Why you should not have to choose between document automation and AI

Second in a series of three on AI, document automation, and the people who make both work.

Legal teams are being asked by their boards, their partners and their leadership to have an AI strategy. That's fair enough, as AI has arguably spread through firms faster than any legal technology before it. At the same time, there's a feeling that AI is somehow taking over from document automation, but that tends to be a mistake.

Even so, many legal operations teams are now questioning whether to move budget away from document automation and towards AI, but it's not a case of one or the other. The two tools complement each other, and the teams that understand this get significantly more value from both.

Why it's not either/or

Document automation and AI do different jobs, and each is good at different things. Each is also, arguably, not very good at some of the things the other does well. Neither covers the whole drafting process on its own, from creating a document through to negotiating changes and getting it signed.

The case for document automation. Document automation is at its best when a document has to follow an agreed form, whether that’s a standard NDA or a complex SPA. Simpler agreements can be produced quickly, with the firm’s agreed wording, its branding and layout, and numbering and cross-references that hold. For more complex documents such as SPAs or loan agreements, the questionnaire behind the template makes sure critical clauses aren’t left out of the first draft, which leaves much less room for technical error than a draft produced from a prompt alone. It also lets non-lawyers produce documents without drifting from the standard position.

Where it struggles is anything that hasn't been built yet. An unusual joint venture agreement has no template to start from, and building one for a deal you might only do once can take longer than drafting it. The same goes for matters that change shape as they go, and for working through the other side's mark-up.

AI is better suited to most of those. It copes with one-off and non-standard documents, lower-risk amendments and reviewing what comes back in negotiation, and it can help you work out what your standards and playbooks should say. It is less good at keeping one clause worded the same way across hundreds of agreements, or keeping Word styling and numbering intact. Moving documents that already work as templates over to AI is also more work than it looks. Each one has to be rebuilt as prompts and playbooks, tested until the output is consistent, and then tested again whenever the model changes.

So going all in on either one means missing out on something. Most teams have both kinds of work, often in the same week. Put everything into templates and the one-off work and negotiation stay manual. Put everything into AI and you spend time rebuilding the consistency your templates already gave you. Often both apply within a single agreement, with an agreed core that shouldn't move and matter-specific detail that always does.

For teams without document automation

Start with AI. If your organisation does not currently have doc auto, AI is a reasonable place to start. It's accessible, and it produces useful output quickly. There is no build phase and no template to design first, so a team can have something in front of its lawyers within hours.

Understand the risks. But teams using AI for drafting without document automation need to be clear-eyed about the risks. Without a template behind it, each draft is only as consistent as the prompt that produced it. That leaves whoever is checking with nothing fixed to compare against, so every draft is checked from scratch. In our experience, they are looking for wording that varies between users, Word styling that comes apart on paste, legal numbering and cross-references that break once a document is amended, and hallucinations.

Upfront investment pays dividends. If this is where you’re starting, our advice is to spend a little time on some standard language and document types first. A playbook to check drafts against is a good start. Better still is a playbook and a template the AI can start from, so its first draft begins from wording you’ve already agreed.

Know your users. In-house teams need to be particularly careful. If business users are drafting with AI and there's no standardisation behind it, they can create serious problems that legal doesn't know about (yet). Setting that standardisation up safely also takes a level of AI expertise that many legal teams don't have yet.

For teams with document automation already

If your organisation already has good, up-to-date doc auto, you're in a good position. You have all the safeguards that come with standardised templates, and you can now layer AI in where it works best.

Layering AI in usually starts with the parts of the process that document automation isn't designed for: negotiation, review and non-standard drafting. The teams getting the most value are doing that while also feeding data into their templates more effectively and keeping them current. Once AI is working well on clause language in negotiations, the natural next step is a playbook built from those templates. AI can use it to check the other side’s comments, and even show how far the final draft has moved from the original.

The template audit. If your document automation isn't in that state, it's a good idea to find out before AI gets layered on top. Many of the platforms can show which features are being used. So if entire sections are hardly touched, or lawyers have drifted back to their own precedents, any AI built on top of those templates will carry the same problems. If the templates are out of date, inconsistent or no longer trusted, whatever AI produces from them will be too.

The real opportunity: using both together

Used together, you get the consistency of standardised templates and the flexibility to handle one-off work efficiently, without trading one for the other. The templates keep AI's output within positions the firm has already agreed, and AI covers the work no template was built for. They also make each other better over time. AI makes standards quicker to build and keep current, and the better those standards are, the more reliable AI's output becomes.

What that looks like depends on where you're starting from. Teams without document automation can use AI to build the standards they're missing, and teams with it can build on what's already there: