340 firms, 28 countries: what changed in a year, and what didn’t
A year ago we had just over a hundred accounts. Now there are 340 in 28 countries. Some of what we expected to happen did. Much of it didn't, and that is the more useful list.
Twelve months ago, there were 112 firms and legal teams on ZAAN, in nine countries. This week, there are 340 across 28. The growth is not the interesting part. What the growth revealed is.
We keep a running list, internally, of the assumptions we held a year ago and whether they survived contact with three times as many accounts. This note is that list, lightly tidied.
What changed
Playbooks became the product. A year ago, firms came to us for Review and treated the playbook as setup. Now the playbook is the thing firms talk about. The practice heads who onboarded in the last two quarters spent more of their onboarding time on positions and fallbacks than on the Word pane. The playbook made explicit what the practice had been doing by instinct and inconsistency.
This changed how we build. Review 3.6 moved severity into the playbook. The rule editor has had more engineering attention in the last six months than the review pane itself.
The spread of practice areas widened. In early 2024, accounts were overwhelmingly M&A and commercial. Employment, real estate and construction teams now make up around a third of active playbooks. Each brought clause types we had not prioritised. A construction practice cares about liquidated damages schedules and time-bar clauses in a way an M&A team does not. The must-survive list in a real-estate playbook looks nothing like the one in an MSA playbook. We have added clause-type coverage accordingly, and the gaps that remain are mostly in the areas we have most recently entered.
Smaller firms arrived. The median account a year ago was a firm of around 150 lawyers. The median now is closer to 60, and we have a growing number of in-house legal teams of fewer than ten. The smaller the team, the more they lean on Draft rather than Review: a two-person legal function does not have a stack of counterparty drafts to review, it has a stack of agreements it needs to produce.
Recall got used for things we did not design it for. We built Recall to find precedents. Across our active accounts, about a quarter of queries are now something else: who in the firm has acted on a matter of this type, what did we charge on comparable work, when did we last see this counterparty. The search works for these questions because the matter history contains the answers, but we had not thought of them as the use case. We are now.
What did not change
Reviewers do not trust a flag without a reason. This was true at 112 firms and is true at 340. Every flag in Review carries the playbook rule that triggered it and the clause text it matched. We tested, briefly, a condensed view that hid the reason behind a click, on the theory that experienced users would not need it. Acceptance of flags fell. The reason is not decoration; it is how the reviewer decides whether to agree.
The first four weeks are hard. Every account goes through a period where the flags are too many, or wrong in a way that reflects the playbook rather than the model, and the team’s patience is tested. We have shortened this period, mostly through better playbook templates by practice area, but we have not eliminated it, and we no longer expect to. A playbook written in a week will be wrong in places. The flags are how you find the places.
Partners accept roughly the same share of redlines. Across accounts that have been live for more than three months, the partner accept rate for proposed redlines has sat in a narrow band through the year, and it does not vary much by firm size or jurisdiction. It varies by playbook maturity. A firm six months in accepts more than a firm six weeks in, and the difference is almost entirely in the playbook, not the model.
Security questions are the same questions. Zero retention, where the data lives, who can see what, what happens on termination. The questions have not changed in a year. What has changed is who asks them: a year ago it was IT; now it is often the risk partner or the general counsel directly, and the conversation is shorter because the answers are documented.
What surprised us
Twenty-eight countries means twenty-eight sets of assumptions about what a contract looks like. We expected the challenge to be language. It was, partly; we now support 38. But the harder differences were structural. Civil-law drafting leans on the code and leaves things unsaid that a common-law draft spells out. A German services agreement is often half the length of its English equivalent, not because it does less but because the BGB does more. Review tuned on English and US paper flagged the absence of things that did not need to be present. The fix was not translation. It was jurisdiction-aware playbook defaults, and we are still building them out.
The second surprise was how little firms wanted to share. We had assumed that an anonymised, cross-firm view of market positions would be valued. Almost nobody asked for it, and several firms said explicitly that they would not want their positions contributing to it, however anonymised. A firm’s playbook is, in a real sense, its competitive position. We have shelved the idea.
What we take from this
The model matters less than we thought and the playbook matters more. The surfaces matter less than the reasons behind each flag. Growth across jurisdictions is a legal problem before it is a language problem. And the questions that a careful firm asks before trusting a tool with its drafts are the right questions, and they do not get easier with scale; they get asked by more senior people.
A year from now this list will look naive in places. We will write that one too.