How Should M&A Attorneys Scope Flat-Fee Work That Includes AI-Assisted Tasks?

Start with a task inventory, not the label AI-assisted

A flat fee is easier to defend when the work is described as a set of observable tasks. Begin with the matter stage and list the deliverables counsel will provide: for example, reviewing an initial information package, preparing a diligence request list, analyzing specified contract provisions, marking a draft purchase agreement, or coordinating comments through a defined version. Do not make “AI review” a deliverable by itself. The client needs a legal result or work product, while the firm needs a bounded production method.

For each task, record the expected document types, approximate source boundaries, number of planned review cycles, and the decision that ends the task. A task can include an AI-assisted first pass, but the priced unit should include prompt or workflow design, source selection, lawyer review, corrections, and delivery. ABA Formal Opinion 512 identifies competence, confidentiality, supervision, communication, and reasonable fees as obligations that remain relevant when lawyers use generative AI. [2]

Define the source of truth and input assumptions

A scope worksheet should name the records on which each output will be based. For contract analysis, that might be the client-approved data room export, the executed agreement, amendments, schedules, and a list of known side letters. For a draft, it might be a marked term sheet, the latest approved precedent, and written client instructions. Identify who supplies each record, in what format, and by what cutoff. State whether counsel will reconcile duplicates, missing pages, inconsistent versions, or documents received after the cutoff.

The worksheet should also separate extraction from legal judgment. An AI tool may locate clauses or summarize a document, but the lawyer should decide what the clause means for the transaction and whether the source is complete. A useful check is to require every material conclusion to point back to a source document and location. If the source set is incomplete or unverified, the deliverable can be a preliminary issue list rather than a final legal conclusion. That distinction prevents a flat fee from silently covering an unlimited reconstruction exercise.

Assign review and rework responsibility

Name the accountable human reviewer for every AI-assisted task. The reviewer should have authority and skill appropriate to the work, and the scope should say whether review is substantive, citation-level, comparison-level, or editorial. For instance, a first-pass extraction of change-of-control provisions may require checking each flagged agreement against the original, confirming defined terms, and recording false negatives found during sampling. A drafting task may require line-by-line legal review, not a quick read for grammar.

Price the expected review standard and state the rework rule. One included correction cycle is materially different from repeated revisions caused by new instructions or a changed transaction structure. If an output fails the agreed error check, correction of that failure belongs inside the original task. If the client changes the legal position, adds a new document population, or requests a new deliverable, the change process should apply. California’s 2026 guidance, which is jurisdiction-specific, emphasizes that lawyers retain professional judgment and responsibility for AI-assisted outputs and must review, verify, and correct them. [3]

Make confidentiality and permissions a priced gate

Before any client information enters an AI tool, identify the permitted data class, the authorized users, retention settings, access boundaries, and whether the client has imposed a contractual or policy restriction. A public or general-purpose tool may be suitable for a fictional template but not for confidential transaction records. The worksheet should provide a non-AI path when the permission or confidentiality gate is not satisfied, rather than assuming that a tool’s marketing description resolves the issue.

ABA Formal Opinion 512 says the risk assessment is fact dependent and should consider the likelihood of disclosure or unauthorized access, the sensitivity of the information, available safeguards, and the effect of safeguards on representation. [2] California guidance similarly directs lawyers to understand how a product collects, uses, stores, and discloses information, and warns that generalized assurances may not be enough. [3] These are reasons to scope the review of terms, vendor documentation, and client instructions as an actual task. They are not a basis for promising privilege, security, or compliance.

State exclusions and external-cost treatment

A useful flat-fee proposal says what is not included. Possible exclusions include new diligence after the agreed cutoff, tax or accounting analysis, specialist regulatory advice, litigation strategy, negotiations beyond a stated number of rounds, filings, travel, translation, data remediation, and review of materials outside the defined transaction. Use the client’s actual matter to select exclusions. Do not use a generic list to disguise a missing scope decision.

Separate ordinary firm overhead from a matter-specific vendor charge, and explain the treatment in the fee agreement. ABA guidance states that the reasonableness and disclosure analysis still applies when generative AI is used, and California guidance discusses how general office functionality and matter-specific costs may be treated differently under its jurisdictional framework. [2] [3] The proposal should identify whether a cost is included, passed through at actual cost, subject to prior approval, or excluded. Any local rule, client procurement requirement, or engagement term may change the answer.

Use a controlled pilot with stopping conditions

If the firm has not used a workflow on this task type, define a small pilot before committing the entire matter to it. The pilot should use an approved sample, a named reviewer, and a comparison against the source records. Measure the checks that matter to the task, such as missed provisions, unsupported citations, wrong version references, incomplete issue categorization, or disclosure of information outside the permitted boundary. Do not turn the pilot into an unsupported promise of speed, accuracy, or savings.

Set a stop condition before starting. Stop the AI-assisted path if the sample reveals a material missed issue, the tool cannot preserve the required source trace, permissions are unclear, outputs cannot be reviewed to the agreed standard, or the reviewer cannot complete the control within the priced assumptions. Then switch to the defined manual or narrower workflow and notify the client if the change affects scope, timing, or fees. NIST describes its AI Risk Management Framework as voluntary guidance organized around Govern, Map, Measure, and Manage functions. [4] That structure can help organize a pilot, but it is not a legal requirement or a certification.

Write change triggers that match deal reality

Scope changes should be tied to observable events rather than a vague promise to discuss surprises. Triggers may include a document population above the agreed boundary, a new acquisition structure, a material change to the term sheet, a request for additional negotiation rounds, a new jurisdiction, a requirement to analyze a previously excluded issue, a new client data policy, or a tool becoming unavailable or unsuitable. Also state who can authorize a change and whether work pauses until the revised scope is accepted.

The fee notice should explain the consequence of each trigger. That may be a revised flat fee, a separate task, an hourly fallback, a new deadline, or a decision not to perform the work. ABA Model Rule 1.5 says the scope of representation and the basis or rate of the fee and expenses should be communicated, preferably in writing, and that changes to the fee or expenses should also be communicated. [1] The rule is a model rule, so counsel must check the governing jurisdiction and the actual engagement terms rather than treating it as a universal rule.

Turn the worksheet into an engagement conversation

Before sending a proposal, review each task with the client in plain language. Confirm what the client wants counsel to decide, what the client will supply, what the firm will verify, and which decisions remain with the client. Explain AI use when the facts, tool, risk, or client sophistication make communication appropriate, and follow any client instruction restricting AI. Do not imply that the tool is the provider of legal advice or that automation removes the lawyer’s responsibility.

The final worksheet should be short enough to use during intake but detailed enough to preserve the assumptions behind the fee. A practical row contains: task and deliverable; source of truth; input cutoff; tool permission; accountable reviewer; review and rework standard; included cycles; exclusions; vendor-cost treatment; pilot check; stop condition; change trigger; and client approval point. Acquivo is a software platform that delivers the entire sell-side M&A process, valuation, marketing materials, buyer matching, and deal management, on one platform for business owners, M&A attorneys, and boutique investment banks. The platform is user-operated software, so counsel remains responsible for the legal scope, review, permissions, and client communication.

This framework is not individual legal advice. Professional conduct rules and client obligations vary by jurisdiction, matter, tool, and contract. Counsel should obtain the review needed for the applicable jurisdiction before using the worksheet or proposing a fee.