AI Contract Intelligence: What Is Hiding in the Portfolio Company's Contracts

1. Where Value Leaks in a PE-Owned Company

Every company is, legally, the sum of the contracts it has signed. Here is the uncomfortable part: in most portfolio companies, nobody currently employed has read them all.

The purchase agreement got read forty times before close. The company's own paper, the customer agreements, supplier terms, leases, and licenses that produce the actual cash flow, mostly got read once, years ago, often by people who have since left.

That gap has a price. It shows up as auto-renewals that lock in another year of a bad vendor without anyone deciding. As price escalators the company negotiated, wrote down, and never once invoked. As change-of-control clauses that surface for the first time in the exit data room. As rebate and volume terms with the largest customer that nobody has tracked since the person who agreed to them departed.

A price escalator you never invoke is a discount you never approved. Multiply that across a few hundred agreements and contract intelligence stops being a legal chore. It becomes a value program with a number attached.

2. The Post-Close Contract Sweep

In diligence, the deal team read the twenty contracts that mattered to the thesis. That was triage, and under a deadline it was the right call. It was not a reading of the company. The deal-side discipline is its own subject, covered in AI for legal diligence and contract review. This guide is about what happens after you own the paper.

The first hundred days are when the full sweep belongs. Ownership just changed, so change-of-control questions are live. Integration decisions are being made. Management is already pulling files for lenders and auditors. You will never have a cheaper moment to gather every agreement in one place.

The sweep is simple to describe. Collect everything: customer agreements, supplier and distribution terms, leases, licenses, employment agreements and non-competes. Then extract the same fields from each. Parties. Term and renewal mechanics. Notice windows. Price terms and escalators. Change of control and assignment. Termination rights. The obligations the company signed up to deliver.

The output is not a memo. It is a table: one row per agreement, one column per field, each cell linked back to the page it came from. Sortable by renewal date. Filterable by clause. The rest of this guide is about what falls out of that table.

3. Revenue Hiding in the Paper

Start with the money owed to you. It is the fastest way to make the sweep pay for itself.

Escalators never invoked. Contracts routinely include annual price increases, a fixed step-up or an index link, that the company never billed. Sales negotiated the right, finance never operationalized it, and every anniversary that passes quietly is margin given away. The sweep finds the entitlement; the billing file shows whether it was ever taken.

Unbilled entitlements. Pass-through costs the contract allows and the invoices never included. Chargeable extras delivered free because nobody checked the schedule. Setup fees waived once and never revisited. None of these are new revenue. They are agreed revenue the company forgot to collect.

The renewal calendar. A renewal you see ninety days out is a negotiation. A renewal you discover after the notice window closed is a term you accepted by default. The most used output of a contract sweep is usually this calendar: every renewal and notice deadline for the next eighteen months, with owners. The findings land with finance, which is why this work pairs naturally with the finance and back-office playbook: extraction finds the entitlement, billing has to collect it.

4. Risk Hiding in the Paper

The risk side is quieter, and it compounds toward exit.

Change of control and assignment. These clauses decide which customer relationships and licenses survive the next transaction. Found three years early, they are a negotiation item at the next amendment. Found in the exit data room, they are a price adjustment, because by then the buyer's lawyers found them first.

Exclusivity and most favored nation. An exclusivity granted to a distributor years ago can quietly block the add-on you are underwriting now. An MFN clause means one aggressive discount to one customer can reprice the whole book, and the customer holding the clause is entitled to go looking.

Termination for convenience. Revenue under a contract the counterparty can exit on thirty days' notice is not backlog, it is a habit. Boards routinely treat contracted revenue as harder than the paper supports. Knowing which share of the book is genuinely committed changes how much to trust the forecast.

5. How AI Extraction Works, and Why Verification Is the Feature

The reason this is now a hundred-day project instead of a two-year one is that reading at scale got cheap. Current models read full agreements, scanned pages included, and pull each clause into a structured record with a citation to the exact page it came from.

The mechanics matter less than the discipline around them. Extraction runs the same clause list across every document, so agreement forty-seven gets the same attention as agreement one. That consistency is the point: a tired associate skims, a model applies the same checklist to every page. It is the same shift already underway on the deal side, where CIM data extraction turned document reading into structured data.

Then the step that separates a serious deployment from a demo: verification. Every extraction carries a confidence score. Anything that touches money or risk, an escalator amount, a change-of-control trigger, a liability cap, gets checked by a person against the cited page before anyone acts on it. The citation is what makes checking fast: a thirty-second look at the source paragraph instead of a hunt through the file.

Be honest about the edges. Dates, parties, terms, and standard clauses extract reliably. Heavily negotiated provisions, indemnity interplay, definitions that reference other definitions, still need a lawyer's read. AI turns ten thousand pages into a short list of pages worth a lawyer's time. That is the win.

6. Roll-Ups: Harmonizing Contracts Across Add-Ons

If the thesis is a roll-up, contract intelligence stops being optional. Every add-on arrives with its own paper: its own supplier terms, its own customer templates, its own inherited surprises.

Five acquisitions in, the platform is buying the same materials under five different price schedules and selling the same service under five different liability regimes. Nobody planned that. It is what happens when integration focuses on systems and people and lets the contracts ride.

A per-add-on sweep, run as part of onboarding each acquisition, makes the comparison possible. The best supplier terms across entities become the target for the rest. Conflicting exclusivities surface before they collide. Customer templates converge toward the strongest version. Procurement synergies are usually modeled on volume; the contract file is where you find the rest of them.

7. Buy or Build: Three Ways to Get Contract AI

Three categories cover the real choices. They solve different problems, and the common mistake is buying one when you needed another.

CLM platforms. Full contract lifecycle systems: a repository, approval workflows, templates, signatures, reporting. The right answer when the company's problem is process, meaning contracts get created inconsistently and stored nowhere in particular. The trade-off is weight: implementation takes months, and the value depends on everyone actually using it.

Document-AI extraction tools. Point them at a folder of agreements and get structured data out. Lighter, faster to stand up, and well suited to the one-time sweep and the roll-up comparison. They answer questions about the contracts you have; they do not manage how new ones get made.

Custom review built on Claude-class models. A build shaped to your clause list, your playbook, and your reporting, using frontier models under commercial terms. The most flexible option, and the only one that molds to how the sponsor actually asks questions. It needs an owner and a verification design, which is why it works best inside a broader AI program rather than as a one-off.

The deciding question is simple: does the company need a system of record for contracting, or answers from the contracts it already has? Process problem, CLM. Knowledge problem, extraction or a custom build.

8. The Clause Map

If you only extract seven things, extract these. Each row is a clause type, why the sponsor should care, and what the extraction should capture.

The list is deliberately short. A sweep that tries to capture everything captures nothing reliably. Start with the seven, verify them well, and extend the list once the first pass has earned trust.

9. Security and Privilege

Contracts concentrate the company's most sensitive facts: customer identities, pricing, employment terms, settlement history. Treat the tooling accordingly. Commercial-grade AI accounts, meaning Team, Enterprise, or API terms that do not train on your data, never a consumer login, with access limited to people who could already open the files. The secure AI adoption groundwork covers exactly this setup.

Privilege deserves its own paragraph. Extraction for business operations, renewal dates and price terms, is ordinary business analysis. The moment a question points toward a dispute, an actual or suspected breach, a termination fight, bring counsel in and let them direct the work, because material prepared under counsel's direction is treated differently from material that is not.

The operating rule is the one that applies to AI everywhere in the company, with a sharper edge here. AI reads and flags. People verify. Lawyers decide what a clause means and what to do about it.

10. The 90-Day Plan

Days 1 to 30: gather and aim. Collecting the paper is genuinely the hardest step. Agreements live in inboxes, shared drives, and a filing cabinet near the controller. Pick the clause list, the seven above are the starting point, and run extraction on the segment that matters most, usually the top customers and suppliers by spend.

Days 31 to 60: verify and monetize. Human-check every extraction that carries money or risk. Build the renewal calendar and put owners on the next two quarters of notice windows. Hand the escalator and unbilled findings to finance with the source pages attached.

Days 61 to 90: complete and wire in. Extend extraction to the full population. Decide the permanent home, a living table, or a CLM if the process problem is real. Wire the alerts: renewals surfacing ninety days before their notice windows, new contracts entering the table as they are signed.

The goal at day ninety is not a report. It is that the next renewal arrives as a decision instead of a surprise.

11. Where to Start

Start smaller than the plan above if you must, but start where the money is: the top twenty customer contracts, the renewal calendar, and change of control. Those three answers justify the rest of the program, and each is checkable within weeks.

For the sponsor, the same discipline scales. The sweep is repeatable at every company you own, and the second run costs a fraction of the first because the clause list and the verification playbook already exist. Where contracts sit among everything else AI can do inside an operating company is mapped in the portfolio company AI playbook.

If you want this built rather than described, Contract Intelligence is the scoped version of everything above: every agreement read, extracted, verified, and delivered as a table the company can query. And if the honest question is whether contracts are the first thing to fix or the third, a Portco Value-Creation Diagnostic ranks them against every other AI opportunity in the company before you commit to any of them.