The more things change: what 25 years of legal drafting technology taught us about adopting AI
First in a series of three on AI, document automation, and the people who make both work.
Someone in your firm has said that AI is going to change how you draft. Someone else has said that checking the output takes as long as writing the thing yourself. We heard both of those arguments in the early 2000s, about document automation.
Between us we have worked on every generation of drafting technology since HotDocs, as vendors, as implementers and as consultants. The objections have held steady.
Most firms and in-house teams we speak to are approaching AI as a fresh problem, with a new team and a blank page. Many of them have done this twice already.
The technology is different this time, but everything that decides whether people use it is familiar.
We remember the early market
The early market was full of ambitious software. The products that gained early traction had something more modest going for them. They sat where lawyers already worked.
HotDocs launched in 1993 and was the first automation tool most of our team encountered. Arguably, the reason it took off above other software was simple: the legal language lived in Word, and that mattered more than anything else on the feature list. Change is easiest to implement when it asks the least of people. By slotting into the way lawyers already worked, this minimised friction and made adoption easier.
By 2000 Julie Saliba, now our Chief Innovation Officer, was inside a large international law firm evaluating platforms, most looking to compete with HotDocs . Documentum’s little known automated content needed developer-grade coding to produce anything usable, which kept it a technology rather than a product. SpeedLegal, later Exari, was XML-based and lived outside Word, which added barriers to adoption. Both alluded to the art of the possible but both asked lawyers to change where they worked, which decided how far they got inside law firms.
Julie later joined Business Integrity, the company behind DealBuilder, working alongside Will Sumners, now our Chief Operating Officer, and Anna Turner, our Director of Delivery. DealBuilder automation worked from inside Word, where the drafting was happening anyway. Linklaters became its first customer and Clifford Chance followed soon after.
DealBuilder became Contract Express, which grew into the leading automation platform of its generation, alongside HotDocs and Exari. Thomson Reuters acquired Business Integrity in 2015.
Around the same time, a new group arrived with three answers to one question. Clarilis took the build work off firms entirely. Avvoka lowered the coding barrier with a cloud platform. Document Drafter competed head on inside Word. Three different bets, and all three found firms they suited. Ownership decided more than architecture did.
The firms that ended up with shelfware had a platform and a single project plan.
The firms that got real value:
- named someone to own the work,
- found champions who carried it into the practice groups,
- kept their templates current, and
- held a clear view of what to automate and when.
AI made starting easy and moved the work later
The chat interface and natural language prompting of AI platforms mean you can have something useful in minutes. That is a real advance, and it explains why AI has spread through firms faster than any legal technology tool before it.
It has also persuaded a lot of organisations that AI asks for less sustained investment than document automation did. But while the upfront investment is lower, the ongoing investment of reviewing each transaction may exceed that in fairly short order. Document automation is paid for upfront, in a business case and then months of template building. AI is paid for afterwards, in building a prompt library people trust, in handling hallucinations and the checking they create, and in grounding output in the firm’s own knowledge.
Template building needs a budget and a signature, so it gets scheduled. Prompt libraries and grounding material need somebody’s attention, which is harder to put in a plan and easier to push to next quarter.
Firms pay for it either way, and the second way arrives as unbilled checking time, and as AI billing models change, ongoing token costs.
Similar problems, very different outcomes
Whatever a firm is rolling out, the agenda reads the same:
- get people to adopt it,
- keep them engaged,
- scale it past the first team.
What moves is where each one gets difficult.
With document automation, the hard part is getting people to start. With AI they likely started months ago, possibly on personal licences. Finding out who is using what takes more digging than writing a launch plan. Those licences keep the usage invisible, and your confidentiality position now rests on tools nobody has reviewed.
Engagement fails in both cases, just in opposite directions. A stale template produces a wrong output, the lawyer notices, and they go back to their own precedents for good. AI produces a wrong answer that reads like a right one, and everybody carries on regardless – until its found by a client or court imposing penalties.
Scaling document automation is a production problem. More templates, more practice groups, more build time, all of it visible and plannable.
Scaling AI is a consistency problem. Two hundred people using it means two hundred ways of using it.
All three problems have the same answer they always had. Somebody owns the standard, keeps it current, finds the people who will carry it, and decides what belongs in the tool. Document automation came with a build plan that made the work visible. AI arrives without one, so somebody has to find the work and schedule it.
Firms that do that work end up with technology people use. Firms that treat the easy start as permission to skip it get away with it for a while, and pay more to fix it later.
What ownership looks like this time
Ownership starts with the standards and the prompt library, and somebody needs time in their week for both. The software already has an owner, usually in IT or procurement. The output needs one too.
Champions do the job they did with templates, on different material. They used to tell you which precedents were worth automating. Now they surface the prompts worth keeping and flag the tasks where checking AI costs more than drafting from scratch.
Maintenance is the part firms underestimate. Document automation needs somebody watching for precedent and legal changes and working them through the templates on a cycle. AI needs that same watch, plus a second one, because the models change underneath the prompts. A prompt that behaved reliably in January can behave differently in June after an update, and the person relying on it tends to discover that during a client deliverable. Either way, somebody has to own the review and be measured on it.
Strategy is the same decision it always was, with a third option added. Work out what to automate, what to prompt, and what to leave to a lawyer with a pen. Most firms need all three, and the boundaries move every year.
This is what we do
Over three decades we have watched firms buy the identical platform and end up in completely different places.
The firms that treated implementation as a starting point are still getting value from decisions they made fifteen years ago, and they will get value from AI for the same reasons.
Most firms we speak to already know roughly what needs doing. What they need is somebody whose job it is to do it. We advise on who should own it, where AI sits alongside the automation you already have, and what to automate, what to prompt and what to leave alone. We build and maintain templates, set the standards AI works from, and train the people using both.
The question worth asking about any of this is:
who owns it the day after you buy it.
If you would like to talk that through for your firm, contact Julie or Will.
FAQs
What is different about rolling out AI for drafting compared with document automation?
Scope, mainly. Document automation is bounded. You choose which documents to automate, build them, and the tool does that job. AI will attempt anything you point it at, so far more time goes into deciding what to use it for and what to leave alone. That work has no real equivalent in a doc auto rollout.
Where the two overlap is the preparation. Automating a precedent set meant getting the precedents in order first. Grounding AI means getting your knowledge and data in order first. The same three problems of adoption, engagement and scale still arise, just in different places. Both take longer than anyone expects, and both decide whether people end up trusting the output.
Does AI need as much investment as document automation?
About the same amount, arriving at a different point. Document automation is paid for upfront, in a business case and then weeks (or months for bigger projects) of template building. AI is paid for afterwards, in building a prompt library people trust, handling hallucinations and the checking they create, and grounding output in the firm’s own knowledge. Template work gets scheduled because it needs a budget and a signature. The later work needs somebody to schedule it deliberately.
How should a law firm approach AI adoption?
As a change management job rather than just a software one. Name somebody to own the standards and the prompt library, find champions in the practice groups who will carry it, review prompts and grounding material on a schedule, and decide which tasks belong in the tool. That list is what separated the firms that got value from document automation from the firms left with shelfware. With AI the same work lands after the rollout rather than before it.