Operated function · contract obligations
It took two months and several expensive people to negotiate. It contains a notice period, an annual price review, a reporting duty and an audit right — each of them dated, each of them now depending on somebody happening to remember at the right moment.
It gets filed. The negotiation team moves on, correctly, because negotiating is their job and the deal is done. Nobody is assigned to the document as an ongoing operational object, and there is rarely a role whose title implies it.
The commitments inside it are dated and the dates are not anywhere. A notice period that requires ninety days is a date; an annual price review is a date; a reporting obligation is a repeating date. All of them live in prose, in a file, and none of them is in a calendar anybody looks at.
Auto-renewal is the one that costs most quietly. A contract renews because the notice window passed unnoticed, and the business is committed for another term on terms it might not have chosen — which is discovered when somebody asks why the invoice is still arriving.
Obligations flowing in the other direction go unclaimed. A right to audit, a service credit, a benchmark review, a volume discount that should have triggered — these are money the business is owed and does not collect, because collecting requires knowing they exist and when.
And when a question arises — what did we agree about liability, is there a change-of-control clause, what is our exit right — the answer requires somebody to find the document and read it, so it takes a day and gets asked only when it is already urgent.
A contract’s commitments exist as sentences inside a document rather than as dated records with owners, so every one of them depends on a person remembering to look — and nobody is assigned to look.
Converting the prose into records is the whole operation, and it is bounded work: read the executed document, identify each obligation and right, capture what it requires, of whom, by when, and how often. That is tedious, high-volume, and exactly what should be delegated.
What it produces is a register, and a register is a fundamentally different object from a filing cabinet. It can be sorted by date, filtered by owner, and asked questions — which terms across our supplier base cap liability, which contracts auto-renew in the next quarter, which give us an audit right we have never used.
The dates then behave like every other dated obligation in an operating business: they surface before they matter, to a named owner, with the clause text attached so the person can see exactly what is required rather than being told something is due.
The rights half is where the money usually is. Businesses are consistently better at meeting their obligations than at claiming their entitlements, because obligations have a counterparty who chases and entitlements do not.
What is never delegated is interpretation. What a clause means, whether an obligation has been met, whether to serve notice and whether to exercise a right are legal and commercial judgements belonging to your own people.
Obligations that exist as dated records with owners — measured by extracted obligations against the contract population, from a baseline where they existed only as prose.
Auto-renewals that happen by decision rather than by lapse — measured by notice windows surfaced before expiry, against renewals discovered after the fact.
Rights actually exercised — measured by entitlements claimed — credits, reviews, audits — against a baseline where they were rarely surfaced at all.
Time to answer a clause question — measured by minutes to retrieve a term across the population, against a baseline requiring somebody to find and read a document.
Obligations discovered already breached — measured by the count found late, before and after the register exists.
Ability to compare terms across a population — measured by contracts whose key terms are queryable, which is what makes a renegotiation strategy possible at all.
any legal interpretation, any assessment of whether an obligation has been met, and any advice on what a clause means. Nothing here is legal advice. It extracts what a document says into dated records and surfaces them — what those words mean and what to do about them belongs to your lawyers and your commercial owners.
Executed documents stay where they are. The register points at them and carries the extracted obligations, so there is one authoritative document and one derived index rather than two documents.
The register is yours: your obligation taxonomy, your owners, your thresholds. It is the asset the engagement produces and it keeps working afterwards, which matters more here than in most operations because a contract register accumulates value for as long as the contracts run.
Every extracted obligation carries a pointer to the exact clause it came from, so an owner acting on it reads the source rather than a summary. A summary is where an interpretation error enters.
Extraction is not interpretation, and the boundary is maintained by always attaching the source clause. An owner sees what the contract says, not a paraphrase of what somebody thought it meant — a paraphrase is how a ninety-day notice becomes a three-month one in somebody’s head.
Where a clause is ambiguous, unusual or does not map cleanly onto an obligation type, it is flagged for legal rather than fitted into the nearest category. A forced classification is worse than an unclassified clause, because it looks resolved.
Contracts contain commercially sensitive terms — pricing, exclusivity, liability caps, sometimes terms your counterparty would not want compared against another’s. Access is scoped and no document leaves your systems.
Operational access is not permission to train. Your contract terms — which collectively describe your entire commercial position — do not become material improving anything serving another organisation, including counterparties who negotiate against you.
General counsel will want the extraction-versus-interpretation boundary in writing, and the source-clause pointer is what makes it real rather than asserted. An obligation register that summarises without citing is a liability.
Commercial leadership owns the obligation taxonomy and the ownership assignment. An obligation with no named owner will not be met regardless of how well it was extracted.
Where an obligation attaches through confidentiality terms in your own contracts — several of which will restrict who may read them — that is established at scoping, because it may narrow the population in scope.
A single contract family — one supplier category or one customer tier — read-only, with obligations extracted into a register and nothing acted on.
The first phase produces the register for that family and nothing else. It acts on nothing, advises on nothing, and typically surfaces two things immediately: notice windows inside the next two quarters that nobody had in a calendar, and rights the business holds and has never exercised.
Both are usually enough to justify continuing on their own, and both are findings your commercial team can act on without any ongoing operation.
If you continue, extraction extends across the population and dated obligations begin surfacing to named owners — with interpretation, notice decisions and any exercise of a right remaining entirely with your people.
Most hold documents and metadata well — parties, value, end date. The gap this addresses is the obligations inside the prose: the notice period, the reporting duty, the price-review trigger, the audit right. If your system already holds those as dated records with owners and source-clause pointers, this is a duplicate. The first-phase extraction on one family settles it quickly.
Interpreting them is, and that is not delegated — nothing here says what a clause means, whether an obligation is met, or what to do. Extraction captures what the document states and attaches the clause so your lawyer reads the source rather than a summary. Where a clause is ambiguous it is flagged rather than classified. If your position is that even extraction requires a qualified reader, the register can be built by your team using the taxonomy, and the surfacing and ownership still apply.
Non-standard is the normal condition and it is why ambiguity is flagged rather than forced into a category — a forced classification is worse than none because it looks resolved. The first phase measures directly what share of one family extracts cleanly and what share needs a lawyer. If most of it needs a lawyer, the honest scope is much smaller and possibly limited to date capture, which is still the thing preventing silent auto-renewals.
The long tail is usually where the unclaimed rights and the silent renewals are, so that is not an argument against it. It is also worth testing the premise: the register on your key contracts frequently surfaces an entitlement nobody had exercised, because the people who negotiated it moved on and the right had no owner. That test costs one family and acts on nothing.