A new workflow is a new hire, not a finished product.

When a new AI workflow goes live in a firm, the most common mistake isn't technical. It's a mismatch of expectations: the system is switched on and then measured, on day one, against a human process the firm has spent years refining. By that standard almost anything new looks like a disappointment, including the associate who started last week.

You wouldn't judge a new associate on their first morning

No one expects a first-year to perform like a senior partner on day one. You onboard them. You hand them the firm's templates and standards. You review their work, give feedback, and correct course. Over months they become genuinely good, because the firm invested in making them good.

A new AI workflow deserves the same patience, for the same reason: it has to learn how your firm works before it can work the way your firm does. Treated as a colleague in training rather than a finished product, it follows the same curve, rough at first, dependable with investment.

A workflow is built, not installed

A freshly deployed system is, almost by definition, half-built. The first version handles the obvious cases; the edge cases, the exceptions, and the firm's unwritten judgment get added as they surface in real work. That isn't a flaw in the build, it is how building works. Anything that touches real legal matters is refined against real legal matters, in cycles. The polish comes from use, not from the install.

The comparison that misleads

Never judge a freshly installed, half-built AI workflow against a human process you have spent years optimizing. The human process looks effortless because the effort already happened, quietly, over a long time, and mostly out of view. The honest comparison is like with like: the new system against where it will be after the same investment, not against the finished state of something far older.

Set the bar at "as good as our best, today" and you will switch off something that was weeks away from being exactly that.

What real results actually require

To see real results, you have to give your AI systems what you would give a human employee: rigorous time, documented procedures, and hands-on training. Standard operating procedures, so the system knows the firm's way of doing things. Feedback loops, so it improves instead of repeating the same misses. People who own it and keep refining it. None of this is overhead, it is the work that turns a promising tool into a dependable part of the practice.

It is also the difference the evidence keeps pointing to. The firms that get a return on AI are not the ones with the highest expectations or the newest tool; they are the ones that approached it deliberately and put in the work. The ones that expected the tool to perform on its own are, overwhelmingly, the ones still waiting.

Patience is the strategy

The firms that win with AI treat a new system the way they treat new talent, with onboarding, standards, and time. Set the expectation correctly at the start, give the workflow the same rigor you would give a new hire, and it earns its place in the practice. Judge it on its first morning, and you will abandon something that was about to get good.

← Back to insights See how we build and oversee the work →