rule. And it proves the coupling: the moment a contract is analyzed, the result feeds the partner's performance scorecard (#13).

It simplifies by using synthetic contracts. The clause text is authored boilerplate across a six-clause bank; there is no real or copyrighted contract anywhere. The model read is downscaled — the prototype runs the recorded OSS model at $0 by default — and the commission figures, confidences, and risk scores are representative and labelled.

Reality contract

Synthetic data only — partners are derived from the shared synthetic properties, and each carries one to three contracts across the five kinds and three formats. No real partner, no real contract, no PII. The two AI stages run dual-mode: a cost-capped, fail-closed cloud call (claude-haiku-4-5) or a recorded OSS model on local hardware at $0; the prototype default is OSS-only. All cost, mode, and token figures are representative and live only in the inspector.

Business Problem & Solution · Partner Contract Intelligence · Abhishek Saxena