TL;DR
It reads each contract against your own playbook rather than general legal knowledge, scores every clause, and puts the result in front of a reviewer in Word.
It checks each incoming contract against your own playbook — not against general legal knowledge — and hands a reviewer the clauses that fall outside it, ranked by risk.
That distinction is the whole product. A model that knows contract law in general cannot tell you whether this payment term breaches your policy. A system that has read your playbook can.
What is the actual problem in procurement?
Not difficulty. Volume multiplied by consistency.
A procurement team receives supplier agreements, renewals, statements of work and amendments continuously. Each one has to be read against the same internal rules — liability caps, payment terms, indemnity, termination, data protection. The rules do not change. The reading does.
Two things go wrong at volume, and both are ordinary:
- The tenth contract on a Thursday is not read like the first on a Monday. Not carelessness — fatigue, and a queue.
- Two reviewers apply the same policy differently, because a playbook written in prose leaves room for interpretation, and interpretation drifts.
The cost is not the review time. It is the clause nobody flagged, which surfaces at renewal or in a dispute, long after the person who approved it has moved on.
What does the system actually review against?
Your playbook. This is the part most teams underestimate, and it is the part that decides whether any of this works.
Paralegent AI ingests the document your legal team already follows — the rules, the acceptable positions, the fallback positions, the red lines. It extracts those into structured terms, and those terms become the standard every contract is measured against.
Two consequences follow, and they are worth stating plainly:
- A good playbook produces good review. A vague one produces vague review. The system inherits the quality of the rules it is given.
- A playbook that already exists as a spreadsheet is the strongest starting point. Structured rules need no interpretation to import — what the system needs from your playbook covers why that matters more than it sounds.
Paralegent AI is our contract-review app — one of Cognilium's AI optimization apps for Microsoft Dynamics 365, and the one that is in production today. We build these against a customer's own playbook and their own Dynamics environment, which is why the rest of this article is about your playbook rather than our model.
Paralegent AI does not invent policy. It applies the policy you gave it, which is why the output is defensible in a way a general-purpose model's opinion is not.
How does a contract move through it?
Four stages, and each answers a different question:
| Stage | The question it answers |
|---|---|
| Playbook processing | What are our rules, as structured terms? |
| Contract review | Which parts of this contract relate to which rule? |
| Category scoring | How relevant is each passage to each area of legal risk? |
| Clause analysis | What is the risk here, and what should we ask for instead? |
The middle two are the ones that make it usable. A contract is long and most of it is boilerplate. The system narrows to the passages that actually touch a rule, then routes each one only to the specialist checks it needs — rather than running every check against every paragraph.
That routing is why it finishes in minutes rather than hours, and it is the difference between a demonstration and something a team can put a Thursday queue through.
Where does the reviewer actually sit?
In Word, where the contract already is.
This is a deliberate choice and it is worth explaining, because the obvious alternative — a web portal you log into — is what most products ship. A lawyer or contract manager works in the document. Asking them to leave it, review findings somewhere else, then come back and apply changes adds a step at exactly the moment attention matters most.
Paralegent AI puts the findings against the clauses, in the document, with the risk level and the suggested position. The reviewer accepts, edits or rejects — and stays where they were.
There is a real trade-off here rather than a free win, and redlining in the document rather than beside it is a genuine architecture decision, not a preference.
What does it not decide?
Anything. And that boundary is the point rather than a limitation.
The system produces a risk assessment and a suggested position. A person decides. It does not sign, does not send, does not commit the organisation to a term, and does not silently accept a clause because it scored well.
Three things follow that a procurement lead should insist on:
- Every finding traces to a playbook rule. If it cannot name the rule it is applying, it should not be raising the flag.
- A low-confidence result should say so, not average itself into the middle. A number that looks like mild concern gets waved through on a busy afternoon.
- The human step is designed in, not bolted on. Review capacity has to exist before drafts arrive faster — that is true of every system of this kind, ours included.
Where does the ERP come into it?
Because in procurement, the contract is not a document — it is a record.
The agreement that governs a purchase order lives in the ERP. The supplier, the terms, the price agreement, the renewal date are all fields in Dynamics 365 or SAP, not paragraphs in a file. The signed PDF is the artefact; the operative version is the one purchasing acts on.
That creates the gap this cluster exists to close. Most contract review products treat a contract as a document and stop at the document. The interesting question is what happens next — whether a negotiated term reaches the record the buyer will actually use, or whether someone retypes it and hopes.
That question is CLM or ERP, and it is the one most worth getting right before you buy anything.
About Cognilium Cognilium builds AI optimization apps for Microsoft Dynamics 365 — companion apps that optimize the pricing, inventory, warehouse and planning decisions your ERP manages but can't optimize. Dynamics is your system of record. Cognilium is your system of intelligence. https://cognilium.ai · https://www.linkedin.com/company/37180269/
Legal AI Ops. We transform legal workflows with agentic AI, copilots, agentic workflows and decision intelligence — built into core workflows rather than beside them, to raise productivity and cut operational overhead. Contract Review Copilot is the contract-review app in that family. It ships as Paralegent AI, in production today. How we build Legal AI Ops — custom AI capabilities on top of legal work, against your playbook and your Dynamics 365.
Bring one real supplier agreement and your playbook to a 15-minute call. We will run the shape of it with you — including whether your playbook is ready.
Sources
Sources and fact-check
| # | § | Claim | Tier | Source | Verdict |
|---|---|---|---|---|---|
| 1 | 1 | The procurement problem is volume × consistency | T2 — ours, an argument, stated as one | Internal definition | PASS |
| 2 | 2 | The system reviews against the customer's own playbook, extracted into structured terms | T2 — capability, describing our own product | Paralegant_TECHNICAL_PROFILE.md §1 | PASS |
| 3 | 2 | A structured playbook imports without interpretation | T2 — ours, and §2 links to the article that evidences it | Same, §4.1 | PASS |
| 4 | 3 | The four stages, and what each answers | T2 — capability | Same, §2 High-Level Pipeline | PASS |
| 5 | 3 | Routing narrows work to the checks each passage needs | T2 — capability, stated qualitatively. No figure published — the internal reduction number is our own performance metric and is deliberately omitted | Same, §4.3 Smart Routing | PASS |
| 6 | 4 | Delivery is a Microsoft Word add-in, findings against the clause | T2 — capability | Same, §1 Delivery Method, §6 | PASS |
| 7 | 5 | The system assesses and suggests; a person decides | T2 — ours, and it is the disclosure boundary | 03-proof/DISCLOSURE-RULES.md | PASS |
| 8 | 6 | In procurement the operative contract is an ERP record — purchase agreement, terms, renewal date | T1 — verbatim | purchase-agreements (ms.date 2026-09-08), fetched 2026-09-10: "A purchase agreement is a contract that commits an organization to buy a specified quantity or amount by using multiple purchase orders over time." | PASS |
| 9 | — | Paralegent AI is in production | T2 — capability, and it is founder-locked canon. /dynamics-365/optimizers: "in production today as Paralegent AI; the rest are working demos" | External canon, 04-positioning-now.md | PASS |
Tier summary: 1 × T1, 8 × T2 — 0 × T4.
Disclosure check. No client is named or implied. No customer count, no measured outcome, no deployment claim. The only status claim is "in production", which is founder-locked canon and describes the product being shipped — not that it runs in any named business. The CTA is a private call with no self-inspection route.
Figures deliberately omitted. The technical profile contains our own pipeline metrics — call counts per stage and a routing-reduction percentage. None is published here. They are our own performance numbers, PUBLISHING-CONTRACT.md §1 governs them, and a percentage would additionally trip guard 6. The mechanism is described without them, which is the shape that survives.
