The Judgment Contract Redlining Still Requires

Contract lifecycle platforms have spent 15 years trying to move redlining inside their walls. Every demo shows the same thing, which is a negotiation happening in a browser, both parties commenting in a shared workspace, the version history tidy and centralized. Then the deal starts, and somebody emails a Word document.

Redlining is the one contract workflow that has refused to move. Not because lawyers are stubborn, and not because the software is bad. It refuses because a redline is the only contract activity requiring the cooperation of someone who does not work for you. You can standardize your intake, your approvals, and your storage. You cannot make the counterparty adopt your platform.

That constraint explains more about contract redlining than any feature list, and it shows up plainly in what happens after a legal team adopts AI. Time spent reviewing contracts falls fast and keeps falling. Time spent negotiating them holds roughly where it was. Review is something a team does alone. Negotiation moves only when the other side moves too. This article explores where the automation actually landed, what a returned markup hides, whose standard the first pass gets measured against, and the one condition under which the exchange finally changes shape.

Contract Redlining is a Two-Party Problem

Contract redlining is the process of marking proposed changes to a contract so both parties can see exactly what was added, removed, or reworded. Each side edits with tracked changes and explains material revisions in comments, exchanging versions until the language works for everyone. The original wording stays visible alongside every proposal.

What separates a markup from an ordinary edit is that nothing disappears. An edit replaces language. A markup shows the original and the proposal together, so the other party can see the delta and weigh it.

Redlining sits inside a larger sequence. Contract review identifies the issues in a draft, redlining resolves them on the page, and negotiation settles whatever the markup cannot. A lawyer moves between the three without announcing the transition, which is why the distinction gets ignored. Hold onto it, because everything explored below depends on it.

The Redlining Workflow That Stayed in Word

Consider what a redline actually is. Two organizations, each with its own document management setup, its own approval chain, and its own counsel, converge on a single file and take turns changing it. That file is the only shared infrastructure either side can rely on.

This is why tracked changes have outlasted every attempt to replace them. A .docx opens everywhere. It carries its own edit history. It needs no login, no seat license, and no procurement conversation with the other side's IT department. Any tool requiring the counterparty to create an account has already lost the negotiation it was designed to host.

The consequence for buyers deserves stating plainly. Half the workflow belongs to somebody else, so contract redlining software promising to own your redlining is promising something it cannot deliver unilaterally. Your own half is where its influence actually reaches. That means how fast your side produces a markup, how well it reflects your positions, and how quickly a lawyer can verify it before it goes back out.

Harvey was built to accept that constraint. It works natively inside Microsoft Word and Outlook, and connects to iManage, NetDocuments, SharePoint, and Google Drive, which means the intelligence arrives where the redline already lives. Nobody has to be told to work somewhere new, least of all the party across the table. A lawyer marking up a supplier agreement stays in the document they were always going to mark up.

Buyers can get this wrong in a predictable direction. A repository gets evaluated on where contracts sit. Legal AI gets evaluated on what happens to the language inside them. Confusing the two produces a purchase that organizes agreements beautifully while leaving lawyers marking up drafts exactly as they did before. Both categories are worth owning, and teams frequently run both, but only one of them changes the redline itself.

There is a second-order effect worth naming. Because you cannot standardize the counterparty, every improvement has to come from your own side of the exchange. That makes the quality of your first pass and the speed of your verification the entire competitive surface. It is a narrower place to compete than the category usually advertises, and a more tractable one.

Review Compressed and Negotiation Did Not

Reviewing a draft and negotiating it feel like one continuous task from inside the work. Put the right tools behind both and the two separate, with review time compressing by multiples and negotiation time moving by degrees. Start with the review half, where the change is easiest to measure.

The European insurance group Talanx cut review time on certain contract tasks from two hours to 15 minutes, which added up to more than 400 external consultant hours saved in a year. The same pattern shows up at Bridgewater, where a supplier agreement that had taken two days to review came back in two hours. Those are eight-fold and twenty-fold compressions on a single task.

What links them is the kind of work being compressed. Reading a draft, checking it against a standard, and flagging what departs from it is the half of contract work a team performs alone. Neither organization needed the counterparty to agree to anything for the gain to arrive.

Now look for equivalent numbers on the negotiation half. According to The Accelerating Impact of Legal AI, an independent RSGI study commissioned by Harvey, 91% of in-house teams using Harvey reported spending less time reviewing contracts. Only 44% reported spending less time negotiating them. No customer figure anywhere shows negotiation compressing the way review did.

Redlining straddles the two, which makes it the most instructive workflow in contract practice. Producing a first-pass markup is largely a review task and compresses accordingly. Deciding which of those proposed changes to send, which to soften, and which to trade away is entirely a negotiation task and does not compress at all.

The economics follow the same split, and they show up first in external spend. The Adecco Group supports more than 300,000 contracts across 45 countries. Its legal and compliance team reports up to eight hours saved a week per lawyer on routine work, alongside a 5 to 10% reduction in outside counsel spend. The saving comes from keeping work in-house that used to go out, which is a review decision before it is a budget decision. Capacity freed on the reading side gets spent on the deciding side, where the commercial value sits.

Teams expecting negotiation cycles to compress by the same proportion are disappointed, and the disappointment is predictable. Rounds are governed by how far apart the parties are commercially, and no tool closes that distance. The realistic promise is that your rounds get spent on genuine disagreements, with nothing lost to something somebody missed. That is a smaller claim than the market usually makes and a more reliable one.

The Four Ways a Redline Slips Past Review

Producing a markup and reading one are different skills, and the risk lives in the second. The changes that matter most often arrive in the quietest form on the page. Volume is a poor guide to danger, since the heaviest markup is usually the most harmless. Four failure modes account for most of what gets missed, and they share a root cause worth naming up front.

Changes that arrive untracked

Edits made with tracking turned off leave no visible mark. Sometimes this happens by accident, when someone works in a copy with tracking disabled. Sometimes the omission is deliberate. Either way, the only reliable check is a document comparison between what came back and what you last sent.

Definition drift

A defined term amended once in the definitions section changes the meaning of every clause that uses it. The markup shows a single edit near the front of the document. The effect runs through 40 pages. Narrowing "Confidential Information" by six words can strip protection from a whole category of material without anyone touching the confidentiality clause.

Broken cross-references and carve-outs

A deleted subsection leaves cross-references pointing at the wrong place, so a clause that once referred to Section 8.2 now points at whatever moved into that slot. Carve-outs behave the same way. Removing confidentiality from the exceptions to a liability cap changes your exposure without editing the cap, and the markup shows only a short deletion inside a list.

Deletions that read as agreement

A clause amended shows heavy markup. A clause deleted shows a single strikethrough. The highest-impact change on the page can be the smallest, which is how an entire indemnity or audit right disappears without drawing attention. Work the deletions first and the amendments second.

Notice what these four have in common. Every one is a comparison failure wearing the costume of a reading failure, which is why they survive a careful lawyer and why software catches them reliably. This is the kind of task AI agents for legal work handle reliably. Harvey compares a returned draft against the version you sent and surfaces what moved, including edits tracked changes never displayed.

This is the least glamorous capability in legal tech and quite possibly the highest value one. A missed liability carve-out does not announce itself for three years. By the time it surfaces, the cost of the omission is being litigated, and nobody is pricing it any more. Comparison is also the capability least dependent on model sophistication, which means it is available to any team willing to build the habit of running it.

The Standard Behind the Markup Decides Everything

Ask a general-purpose AI to redline a services agreement and it will produce a competent markup against the market standard. That is the problem, and it is what separates a general model from legal drafting AI measured against your own positions. Market standard is an average, and no organization negotiates at the average. The gap between the two is where the review work returns.

Your organization negotiates against its own approved positions, which reflect its insurance program, its regulatory exposure, its risk appetite, and 10 years of deals it wishes it had papered differently. A markup measured against generic best practice creates work, because a lawyer then has to reconcile every proposed change against what the organization would actually accept. A markup measured against your own approved fallback positions can be checked and sent.

Harvey Playbooks encode those positions so a first pass reflects your standards, and Harvey's Contract Intelligence applies them across the agreements a team handles at volume. The platform stays the same across every customer. What changes is the precedent, the standards, and the codified expertise each team brings to it.

RSGI's Emerging Trends for the Evolving Business of Law makes the strategic version of this point. Competitive advantage depends less on hours an organization can deploy and more on how well it captures what it already knows. Applied to contracts, your accumulated positions are the asset. The software is identical for everyone who buys it, and what differs is what each organization brings to it.

That reframes configuration as something more than setup. Encoding standard positions means documenting a process well enough to repeat it, which Harvey's governance research names as the precondition for reliable agent output. An organization that has never written its positions down cannot encode them, so the exercise forces an articulation that was always tacit. That articulation holds its value whichever tool prompted it.

This is also why contract review software selection turns on configurability. The question worth asking a provider is whose standard the draft gets measured against, and how much work it takes to make that standard yours. Legal drafting capability is table stakes across the whole category.

Where AI Can Help With Contract Redlining

Once the first pass is automated, the constraint moves to checking it. A markup a lawyer cannot verify quickly is worse than no markup, because it carries the authority of having been produced by something and the risk of never having been read closely. Two tools can generate identical markup and produce opposite outcomes, depending entirely on how fast a reviewer satisfies themselves that each flag is right.

Grounding matters more than fluency here. Harvey ties every flagged change to the source language it came from, so a reviewer opens the clause, sees the original text beside the proposal, and confirms or rejects it in seconds. Across 47 tracked changes, the gap between traceable output and merely plausible output is the gap between a tool that saves a morning and one that costs an afternoon. Harvey's own evaluation criteria put accuracy in these terms, as output grounded in verifiable sources and built to be reviewed. Additionally, a qualified lawyer should review any AI-generated redline or analysis before your organization relies on it.

The verification burden explains an uneven pattern in the adoption data. According to Defining the Impact of Legal AI, RSGI's November 2025 study of Harvey customers, a typical Harvey user at a law firm saves 15.7 hours a month while a power user saves 36.9. The in-house split runs 11.8 against 28.3. Power users run the same software as everyone else. What separates them is habit, including configured positions, saved workflows, and a settled routine for what gets verified and by whom.

CMS treated that as the actual project. The firm ran a 12-month pilot with 300 lawyers before scaling, approached the rollout as a transformation program, and reached more than 3,000 lawyers with over 95% of them active. Head of Innovation, Knowledge and Technology Bert Vries has been explicit that the sensitivity of client data made security and governance a selection criterion in its own right.

Auditability belongs in the same conversation. Extending Governance to Legal AI Agents describes AI governance in legal practice as resting on two layers, written policy that sets expectations and the programmatic controls that turn those expectations into practice. The second layer covers layered permissions, access boundaries, audit trails, and adoption visibility. For redlining specifically, where a markup carries the commercial position of both parties, being able to show what the tool did and who approved it is part of whether it can be used at all. Deutsche Telekom ran a full privacy and information security assessment before adopting Harvey in early 2024, and General Counsel Dr. Claudia Junker has been direct that encrypted data held in Germany was a condition of proceeding.

The Redline as the Record of What Was Agreed

Everything above accepts the constraint that you cannot control the other side's tooling. There is exactly one condition under which that stops being true, and it is not a purchase. It requires the counterparty to want the same thing you want, which is rarer than any procurement process can manufacture.

Deutsche Telekom and Gleiss Lutz have worked together for years, drafting side by side on complex matters. Both organizations identified the same friction, which came from collaboration running through email, tracked changes, and version control at precisely the moments that mattered most. Their account of that work describes moving toward shared context inside Harvey, with Junker describing the goal as working in the same space, on the same material. Gleiss Lutz’s Co-Managing Partner, Dr. Ralf Morshäuser, names the enabling condition, which is that collaboration works when both sides operate at the same level.

Think of it as two tiers. With trusted counterparties on repeat work, the exchange can genuinely collapse into shared context and the version-passing goes away. With everyone else, the markup exchange remains, and your advantage comes entirely from your own half of it. Most contracts functions will live in the second tier for years. That is a reason to build the half you control, and a poor reason to wait.

Handle the exchange as a negotiation with a shape, read what comes back against what you sent, and check the execution version before anyone signs. Harvey marks a draft against your organization's standard positions, surfaces what a returned redline moved, and grounds every flag in the language behind it. It does this inside Word, where the redline already is, which is part of why Harvey holds a 92% monthly adoption rate across its customer base. To see it run on the kind of markup your team handles every week, request a demo.

Frequently Asked Questions About Contract Redlining

What is contract redlining?

Contract redlining is how two parties negotiate an agreement in writing. One side marks proposed edits as tracked changes, explains the material ones in comments, and returns the draft. The other side accepts, rejects, or counters, and the versions pass back and forth until the language works for both. Nothing gets overwritten, so the record of what each party asked for survives.

When should you redline a contract?

Redline once you know which terms fall outside what your organization will accept, which puts the markup after review and not during it. Low-risk standard forms may need no markup at all. Marking everything on a routine agreement spends credibility you need for the clauses that carry real exposure.

How do you redline a contract?

Turn on tracked changes, then mark it against your organization's approved positions. Contract redlining best practices cover the mechanics in full, though the yardstick matters more than the technique, since no organization negotiates at the market average.

What is the difference between contract review and contract redlining?

Contract review is the assessment of a draft against your organization's requirements and risk tolerances. Contract redlining is the markup that follows, where a reviewer proposes edits as tracked changes and explains them in comments. Review finds the issues. Redlining resolves them on the page across successive versions.

Can AI redline a contract on its own?

Contract redlining AI can produce a first-pass markup against a defined set of positions, though automating contract redlining depends more on documenting those positions than on the tool. Harvey measures that pass against your organization's own approved standards, which is what makes the output checkable. A qualified lawyer still reviews it before it reaches the other side, and the research consistently shows review time falling faster than negotiation time.

Does contract redlining have to happen in Word?

In practice, almost always. Tracked changes travel in a file both parties can open without adopting each other's software, which is why the format has outlasted every attempt to replace it. Harvey works natively inside Word and Outlook for that reason.

Who should redline a contract in your organization?

Ownership should follow the risk in the agreement. Business owners can clear low-risk standard forms against a short list of deal-breakers, while qualified legal reviewers handle negotiated agreements and senior counsel owns high-value deals. Whoever holds the pen should also run the version comparison, a step Harvey automates so nothing moves unseen.