The Preparation-Judgment Split
Most AI tool failures happen because we're automating the wrong half of the work.
I keep seeing the same pattern. A team picks an AI tool, points it at a workflow, and six weeks later quietly stops using it. The tool worked fine. The problem was which half of the work they aimed it at.
Most knowledge work has two distinct phases. There’s preparation — research, synthesis, drafting, pulling data together, summarizing what exists. And there’s judgment — making the call, weighing trade-offs, reading the room, deciding what matters.
AI compresses the first phase dramatically. It barely touches the second.
Where this shows up
I tested this most clearly with contract review. A two-person legal team was spending roughly 12 hours per contract on initial review: reading the document, extracting key terms, comparing deviations against their playbook, flagging risk areas. That’s preparation work. When we pointed Claude at it, review time dropped by about 80%.
But the next step — deciding which deviations actually matter, choosing a negotiation strategy, assessing counterparty risk — that stayed entirely human. The tool had nothing useful to say about whether a particular liability cap was worth fighting over given the relationship dynamics at play.
The preparation-to-judgment ratio in contract review is roughly 80/20. That’s why AI delivers such outsized returns there.
Now compare that to a sales negotiation. The preparation phase — researching the prospect, pulling together talking points, reviewing past interactions — is maybe 20% of the real work. The other 80% is reading the buyer, adjusting your approach in real time, knowing when to push and when to hold. AI helps with the prep, but the core work stays human.
The practical test
Before you deploy an AI tool on any workflow, ask one question: what percentage of this work is preparation versus judgment?
If the answer is 70%+ preparation, you’re looking at a high-ROI automation target. If it’s the reverse, AI might save your team 20 minutes on research while leaving the hard part untouched — and the team will wonder why they adopted a new tool for marginal gains.
This isn’t a knock on AI. It’s a calibration tool. The companies getting real value aren’t the ones buying the most tools. They’re the ones mapping their workflows honestly and starting where the preparation ratio is highest.
Start there. The 80% gains are real, but only when you’re compressing the right half.