Your contracts keep spending after you sign them

Somewhere in your company's shared drive there is a folder with a name like Contracts, Final. (It is neither.) Inside it are a few hundred PDFs that everyone treats as finished objects: negotiated, signed, filed, done. Then a 12-month renewal invoice arrives, 40 percent larger than last year's, and someone opens one of those finished objects to find out why.

The why is on page 14. Auto-renewal unless notice is given 60 days before term end. The notice window closed three weeks ago. Nobody missed it through carelessness; they missed it because no system on earth was watching it. The renewal date lived in a PDF, the calendar lived everywhere else, and the contract executed its default exactly as written.

A signed contract is a program that keeps running. Most companies close the file the day it compiles.

The renewal is the famous version, but the same folder leaks in quieter ways. Escalator clauses compounding a point above what anyone budgeted. A vendor applying seven percent where the negotiated cap says three, unchallenged, because who cross-checks an increase against clause 6.4 of a three-year-old amendment. SLA credits earned and never claimed. Volume discounts that trigger at thresholds nobody monitors. Signed leverage, unclaimed, expiring quietly on a schedule.

Most legal or finance teams have tried the obvious fix: an intern, a spreadsheet, a heroic quarter of reading. It works until the week it stops being maintained, which is usually the week the intern leaves. The failure is structural. Obligations do not live in one document; they live in a chain. The 2022 MSA sets a renewal term, the 2024 amendment changes it, order form three modifies pricing for one product line only, and the answer to "when can we exit this" is a join across four documents that were never designed to be joined.

Reading one contract is a task. Maintaining a truthful, current picture of four hundred contract families as amendments land is a data pipeline, and pretending otherwise is how the spreadsheet quietly becomes fiction.

The method is to treat the executed portfolio as a dataset and extract it into a register: one row per obligation, typed and dated. Renewal date, term length, notice window, auto-renew flag, escalator formula and cap, termination rights, SLA remedies, data-return duties. Each row carries a citation back to the exact clause and page it came from, because a register legal cannot verify in one click is a register legal will not trust, and an untrusted register is the spreadsheet again with better branding.

Two properties separate a register that works from a demo. First, amendment supersedence: extraction has to resolve the document chain, so the register reflects what the relationship currently says, not what the original said. Second, calibrated abstention: where the language is genuinely ambiguous, the system flags the clause for human reading instead of guessing, which is the difference between a tool lawyers adopt and one they audit once and abandon. Downstream of the register, the rest is almost boring: notice-window alerts with named owners and enough lead time to actually negotiate, invoice checks against the escalator formula, a renewal calendar that finance sees before the quarter starts rather than inside it.

We built a register like this over a portfolio in the mid hundreds of contracts. The first automated pass surfaced a dozen auto-renewals landing inside the next quarter and two vendors billing above their negotiated caps, one of them for over a year. The recovered overbilling alone was a meaningful multiple of what the extraction work cost.

Renewal leverage has an expiry date, and it is always earlier than the invoice.

The honest limit: extraction does not negotiate, and a register full of alerts nobody owns is a nicer way to miss the same deadlines. The system buys back the leverage; a human still has to spend it.

Pick your ten largest vendor contracts and answer three questions from memory: renewal date, notice window, this year's permitted increase. Most executives can answer for two of the ten. The other eight are running unattended.

TensorLabs builds these extraction pipelines, clause-grounded, amendment-aware, wired to alerts that arrive while options are still open. If your contracts folder has ever surprised you with an invoice, write to us and say roughly how many agreements are in it. That number sets the size of what is recoverable.