Clause governance: from scattered precedent to governed firm IP

Sixty-nine per cent of legal professionals now use generative AI in their work. Nine per cent work at a firm with a written, enforced AI policy. Those two figures come from the 2026 Legal Industry Report, and the gap between them is where most of the current governance conversation sits. It is a real gap. It is also not a new one. Firms have been drafting from ungoverned content for twenty years — precedent nobody owns, clauses nobody approved, wording nobody can date. What AI has done is make the consequences arrive faster and in more places at once. This piece argues that clause governance is not a response to AI. It is the older, duller problem underneath it, and the firms that fix it will get more out of everything they build on top.

The governance gap is older than the technology

Ask a firm where its standard indemnity wording lives and you will usually get a person’s name rather than a location. That is not a technology failure. It is the natural end state of a profession that produces its best drafting content as a by-product of client work and then leaves it inside the matter it was written for.

Ron Friedmann made the point plainly in his overview of legal knowledge management: “busy lawyers lack the time to convert client-specific documents into a more general precedent”, and the materials that do get saved “have less reuse value than one might expect” because “the context in which they were originally used is very important… rarely do lawyers capture that context”. That was written more than a decade ago and it still describes most firms accurately. The content exists. The context, the ownership and the currency do not travel with it.

So when a firm now asks how it should govern AI-assisted drafting, it is really asking a second-order question. The first-order question is whether the drafting content itself is governed at all.

💡 Insight

It is worth separating two things that often get discussed together: controlling what a tool is allowed to do, and controlling what it is allowed to draw on. Most firms have started on the first. Firms often find the second is where the durable benefit sits, because it improves manual drafting at the same time.

Scattered precedent is a risk problem before it is an efficiency problem

The usual case for fixing precedent is time saved. That case is real but it undersells the position, and it tends not to move partners.

The risk case is sharper. When the same obligation appears in four different forms across a firm’s recent work, the firm has four positions and no view. If one of those forms was drafted under a superseded statutory regime, it is still in circulation, and nothing in the system knows it is wrong. If a clause was negotiated down for a particular client in a particular deal, and it has since been copied into a template, the firm has quietly adopted a concession as its opening position.

None of that shows up in a report. It shows up in a claim, a regulatory review, or a due diligence exercise where a client’s counsel notices that your last three agreements say different things about the same point. The question that follows — who decided this, and when — is the question ungoverned precedent cannot answer.

Precedent is not the same thing as firm IP

The distinction that matters is between content a firm happens to hold and content a firm can stand behind. Three things separate them, and all three are governance rather than technology.

Ownership. A named person — a partner, a PSL, a practice-group lead — is accountable for the wording and has allocated time to maintain it. “The banking team owns it” is not ownership; it is an absence of ownership described in the plural.

Approval state. The firm can say whether a given piece of wording is approved for general use, restricted, under review or withdrawn. Approval only carries information if some content does not have it.

Currency. The firm knows when the wording was last reviewed and against what. Legislation moves, house style moves, and negotiated market positions move. Content without a review date is content of unknown value.

Clause governance is simply the operating discipline that keeps those three true across a body of drafting content as it grows. It is the difference between a precedent bank and firm IP — and it is the reason we describe structured legal precedent management as an asset rather than a filing convention.

What clause governance looks like when it is working

In practice, clause governance shows up as four mechanisms rather than a policy document. Firms tend to underestimate the fourth.

Ownership and approval states, so every clause has an accountable person and a status that controls whether it is available for general use.

Approved variants and fallback positions, so the firm’s opening position, its acceptable middle and its escalation point are all recorded rather than rediscovered under time pressure.

Version history and an audit trail, so the firm can reconstruct which wording was used on a matter and why that was the current position at the time.

A review rhythm, so the library is maintained as a standing process rather than delivered as a project. This is the one that decides whether the other three survive contact with a busy year.

XpressDox Intelligent Clause Management implements these inside Microsoft Word, which is a deliberate choice rather than an incidental one. A governance system that requires lawyers to leave the document they are drafting is competing with copy and paste, and copy and paste is very good at winning.

💡 Insight

Firms often find the approval workflow needs more design than the library itself. Two or three states used consistently will outperform a detailed workflow that nobody completes. It is worth deciding early what a clause without approval actually means in your firm — unavailable, or available with a visible caveat.

Why this now decides what AI is worth to your firm

AI-assisted drafting is only as good as the content it is allowed to reach. A suggestion engine pointed at a firm’s undifferentiated document history will surface the plausible rather than the approved, and it will do so with a confidence the underlying content has not earned. Point the same capability at a governed library and the behaviour changes: every suggestion carries an owner, an approval state and a version.

That is the practical argument for governed AI workflows, and it is why we position AI as a capability within clause governance rather than a replacement for it. The value is not that the retrieval is clever. It is that the content being retrieved is content the firm has already decided it stands behind.

It also explains the figures at the top of this piece more usefully than a policy gap does. Individual adoption is running ahead of firm-level adoption because individuals can adopt a tool in an afternoon and a firm cannot govern its drafting content in an afternoon. The constraint is the content, not the appetite.

Common mistakes firms make when they start governing clauses

These five come up repeatedly, and all of them are avoidable.

1. Starting with a bulk import. Moving every clause in the precedent bank into a new library reproduces the original problem with better search. The first hundred clauses should be chosen, not inherited.
2. Governing the wrong clauses first. Firms tend to start with the clauses that are easiest to collect rather than the ones that carry the most risk. Limitation of liability, indemnities, termination and anything touching a current regulatory change are worth more governance than a definitions section.
3. Treating approval as an administrative step. If everything is approved, approval means nothing. If nothing is, lawyers fall back on the document they used last time and the library becomes a second place to look rather than the first.
4. Building governance away from the drafting environment. A well-structured library in a separate system is still a detour, and detours lose to habit. Adoption is usually decided in the first week.
5. Running it as a project. Projects finish. Legislation, house style and market positions do not, so a library without a review rhythm degrades quietly and the first visible symptom is a lawyer quietly reverting to their own folder.

💡 Insight

It is worth considering how a new clause enters the library, not only how the initial set is built. Firms often find that the submission route — a lawyer proposing wording from live work, an owner reviewing it — is what keeps a library current. Without it, the library dates from the month it was created.

Our view: governance is an accountability question that happens to need software

After fifteen years working on document automation with firms in more than 25 countries, our position is that the technology is rarely the binding constraint. Firms that succeed at clause governance have usually made one decision the unsuccessful ones have avoided: they named people, gave them time, and accepted that a smaller governed library beats a larger ungoverned one.

We would go further. The firms treating this as a knowledge management initiative are underselling it internally. Governed drafting content is the foundation for automated templates, for AI-assisted drafting, and for any credible answer to a client asking how the firm ensures consistency across offices. Run Advanced Document Automation and Intelligent Clause Management together and there is one approved position feeding both automated and manual drafting — which is the point at which the argument stops being about knowledge management and starts being about how the firm operates.

The uncomfortable part is that this is not primarily a purchasing decision. A firm can buy a clause library on Monday and still not have clause governance on Friday. What it can buy is the structure that makes governance possible to sustain — and for most firms, sustaining it is the part that has failed before.

Where to start

A reasonable first test costs nothing. Name the ten clauses that carry the most risk for your firm. Then try to say, for each, who owns the wording and when it was last reviewed. Most firms can complete the first list and stall on the second, and the gap between them is the scope of the work.

That is a smaller starting point than a library project, and it produces something useful whether or not you change any technology. If it would help to talk through where your firm sits — what to govern first, who should own it, and what a realistic review rhythm looks like — our team spends most of its time on these questions. You can book a demo or write to us at support@xpressdox.com.

Frequently asked questions

What is clause governance?

Clause governance is the discipline of keeping a firm’s approved drafting content owned, approved and current. In practice it means every clause has a named owner, an approval state that controls whether it can be used, recorded variants and fallback positions, and a version history that shows what changed and when. It is the difference between storing clauses and being able to stand behind them.

Is clause governance different from knowledge management?

It is a subset of it, and a more operational one. Knowledge management covers everything a firm knows, including who knows it. Clause governance is specifically about the drafting content the firm reuses — the wording that ends up in client documents — and about the ownership and approval that make that wording safe to reuse.

Do we need a knowledge management department to do this?

No, though you do need named ownership. Smaller firms typically appoint one or two experienced fee earners as clause owners for their main practice areas and set a review rhythm rather than creating a function. What does not work is leaving ownership at team level, because a team cannot be accountable for a review date.

Where should a firm start if the precedent bank is already large?

Start with risk rather than volume. Identify the clauses where an out-of-date or inconsistent position would matter most — typically liability, indemnity, termination and anything affected by a recent regulatory change — and govern those first. A bulk import of everything tends to recreate the original problem in a new system.

How does clause governance affect AI-assisted drafting?

It determines what the AI can draw on. An assistant pointed at an ungoverned document history will surface wording that looks plausible but carries no approval; pointed at a governed library, every suggestion arrives with an owner, an approval state and a version behind it. Governance is what makes AI-assisted drafting predictable rather than merely fast.