Most mid-size law firms are not differentiated by hours worked. They are differentiated by which hours.
Some of those hours create direct value for the client: the negotiation strategy on a high-stakes deal, the judgment call on whether a clause is worth fighting, the partner-level read of where a transaction is actually heading. Other hours are mechanical. Reading every line of a 40-page master services agreement to make sure none of the boilerplate has drifted from market. Summarizing the deal terms for a co-counsel. Parsing a redline from the other side to figure out what actually changed. Those hours pay the bills, but they are not the thing the client is paying for.
AI applied to the early end of a deal cycle does not change the lawyering. It changes the ratio of mechanical hours to judgment hours.
What contract review looks like at a mid-size firm
At a firm of 50 to 200 attorneys, transactional work comes in waves. A typical deal might have a corporate partner, two senior associates, two or three junior associates, and a paralegal or two coordinating the document flow. Once an agreement comes in from the other side, or once the firm's own first draft goes out, the first wave of review happens at the associate level. They read every clause, flag deviations from the firm's templates, summarize material changes, and surface anything they think the partner will want to look at.
This first pass takes time. A 60-page acquisition agreement might run an associate eight to twelve hours, sometimes more if the deal is unusual. By the time the partner sees it, the associate has built a summary memo, marked up the document, and identified the three or four issues that actually need partner judgment.
The next wave is the same logic at the next level. The partner spends real time on the issues the associate flagged, asks for follow-up research on the unclear ones, and starts thinking about how to position those issues with the client and with the other side. The strategic part of the deal happens here.
The cycle is not broken. It is front-loaded. The first wave is structural work that requires legal training but does not require the senior judgment that the client is actually paying for.
Figure 1 · Where associate hours go on a 60-page transaction
Illustrative proportions, not measured outcomes from a specific firm.
Where AI fits
A 2024 report from the Thomson Reuters Institute surveyed legal professionals and found that 73% expected AI to have a high or transformational impact on legal work within five years, with contract review and document analysis topping the list of specific applications (Thomson Reuters Institute, Future of Professionals Report, 2024). Stanford's CodeX center has tracked how large language models perform on legal reasoning tasks since 2022; their work has shown both the capability and the failure modes in detail (Stanford CodeX, ongoing research).
For a mid-size firm, AI applied to contract review compresses the front of the cycle. The first-pass structural read, the comparison against firm templates, the flagging of deviations, and the summary memo are all generated by an AI system before the associate sits down with the document. The associate does not start with a blank document. They start with a draft analysis already in hand.
What gets surfaced by that first pass tends to fall into a few categories. There are deviations from the firm's standard language, which the system identifies by comparing the document to a library of past redlines and approved templates. There are unusual provisions: language that does not appear in any of the firm's templates and may indicate the other side is testing something new. There are inconsistencies within the document itself, definitions used in section 8 that contradict the same term in section 12. And there are cross-references and dependencies, where a change to one clause cascades into others that have not been updated to match.
These are all things the firm catches today. They are just being caught earlier in the cycle.
Where the firm's expertise stays central
It would be a misreading to think the AI replaces the associate or the partner. The shape of the work changes, not the existence of it.
For associates, the first-pass hours do not disappear. They shift. Instead of reading the document to identify what is flag-worthy, the associate reads the AI's flags to assess which ones actually matter. That is a more senior task, and one that builds judgment faster. The associate is, in a sense, getting reps on the part of the work that compounds.
For partners, the practical effect is timing. The AI-prepared brief gets to the partner faster, which means partner judgment enters the deal earlier. The deal gets more strategic attention per dollar of fees, or the firm handles more deals per quarter with the same partners. Most firms split the difference.
For paralegals, the role shifts toward quality control on the AI's outputs and ensuring the firm's voice and style hold across documents. Paralegals who already had deep institutional knowledge of what the firm's templates say become even more central, because they are now the ones validating that the AI is referencing them correctly.
Figure 2 · What each layer of the review catches
| In the document | AI first pass | Associate | Partner |
|---|---|---|---|
| Deviations from firm template language | Surfaces reliably | Confirms severity | Positions with client |
| Internal inconsistencies (term defined two ways) | Surfaces reliably | Decides which to keep | — |
| Cross-reference dependencies between clauses | Surfaces reliably | Validates the chain holds | — |
| Unusual provisions not in any firm template | Flags as novel | Researches precedent | Strategic judgment |
| Implied negotiation tactics in a redline | — | Pattern-matches past deals | Reads the room |
| Material commercial trade-offs | — | Summarizes options | Decides |
Illustrative breakdown of where final judgment sits at each stage of review.
Where the boundaries matter
Two boundaries are non-negotiable.
Confidentiality is the first. Most mid-size firms cannot run client documents through a consumer-grade AI service. The document handling has to be on-premises or through an enterprise tier that contractually guarantees the firm's data is not used for training and is not accessible to anyone outside the firm. This is increasingly available; the largest providers have built enterprise legal offerings specifically because firms need it. A McKinsey review of AI adoption in legal services found this concern was the single biggest gating factor for mid-size firm adoption, ahead of cost or capability (McKinsey & Company, The state of AI in legal services, 2024).
The second is hallucination risk. Large language models can and do fabricate clause references, miscite cases, and invent precedent that sounds plausible but does not exist. The 2023 Mata v. Avianca case, where attorneys submitted a brief with AI-generated case citations that turned out to be fictional, became the canonical example (Mata v. Avianca, Inc., 22-cv-1461, S.D.N.Y., 2023). For contract review specifically, the failure mode is more subtle. The AI might correctly identify that a clause deviates from standard, but suggest a remedy that does not quite match how the firm has handled similar deviations historically. Every output that leaves the firm has to be reviewed by the human who would have done the work anyway.
The AI is an associate-level assistant. It is not the partner. It never makes final calls, never sends documents to the client without human review, never signs off on anything that lives outside the firm.
What changes at the firm level
The change at the firm level is mostly about capacity and positioning.
A firm that integrates AI into the early end of its review cycle handles more transactional matters per quarter with the same headcount, or handles the same volume with significantly more partner attention per deal. Either is a competitive advantage in the mid-size segment, where firms compete with both larger firms (who have more associates) and smaller firms (who can move faster on smaller deals).
The firm's brand also shifts in a quieter way. Clients who experience a faster turnaround on first drafts and tighter partner attention on strategic issues do not usually attribute the change to AI. They attribute it to the firm being responsive and well-staffed. The internal mechanism is invisible to them, which is the right outcome.
The firm becomes, in effect, leaner at the front end of every deal and more concentrated at the back end. Less time spent reading the boilerplate, more time spent on the things that actually compound for the client and for the firm's reputation.
Where this is heading
The trajectory across the industry is clear, even if the rate of adoption varies by firm. The 2024 American Bar Association Legal Technology Survey Report found that 23% of law firms surveyed were already using some form of AI in contract review or document analysis, up from 12% a year earlier (American Bar Association, 2024 Legal Technology Survey Report). Among mid-size firms specifically, the curve is steeper because the operational pressure is higher. Mid-size firms compete on responsiveness, and AI on the early end of the cycle is one of the highest-leverage operational changes available right now.
Figure 3 · AI in contract review, % of firms using
Anchored to ABA 2024 Legal Technology Survey Report; per-segment splits illustrative.
The second-order effect, which is less discussed but probably more important, is what it does to the firm's expertise mix. As the mechanical work compresses, the firms that succeed will be the ones whose senior judgment scales with the new throughput. The partners and senior associates whose judgment is hard to replicate become more central to the firm's value, not less. The firms that try to use AI to thin out their senior staff will discover quickly that they have cut the thing the client was actually paying for.
Figure 4 · AI in contract review, 2020–2027 trajectory
2020–2024 anchored to ABA Legal Technology Survey series; 2025–2027 projected at observed growth rates.
The firms that get this right are operating on what is possible today. Most firms are not there yet. That gap, between what the technology supports and what most firms are running, is the operational space where the next round of competitive sorting is going to happen.
Sources
- Thomson Reuters Institute, Future of Professionals Report, 2024.
- Stanford CodeX Center for Legal Informatics, ongoing research on LLM legal reasoning.
- McKinsey & Company, The state of AI in legal services, 2024.
- Mata v. Avianca, Inc., 22-cv-1461, S.D.N.Y., 2023.
- American Bar Association, 2024 Legal Technology Survey Report.