Six Week AI · a worked example: Legal
AI for contract review, shown honestly, and the method behind it.
Contract review is one worked example of how I work: pick the one place AI pays off now, build it on your rules, measure it honestly, hand it over. No platform, no lock-in.
Below: the questions that stall a legal-AI project, each with a worked answer, and a live demo on synthetic contracts with the real accuracy numbers published, gaps included.
The finished record · every value cited
Section 7.2 · page 6 · quoted verbatim
…each party's aggregate liability under this Agreement shall not exceed the total fees paid by Customer in the twelve (12) months preceding the claim, except for breaches of the confidentiality obligations in Section 9…
What goes wrong
The questions that stall a legal-AI project, each one answered.
Not a list of fears. Each is a specific, recurring obstacle, and each links to how this system handles it, with the demo behind it.
- HallucinationA confident term that isn't in the document.Fluent and wrong
- TraceabilityEvery value traced to its clause, or the lawyer re-reads the whole contract.Trust the extraction
- Human oversightDoes it know which contracts it's unsure about?Human in the loop
- EvaluationHow to tell a real accuracy number from a marketing one.How you know it works
- EU AI ActWhere contract review actually lands on the risk tiers.AI Act risk tier
- Data governanceWhere the data goes, what's retained, who's liable.Data residency
- Build vs buyA $150K platform, a generic chatbot, or a system your team owns.Build vs buy vs DIY
For your engineer: the full technical build and evaluation →
The approach
The win is one step, not a new platform.
Your work already moves across several systems and people; we don't replace that. We find the one step where AI pays off, often something small like re-keying the same terms into the CRM, your team builds it on your rules, I make sure it's accurate, and they own it. Build, educate, hand over. No platform, no lock-in.
The proof
See how it works.
Read an incoming contract, pull its key terms, and flag what needs a human. Every value traced to its clause, risk scored against an editable playbook, and the doubtful cases routed to a person. Twenty synthetic contracts, nothing gated, and you can re-run it live.
Now
Every value cited to its clause
Before
a 30-minute first-pass read
Now
Each risk tied to the rule that fired
Before
“the AI says it's risky”
Now
The doubtful cases routed to a human
Before
trust it all, or check it all
Current measured run: 87.1% citation accuracy (section and page), 87.8% honest-response rate against the golden set, published in full on the demo's eval panel, gaps included.
The first step
First, let's see if we're a fit.
A 20-minute call to see if we are a good fit for each other. No pitch, no slide deck. You put a face to the text and a feel for how I work; I learn enough about your workflow to know where AI could help. A couple of questions each way, then we both decide if it is worth going further.
Need your CIO's sign-off? Here's the page for them →
Not ready to talk? Run any workflow through the Litmus Test, a one-page test for spotting which of your workflows suit AI on your own. One email, no follow-up.
Already had your fit call? Book the 90-minute workshop ($2,500, a standalone session); you secure your slot when you book. Book the workshop →