AI Drafts the Disclosure Schedules, and the 12-Hour Run Is the Point
Marveri, an AI platform for financings and mergers and acquisitions based in Cambridge, Massachusetts, has made its automated disclosure schedule drafter available to law firms. The tool produces a full first draft of the disclosure schedules for a financing or M&A deal from the transaction agreement and the full data room, cites every item to its source document, and exports to Word for the deal team to review and edit. It also runs public searches on the company covering lien filings, court and bankruptcy dockets, business registry status and intellectual property registrations, and folds the results into the relevant schedules.
Since a beta that opened in spring 2026, firms ranging from boutique deal practices to the Am Law 200 have used it on dozens of live deals. A single run can take up to twelve hours. The company states that attorneys need to verify every line, and that the draft flags items for review and places where the AI was unsure. Marveri sells the tool as a paid add-on. Marveri states it is SOC 2 Type II compliant and never trains models on customer data. The release names no customer firm, gives no time-saved figure, no accuracy benchmark, no pricing and no error rate.
What a disclosure schedule actually is
The significance of this product depends on understanding the document, so it is worth stating plainly. A transaction agreement contains representations: statements the seller makes about the company, such as that its contracts remain valid, that it owns its intellectual property outright, and that no litigation is pending. The disclosure schedules are where the seller qualifies those statements, listing the exceptions. A contract that stays valid except for these three, litigation in this one matter, intellectual property licensed rather than owned.
That makes the schedules a document with two properties that matter here. First, they are exhaustive by nature: a missing item is a misrepresentation, and a misrepresentation is a contractual liability rather than a drafting embarrassment. Second, someone builds them by reading everything. An associate preparing them works through hundreds of contracts, consents and corporate documents against dozens of representations, and the work is almost entirely search and comparison rather than judgment.
That is precisely the shape of task that rewards automation and punishes carelessness. It is also why the release’s framing about verification is not a hedge but the central design constraint, and why the mechanisms described around it matter more than the drafting claim.
The twelve-hour run is a deliberate inversion
One number in the release stands against everything the legal AI category has been saying. A single run can take up to twelve hours.
Every other product in this market sells speed. Draft faster, review faster, close faster. Marveri is describing a process that takes half a day, and it is doing so because reviewing an entire data room and then checking the work across multiple models is not a fast operation. The release says the tool checks its work across multiple models, which is a real technique: running the same task through several models and comparing outputs catches errors that a single model makes confidently, because models do not fail in the same places.
That approach is expensive and slow and it is the right one for a document where a missed lien is a liability. A twelve-hour run that catches an item a two-minute run would have missed is worth more than the time it costs, provided the human review it enables is faster than the drafting it replaces. That is the trade the product is making, and it is a more interesting bet than another latency claim.
Three mechanisms that make verification possible
The release states that attorneys need to verify every line in the schedules. That sentence is unusual and correct, and what makes it workable is three mechanisms described around it.
The first is per-item citation. Every disclosure links back to its source document, which means an attorney checking a line can open the contract and confirm it in seconds rather than searching the data room again. A draft an attorney cannot check quickly is not a draft; it is a liability with a timestamp.
The second is uncertainty flagging. The draft marks places where the AI was unsure. Most AI products present their output with uniform confidence, which forces a reviewer to apply equal scepticism everywhere and therefore to skim. Flagging the low-confidence items directs human attention where it is needed, and it comes closer to calibration than anything else a user can act on.
The third is the division of labour the release describes, in which the deal team spends its time on calls that need a lawyer rather than on reading. That is the honest productivity claim in legal AI: not that the machine does the work, but that it moves the human to the part of the work that requires a human. Connor Acle, the chief executive and a former deal lawyer, makes that point from experience, and Mike Dockery, a former corporate partner, frames it as time going to review and judgment instead of searching the data room.
The public records search is a different capability
One feature in the release deserves separating from the drafting, because it solves a different problem. Marveri runs public searches on the company covering lien filings, court and bankruptcy dockets, business registry status and intellectual property registrations, and incorporates the results into the relevant schedules.
An associate doing this manually works through a set of registries and dockets one at a time, and the work is both tedious and easy to do incompletely. A Uniform Commercial Code lien search in the wrong jurisdiction returns nothing and looks like a clean result. Automating the sweep and folding the output into the schedule addresses the failure mode that matters most in this document, which is not a wrong answer but a missing one.
It also changes what the product is. Drafting assistance makes an existing task faster. Automated public-record verification finds things a human might not have looked for, which is a different kind of value and the more defensible one commercially.
The liability question the release does not address
Disclosure schedules are representations to a counterparty, and an error in them is a contractual problem. That makes the question of who is responsible for a missed item the most consequential one in this product category, and the release does not address it.
The company’s position is clear enough in substance: the attorney verifies every line, so the attorney remains responsible. That is correct and it is also the arrangement every professional services firm expects. What the release does not describe is how the tool behaves when a reviewer accepts a flagged item without checking it, or what evidence exists afterwards that verification happened. For a product that firms with professional indemnity exposure buy, an audit trail of what was reviewed and by whom is the feature that turns a helpful draft into something a firm can defend.
Why this document class was left alone
It is worth asking why a task this well suited to automation went unautomated for so long, because the answer explains both the opportunity and the difficulty.
Legal AI has concentrated on research, contract review and drafting, and those are tasks where a plausible answer has obvious value. A disclosure schedule is different. Its output is a list of exceptions that a counterparty will rely on, so an incomplete answer is worse than no answer, and the failure is silent: a schedule that misses a lien looks exactly like a schedule that found none. That asymmetry is why the document attracted less attention than contract review, and it is why a product claiming to do it invites more scepticism than one claiming to summarise a case.
The release’s emphasis on mechanism rather than accuracy is therefore the right rhetorical choice even if it is an incomplete one. Naming the citation, the uncertainty flag and the cross-model check tells a sceptical buyer how the tool fails safely. It does not tell them how often it fails, and that remains the question.
What would complete the picture
Five additions would let a firm evaluate this rather than admire it.
A named customer is the first. Firms ranging from boutique practices to the Am Law 200 describes a category, and one organisation willing to go on the record would establish whether this serves small deals as a convenience or large ones as a necessity.
Time saved is the second and the most important commercially. The claim that review replaces searching is credible and unquantified. Hours per deal, and how that scales with data room size, is the figure a practice group leader needs to justify a paid add-on.
An accuracy benchmark is the third. The release describes the mechanisms that make errors findable and gives no rate. How often the draft misses an item that a careful associate would have caught is the number that determines whether twelve hours is well spent.
Pricing is the fourth, since Marveri sells the tool as a paid add-on with no published price.
And the fifth is the audit trail. What record exists of which lines an attorney reviewed, which flags they overrode, and when. For a document that carries liability, that record is part of the product.

What to watch
Whether Marveri names a firm, because that is what converts a category description into evidence. Whether the twelve-hour run holds as data rooms grow, since a product whose runtime scales badly becomes unusable on the largest deals, which are the ones that pay most. And, whether an accuracy figure appears, because a disclosure drafter that cannot state its miss rate has not yet faced the question that matters. And whether the uncertainty flagging proves calibrated in practice, because a tool that flags too much trains reviewers to ignore the flags and a tool that flags too little removes the benefit entirely.

