Use case · Contracts
Contract intelligence that you own, not rent.
A token factory shape for contract review, extraction, and comparison — built for organisations where the contract is the liability, not a convenience document.
The problem
What we are actually solving.
Contracts are the paperwork of consequence. Master services agreements, data processing addenda, framework agreements, ISDA schedules, supplier terms, licensing deals — the organisation runs on them, legal owns them, and almost everyone else has to read them occasionally under time pressure.
The obvious AI pitch — “upload your contracts, ask questions” — is the wrong problem framed. The real problem is that contract work is adversarial-grade: the cost of a confident wrong answer is materially higher than the cost of a slower right one. That rules out vendor chatbots that cannot show their work, and it rules out architectures that cannot be audited end-to-end.
For Nordic banks, insurers, and DORA-scoped financial entities, there is an extra constraint: contracts with ICT providers are themselves a regulated artefact. Which contracts exist, what they say about sub-contracting and exit, and whether they comply with the DORA Article 30 checklist — those questions need defensible answers, not plausible ones.
The approach
The shape of the token factory for this use case.
The context engine ingests the canonical contract corpus with clause-level chunking. Headings, numbered sections, defined terms, schedules — the document’s own structure dictates the retrieval units, not an arbitrary token window. Every chunk carries provenance: which contract, which version, which clause, signed when, by whom.
The prompt compiler maps legal intents — “extract termination triggers,” “compare liability caps across these five agreements,” “flag clauses that drift from our standard template” — to provider-native payloads. Intent is a first-class concept because legal work is intent-shaped: the same question asked of ten contracts is ten different retrievals, not one.
The router picks a model per intent class and per stakes level. A low-stakes summary can run on a cheaper long-context model; a high-stakes extraction that will end up in a regulatory filing runs on the current frontier model with a stricter eval gate.
The eval harness here is the load-bearing part. A contracts token factory without a gold set of legal adjudications is a liability generator. The gold set is built once with the legal team, replayed on every model change, and owned by the client.
Outcomes
What changes.
The shift you should expect, if the factory is built well, is from “ask legal for a summary” to “ask the factory, then escalate the hard third to legal.” The win is not raw speed; it is a higher fraction of legal’s hours spent on questions that actually need legal judgement.
The trade-off is honest: a contracts factory is the most opinionated token factory to build, because it touches the organisation’s risk register. Expect a longer discovery phase, a stricter eval bar, and a shorter list of intents in the first production slice than a generic “chat with your documents” vendor would promise.
Where regulation lands. DORA Article 30 requires specific clauses in ICT contracts and a live register of information. The EU AI Act places contracts-review AI above the minimal-risk tier once it materially informs decisions on individuals or on regulated services. A token factory gives you the auditability both frameworks assume — citations back to the clause, logs of which model answered, a versioned eval set — without buying a separate compliance product.
Honesty
Expert approach based on adjacent delivered work. I have not shipped a production contracts token factory under Blue Abraham; the approach above is how I would build it given what I delivered in aviation technical documentation. See Technical documentation for the delivered parallel.
The aviation engagement and a contracts engagement share the same load-bearing constraints: high-stakes domain, regulatory scrutiny, long-lived corpus, traceability non-negotiable. The differences are in the intents and the legal review loop, not in the architecture.
Next step
Talk about contract intelligence that you own, not rent..
The usual first engagement is an architecture review. Say so on the form, or write directly.
Get in touch